construction theoretic approaches to grammar: lessons … · construction theoretic approaches to...
TRANSCRIPT
1
1
Budapest Szuperkurzus – June 4-8, 2007 Farrell Ackerman UC San Diego
Construction theoretic approaches to grammar: Lessons for Syntax from Wordbased (Word and Paradigm) Morphology1
The aim of science is not to open the door to infinite wisdom, but to set a limit to infinite error. -Betholt Brecht, Life of Galileo
“…individuals are quite stupid compared to the complexity of the problems we aspire to solve… All anyone can hope to do is to make canny simplifications that do minimum damage to understanding.” Peter J. Richerson & Robert Boyd 2005:248 Not by genes alone: How culture transformed human evolution.
1. INTRODUCTION
1. Course Goal
Making sense of language: identifying the conceptual alternatives and analytic toolkits associated with competing approaches to morphology and syntax and arguing for a construction-theoretic approach to grammar analysis.
Argue, using morphological theory, that morpheme-based (realizational) proposals, in which words and paradigms are epiphenomena, are less preferable than stem-based realizational proposals, and that word-based proposals complement both.
Word-based proposals are construction-theoretic and require a similar toolkit to that required for construction-theoretic syntax.
Focus on lexical/morphological data displaying single word (synthetic) versus multi-word (periphrastic) expression for derivation ( lexeme formation) and inflection.
1 These lectures represent collaborative research with Jim Blevins, Jeremy Boyd, Mark Gawron, George Gibbard, Rob Malouf, Irina Nikolaeva, Greg Stump, and Gert Webelhuth, as well as benefiting from the research of Alice Harris and Andrew Spencer. Naturally, only I am responsible for all misunderstandings and misdirections.
2
2
2. Two guiding hypotheses Concerning theory, construction-theoretic approaches to grammar contain the toolkits that permit profitable interaction with developmental theories in learning and biology.
Encourage the consideration that words and syntactic structures represent patterned configurations of meanings (= constructions) in different domains and that each enter into systems of paradigmatic relatedness , contra the standard mainstream generative grammar (MGG) view according to which words, sentences, and paradigms are epiphenomena.
If one listened to workaday morphologists (Aronoff 1994; Anderson 1992; Blevins
2005, 2006; Booij 2002, 2005; Corbett; Gurevitch 2005; Harris; Matthews 1991; Pullum and Zwicky 1992, Robbins 1959; Spencer 2005, 2003, 2001; Stump 2001, 2003; Zwicky 1992, among many others), as well as to lexicalists (constructionists) advancing correspondence theories of grammar (L(exical) F(unctional Grammar), H(ead-Driven) P(hrase) S(tructure) G(rammar), Jackendoff & Culicover’s constructional grammar 2005, Hudson’s word grammar 2007, Fillmore et. al. ), i.e., multiple independent levels in correspondence with each other, rather than syntacticians analyzing morphology (Marantz, et. al.), , mainstream generative assumptions would seem odd and many issues concerning learnability and the organization of grammars would interface elegantly with independent proposals in learning theory and developmental biology.
Concerning methodologies, we will not gain much by utilizing new quantitative and experimental methodologies if we simply apply them to familiar phenomena from the usual languages in order only to bolster cherished beliefs, rather than permitting them to help us identify new objects of analysis that may require new theoretical toolkits, especially when the latter connects linguistics with cutting edge research in cognitive (neuro)science. 3. Old goals, modern goals: 1
“That science [linguistic science] strives to comprehend language, both in its unity, as a means of human expression and as distinguished from brute communication, and in its internal variety, of material and structure. It seeks to discover the cause of the resemblances and differences between of languages, and to effect a classification of them, by tracing out the lines of resemblance, and drawing the limits of difference, It seeks to determine what language is in relation to thought, and how it came to sustain this relation; what keeps up its life and what has kept it in existence in past time, and even, if possible, how it came into existence at all.” W. D. Whitney 1875:4
a. Language as a species specific artifact and capacity of humans.
Powerpoint: Mayberry on critical period effects for language development.
b. What are the similarities behind the obvious differences among languages and what are their causes?
3
3
c. Are there constraints on how languages can differ, i.e., on the form of a possible human language?
d. What is the relation between language and other cognitive capacities, i.e., thought?
Pattern-based information (Systems of patterns) Tomasello (2003:4) arguing for research strategies informed by tighter relations
between language and other domains observes that skills in human pattern-finding derive, among other things, from:
“the ability to form perceptual and conceptual categories of “similar “ objects and events; the ability for perform statistically based distributional analyses on various kinds of perceptual and behavioral sequences: the ability to create analogies (structure mappings) across two or more complex wholes, based on the similar functional roles of some elements in these different wholes.”
Development: Representational Redescription “My position regarding meta-representation has always been in terms of `redescription’ of representations, rather than simple duplication…The lower levels are left intact; copies of these are redescribed. Redescription involves a loss, at the higher level, of information that continues to be represented at the lower level. Our multiple levels of representation are not, I submit, simple duplicates of lower levels; rather, they involve increasing explicitation and accessibility at the cost of detail of information.” Karmiloff-Smith 1993:569 in Bloom ed.
Strong Competence Hypothesis: Coming full circle in motivating the lexicalist trend
Kaplan and Bresnan’s (1981) desideratum of “strong competence hypothesis”: linguistic theories should be evaluated not only in terms of their ability of parsimonious and comprehensive analysis of data (however this is finally calculated), but with respect to how the ingredients of such analyses relate to research results in domains independent of language analysis, such as acquisition and parsing.
e. How did language arise, i.e., what is the evolutionary origin of language in terms of its biological and cultural sources.
Powerpoint: Elman’s speculation on language emergence.
f. How is language learned without explicit instruction?
“In the natural acquisition of of our native language we hear little by little a number of sentences which are constructed in the same way and therefore constitute a group, and so it comes about that the rule is unconsciously abstracted from these models.” Paul:111 cited in Itkonen 2005:125.
4
4
“Language acquisition is based on the ability to abstract from a corpus of sentences a certain structural pattern, and to construct, from the old materials, new sentences conforming to this pattern.” Chomsky 1955:131 cited in E. Itkonen 2005:125 Powerpoint: Simonyi, Zs. 1889 on the role of analogy. 4. The expression problem for theory construction
The “same” construction finds very different surface expressions in the world’s languages, i.e., they can be expressed either by one word (synthetically) or many (periphrastically).
Lexeme formation with prefixes/preverbs: (Ackerman and Webelhuth 1998 Chapter 1, below on Realization-based Lexicalism)
(3) obxod N ‘round’ (as in ‘make the rounds’) (4) obxodnyj A ‘roundabout’
(8) äraostmatu A ‘incorruptible’ äraostmatus N ‘incorruptibility’ äraostetav A ‘venal, corrupt’ äraostetavus N ‘venality’ Active and Passive:
5
5
Given differences between related predicates concerning meaning, valence,
semantic role inventory and grammatical function requirements and, that such relations trace a cline from idiosyncratic to partially productive to fully
productive, What is the toolkit that can explain these relations while also accounting for
surface differences (exponence) in language particular and cross-linguistic expression?
Is there any reason to privilege one type of surface expression over another, giving
it a primary explanatory role in theoretical analysis? 5. Basic Lexicalist assumptions (largely shared by Head-Driven Phrase Structure Grammar, Lexical Fuctional Grammar and Construction-theoretic Grammars) (1) Principle of morphological integrity2
Syntactic mechanisms neither make reference to a word form’s proper subparts nor can they create new word forms in constituent structure.
(2) Principle of lexical modification3 The lexical properties (meaning, argument structure, grammatical function inventories, and case government patterns) associated with a lexeme are fully determined by lexical stipulation together with rules of lexeme derivation and cannot be altered by items of the syntactic context in which a realization of that lexeme appears.
(3) Principle of morpholexical inflection The morphosyntactic content associated with a lexeme’s realizations is fully determined by lexical stipulation together with rules of inflectional morphology and cannot be altered by items of the syntactic context in which a realization appears.
(4) Principle of unary expression In syntax, a lexeme is uniformly expressed as a single morphophonologically integrated and syntactically atomic word form.
2 We equate this with the lexical integrity principle of Bresnan & Mchombo (1995). 3 This was referred to as the principle of lexical adicity in Ackerman & Webelhuth (1998).
6
6
6. Derivation (lexeme-formation): synthetic v. periphrastic causatives Chichewa:
Catalan:
7. Inflection: Periphrastic (compound) tense
Hungarian: Ackerman 1987 (from Kálmán et. al. 1984) a holló énekel egy dalt a rókának the raven sing a song.ACC the fox.DAT `the raven is singing a song to the fox.’ a holló énekelni fog egy dalt a rókának the raven sing.INF. will a song.ACC the fox.DAT `the raven will sing a song to the fox’ a holló énekelhet egy dalt a rókának the raven sing.MOD.3SG a song.ACC the fox.DAT `the raven can sing a song to the fox.’ a hollónak énekelnie kellett egy dalt a rókának the raven.DAT sing.INF.3SG must.past a song.DAT the fox.DAT `the raven had to sing a song to the fox’
7
7
8. Taxonomy of lexicalist approaches
(1) Morphological
Integrity
(2) Lexical
Modification
(3) Morpholexical
Inflection
(4) Unary
Expression Classical LFG and HPSG (Bresnan 1982b; Pollard & Sag 1987)
yes yes yes yes
Some recent views in LFG and HPSG (Alsina 1992, 1997; Hinrichs & Nakazawa 1989, 1994)
yes no
(periphrastic causatives)
no (periphrastic
tense) yes
Realization-based Lexicalism
yes yes yes
no (periphrasis is
a type of morphological
expression) TABLE 1. Taxonomy of Lexicalist Approaches
Assumptions of Realization-based Lexicalism
• Only morphological and not syntactic rules can create or analyze morphological words (= the principle of morphological integrity).
• Only morphological and not syntactic rules can create new argument structures (= the principle of lexical modification).
• Only morphological and not syntactic rules can associate morphosyntactic content with a lexeme’s realizations, whether these be synthetic or periphrastic (= the principal of morpholexical inflection).
• Lexemes are preferably expressed by single synthetic word forms but can also be expressed by combinations of words; that is, the principle of unary expression is merely a preference principle.
9. The construction-theoretic hypothesis: The Sausurrean Aspects of a Sign4 “ The results flowing from these architectural assumptions and representational tools set strongly lexicalist theories apart favorably from the recent state of the art in the transformational paradigm, where a purely theory-dependent proliferation of categorial heads and syntactic movements has made the promise of unified, concrete, and empirically contentful theory of syntax and morphology a matter of faith against all hope.” Ackerman & Webelhuth 1998:349.
4 This anticipates discussion of Ackerman & Stump 2004 on the mapping between content to form paradigms below.
8
8
A. Content-theoretic Aspects of the Sign5 . Functional-semantic content (lexical semantics, semantic roles, syntactic roles) . Morphosyntactic content (tense, aspect, agreement, case…)
B. Form-theoretic Aspects of the Sign . Categorial form . Morphophonological form
Content is set in many-many correspondences with form, i.e. while content is
similar cross-linguistically, form is not and content cannot be read off of form in a compositional fashion.
Addresses universal contentive aspects of grammar encoded by such
morphosyntactic properties ( = functional categories) as tense, aspect, agreement, etc. without assuming a privileged role for tree-theoretic representations of these notions.
10. The proliferation problem
The tools useful for accounting for one phenomenon are extended and applied to more and more disparate phenomena, with increasing abstractness, i.e., global parsimony
Morphosynyactic properties, i.e., Tense and Agreement, are handled in the lexicon
in unification-based lexicalist theories; mainstream generative grammar postulates the existence of categorial heads which express this information in phrase structure.
5 These distinctions correspond to content-paradigms and form-paradigms in the upcoming presentation of word-based morphology.
9
9
Tree-theoretic representations, defensible in some languages on the basis of classic constituency tests, are reified, with extra assumptions such as uniform binary branching (see below), into (possibly) Universal Grammar representations underlying surface diversity.
Speculation about assuming favored constructs and structures as universals “Now any classification that starts with preconceived values or that works up to sentimental satisfactions is self-condemned as unscientific. A linguist that insists on talking about the Latin type of morphology as though it were necessarily the high-water mark of linguistic development is like the zoologist that sees in the organic world a huge conspiracy to evolve the race-horse or the Jersey cow. Language in its fundamental forms is the symbolic expression of human intuitions. These may shape themselves in a hundred ways, regardless of the material advancement or backwardness of the people that handle the forms, of which, it need hardly be said, they are in the main unconscious. If, therefore, we wish to understand language in its true inwardness we must disabuse our minds of preferred "values" and accustom ourselves to look upon English and Hottentot with the same cool, yet interested, detachment.” Sapir 1921:124
Why not choose any random selectively type as the universal from which all other types deviate?
10
10
11. Old goals, modern goals: 2
“Thus it is that the different uses, in which we have come to be acquainted with a word or a phrase, associate themselves with each other. Thus, to the different cases of the same noun, the different tenses, moods and persons of the same verb, the different derivatives of the same root, associate themselves, thanks to the relationships between their sounds and the meaning;….further, forms of different words with similar functions - e.g., all plurals, all genitives, …all masculines which form their plural by means of umlaut as contrasted with those that form it otherwise… These associations may one and all arise and operate without consciousness, and they must not be confounded with grammatical categories, which are the result of conscious abstraction, though they not infrequently cover the same ground. Paul 1890:6 (in 1970 trans.) a. What is the relation or network of associations between all of the inflected and derived forms of a word? Powerpoint: Simonyi, Zs. 1889. On analogy and frequency b. What are the general strategies or patterns that (classes of) words participate in with respect to the encoding of morphosyntactic and derivational information? c. How do associative relations, as characterized above, and their convenient simplification into categorical distinctions, e.g., discrete parts of speech, semantic roles, etc., correspond to actual knowledge of language ( = competence)? Powerpoint: Sapir on taxonomy of expression types “It is because our conceptual scheme is a sliding scale rather than a philosophical analysis of experience that we cannot say in advance just where to put a given concept. We rnust dispense, in other words, with a well-ordered classification of categories. What boots it to put tense and mode here or number there when the next language one handles puts tense a peg "lower down" (towards I), mode and number a peg "higher up" (towards IV)? Nor is there much to be gained in a summary work of this kind from a general inventory of the types of concepts generally found in groups II, III, and IV. There are too many possibilities.” Sapir 1921:107
The lesson “The table shows clearly enough how little relative permanence there is in the technical features of language. That highly synthetic languages (Latin; Sanskrit) have frequently broken down into analytic forms (French; Bengali) or that agglutinative languages (Finnish) have in many instances gradually taken on "inflective" features are well-known facts, but the natural inference does not seem to have been often drawn that possibly the contrast between synthetic and analytic or agglutinative and "inflective" (fusional) is not so fundamental after all….Facts such as these seem to lend color to the suspicion that in the contrast of pure-relational and mixed-relational (or concrete-relational) we are confronted by something deeper, more far-reaching, than the contrast of isolating, agglutinative, and fusional.” 1921:146
11
11
On the consequences of ignoring unfamiliar (Amerindian) language structures for linguistic theory construction
“To exclude the evidence which their languages offer as to what the human mind can do is like expecting botanists to study nothing but food plants and hothouse roses and then tell us what the plant world is like.” Whorf 1956: 215 (from Science and Linguistics)
In MGG surface forms only interesting or useful inasmuch as they help theoreticians to “discover” the nature of abstract underlying representations and the operations required to (re)achieve surface expression.
Neither the surface forms themselves nor their organization into networks of
relatedness are interesting or instructive about grammar.
Is deriving (classes of) individual words and sentence all that theory needs to explain, or does the organization of words and sentences into networks of relatedness require explanation as well?
12. Four basic questions: (Some answers provisionally provided at the end of the course) Q1: What is syntax? Q2: What is morphology? Q3: What is their relationship between syntax and morphology?
Can we lean about morphology from syntax? Can we learn about syntax from morphology? Is morphology reducible to syntax, i.e., are the representations and assumptions used for syntax the same as those used for morphology? Are neither morphology and syntax reducible to each other, but can both still be analyzed using a similar toolkit?
Q4: What is the relationship between the theoretical “toolkit” used in the analysis of morphology and syntax to language development in children? What is the start-state, what is the end-state and how do we trace the trajectory between them? (nativist v. constructivist approaches to grammar learning)
13. General outline of the course
Readings: These readings are selected to cover the basic territory addressed in these lectures – The focus of the course will be on conceptual issues and argumentation for a construction-theoretic approach, rather on formal implementation. Ackerman, F., G. Stump, G. Webelhuth. To appear. Lexicalism, periphrasis and implicative morphology. In B. Borsley & K. Börjars eds. Non-transformational Theories of Grammar. Oxford: Blackwells Publishers. Ackerman, F. & Gert Webelhuth, 1998. A Theory of Predicates. CSLI Publications/Cambridge University Press. Chapter 1.
12
12
Braine, M.D.S. 1992. What sort of innate structure is needed to “bootstrap” into syntax? Cognition 45.77-100. Carstairs-McCarthy, A. 2005. Basic Terminology. Štekaur, P. & R. Lieber eds. Handbook of Word-formation. Springer. Hay, J. and H. Baayen 2005. 2005. Shifting paradigms: gradient structure in morphology. Trends in Cognitive Sciences Vol.9 No.7 342-348. Stewart, T. and G. T. Stump. 2007. Recommended background: Culicover, Peter & Ray Jackendoff. 2005. Simpler Syntax. Chapters 1-4. Oxford University Press. Recommended background: Spencer, A. 2004. Morphology – An overview of central concepts. A. Spencer & L. Sadler eds. Projecting morphology. CSLI Publications. 6/4
1. INTRODUCTION
2. THEORY CONSTRUCTION: STARTING SIMPLE AND BASIC (G. HOLTON, THE ADVANCEMENT OF SCIENCE, AND ITS BURDENS)
3. INTRODUCTION: WHERE ARE WE, SHOULD WE STAY, AND IF WE DECIDE TO LEAVE, WHERE DO WE GO? 6/5 4. TAXONOMY OF CONSTRUCTION-THEORETIC APPROACHES (ADAPTING GOLDBERG 2006)
5. EVERYTHING IS SYNTAX: CATEGORICAL/UNIFORM (EMBICK 2007)
6. SYNTHESIS VERSUS PERIPHRASIS – MORPHOLOGY IS MORPHOLOGY AND SYNTAX IS SYNTAX (STUMP 2002)
7. COMPETITION AMONG SURFACE EXPRESSIONS: A UNIFORM TREATMENT OF BLOCKING AND OPTIONALITY (BOYD 2007)
6/6
8. SYNTAGMATIC VERSUS PARADIGMATIC APPROACHES TO GRAMMAR
9. SYNTAGMATIC/COMPOSITIONAL/VERTICAL MORPHEME-BASED APPROACHES
10. SUSPICIONS ABOUT THE SYNTAGMATIC/COMPOSITIONAL/VERTICAL (MORPHEME-BASED) APPROACH
13
13
11. PARADIGMATIC/CONFIGURATIVE/HORIZONTAL WORD-BASED APPROACHES 6/7
12. REALIZATION-BASED LEXICALISM: EXPONENCE-BASED AND IMPLICATIVE (ACKERMAN AND STUMP 2004, ACKERMAN, STUMP, WEBELHUTH TO APPEAR)
13. CONSTRUCTION-THEORETIC MORPHOLOGY: WORD-BASED MORPHOLOGY 6/8
14. CONSTRUCTION THEORETIC SYNTAX - SOOPERATING CONSTRUCTIONS IN THE RELATIVE CLAUSES OF EURASIA (ACKERMAN, MALOUF, NIKOLAEVA)
2. THEORY CONSTRUCTION: STARTING SIMPLE AND BASIC (G. HOLTON, THE ADVANCEMENT OF SCIENCE, AND ITS BURDENS)
1. The basic picture of Einstein’s account of theory construction Powerpoint: Representations from Holton’s essay
What counts as E, i.e., the empirical phenomena that the theory is responsible for explaining?
Is there a core/periphery distinction in language?
How does one determine what are important or instructive empirical phenomena?
Is this best determined by the sorts of issues which our toolkit seems best adapted
to address?
Or should our toolkit be determined by detailed understanding of “representative” empirical phenomena?
2. The elements of the basic model and their relations 3. Dimensions of theory construction 4. How are theories supposed to work?
Textbooks tend toward two strategies in presenting theories:
14
14
A: Try to provide a sense of how A is derived from E and then how A leads to deductions concerning E’s that did not play a role in the postulation of A. (Cf. Haegemann 2006 Chapters 1 & 2; Matthews 2007 Chapters 1 & 2)
B: Simply state A and how it’s deductions correspond to Es.
5. External validation
What counts as an early stage and when can one feel that it is responsible and necessary to question thematic (see below) concerns?
6. Inner perfection
The set of axioms associated with the theory on the left is augmented by auxiliary assumptions./modifications etc., required for observational adequacy in the theory on the right.
What counts as prediction? Does the theory address new data without the need to
modify known and independently justified assumptions?
What counts as falsifying evidence, if one is permitted to proliferate auxiliary assumptions?
7. Achieving global parsimony “By going cyclically through several stages of theories, each stage is forced to use conceptions more removed from direct experience (e.g. atomism), The distance from E to the A is larger, the contact with common sense is more and more tenuous. But the fundamental ideas and laws of science attain a more and more unitary character. In the meantime another cost of this process is that the more general the theory becomes, the longer it may have to wait for the correlation of its predictions with the ground of experience… “It may need many years of empirical research to ascertain whether the theoretical principles correspond with reality”; for it may take that long to discover “the necessary array of facts.” (I.O. pp. 223) Holton 1986:49 8. Beliefs and best guesses as guides 9. Theory comparison
Themata, i.e., guiding ideas, in the confrontation of rival theories.
So, we will initially juxtapose competing “themata” and explore some of the consequences of assuming some of the less familiar themata, i.e., the leading ideas of lexicalist/construction-theoretic approaches.
After all, we will never have an accounting of which of several alternatives is really
better unless we explore honestly some of the relevant alternatives.
15
15
10. A cautionary note
Theory comparison is hard, often leading to the dismissal of rival theories on the basis of ignorance of what they’re really doing.
A typical comparative strategy: Select an empirical phenomenon, demonstrate
that a possibly outdated implementation of a rival perspective (presented as the only possible implementation of the guiding ideas of that approach) is inadequate and then demonstrate that the favored theory (enjoying the advantage of cumulative changes made since the time of the dismissible alternative) is the sole explanatory proposal.
Motivation: Not only to propose an analysis of a particular phenomenon, as
illustrative of a class of phenomena, but to show that this analysis is better for this phenomenon and, by implication, any other interesting phenomenon, than any conceivable alternative based on different assumptions.
3. INTRODUCTION: WHERE ARE WE, SHOULD WE STAY, AND IF WE DECIDE TO LEAVE, WHERE DO WE GO? 1. Promising developments in the cross-linguistic study of morphosyntax.
More expansive recognition of the types of information that receive grammatical encoding as well as a more empirically grounded investigation into the principles concerning how such information is expressed morphologically and syntactically.
Language models formulated in terms of inviolable, categorical, but
parameterizable, principles and restricted to so-called core phenomena are gradually being replaced by theories which examine broader expanses of grammar viewed from the perspective of constraint-based systems: these systems, in turn, are viewed as adjudicating relations among (sometimes, stochastically interpreted) violable constraints6 which are sensitive to the often gradient nature of grammatical phenomena.
New approaches designed to capitalize on richly articulated and independent
morphological, lexical and syntactic representations in order to provide insights into networks of lexical and syntactic relatedness displaying varying degrees of generality.
6 See Aissen and Bresnan 2001; Baayen et. Al.: Bresnan 2006. 2007; Albright 2001; Albright and Hayes 2003; Bod et. al. eds. 2003; Boersma and Hayes 2001; Hay, J. 2002, 2003; Keller 2001, Keller & Sorace 2005, S. Kirby 2003, among others.)
16
16
Proposed representations model the complete spectrum of grammatical phenomena from regular to partially regular to wholly idiosyncratic formations using the same “toolkit”, i.e., there has been an effort to explore single system models of grammar (construed either as rules or exemplars), directly designed to address issues of scalability, i.e., identifying the inventories of ingredients and the principled ways in which they can be used to relate hierarchically organized types of grammatical information to their subtypes. 7
2. Linguistics undergoing an important reconceptualization concerning its methodologies, its objects of inquiry, and ideas about theory construction.
The nature of these changes align it more with the biological and developmental sciences in their emphases on the (probabilistic) modeling of complex interacting systems than with the crisper deductive bases of the physical sciences that seem to have served as the guiding metaphorical standard within the predominent Chomskyan program of grammar analysis.8
“The goal of the physicist is to establish general laws and to reduce all phenomena to a minimal number of such laws. General laws, however, play a much smaller role in biology. Just about everything in biology is unique: every animal and plant community, flora or fauna, species or individual. The strategy of research in biology must be quite different from the strategy of the physicist. In this respect, biology may be nearer to such sciences as archeology and linguistics.” Mayr 1976:15 cited in Sussman 1999:6.
The study of biological objects requires concepts and methodologies that set it apart from disciplines such as physics and chemistry. He elaborates his views by observing that biology:
7 See Malouf 2001, 2003; Ackerman and Webelhuth 1998; Sag 1997; Fillmore and Kay 2001; Fillmore, Kay, Michaelis, and Sag, to appear, Culivover and Jackendoff 2005, among others). These approaches take seriously the task of “scalability”, whereby small and more highly constrained generalizations are accounted for with the same mechanisms as used for large and less locally conditioned ones. 8 See E. Mayr 2001 for an enlightening overview of the differences between the biological and physical sciences. Within linguistics these competing perspectives are nicely illustrated by the contrasting approaches to grammar learning evident in Karmiloff-Smith (1997, 2003), Tomasello (2003, 2006) versus Anderson and Lightfoot (2003) and Anderson (2004), despite the convergent appeals for a neurobiological perspective among these authors. The consequences of utilizing natural science assumptions for language analysis have recently been examined in Pinker and Jackendoff’s (2005) criticism of the hypotheses about language advanced in Chomsky, Hauser, and Fitch (2002). This view which underlies and guides the Minimalist Program regards language as a perfect system with, perhaps, ancillary functional utility. Chomsky et. al. suggest that this perspective on languages raises foundational issues for the biological sciences. After all, they note, there is no biological system that resembles the sort of object identified by the Minimalist Program. Pinker & Jackendoff aver that this is a curiously misguided and misleading belief, given the rigor of research and the solidity of results in biology versus those in linguistics. However such issues are finally resolved, for our purposes it is important to see that different conceptualizations of the problems yield radically different views of language analysis. (See the discussion of emergentism below.)
17
17
“ … deals to a large extent, with unique phenomena, such as the extinction of dinosaurs, the origin of humans, the origin of evolutionary novelties, the explanation of evolutionary trends and rates, and the explanation of organic diversity. There is no way to explain these phenomena by laws… With the experiment unavailable for research in historical biology, a remarkable new heuristic method has been introduced, that of historical narratives. Just as in much of theory formation, the scientist starts with a conjecture and thoroughly tests its validity, so in evolutionary biology the scientist constructs a historical narrative, which is then tested for explanatory value.” Mayr 2004:32
Distinction between methodological assumptions in the physical versus biological
sciences is further explored by Camazine et. al. 2001:12 who suggest that: “The mechanisms of self-organization in biological systems differ from those in physical systems in two basic ways. The first is the greater complexity of the subunits of biological systems. The interacting subunits in physical systems are inanimate objects such as grains of sand or chemical reactants. In biological systems there is a greater inherent complexity when the subunits are living organisms such as fish or ants or neurons... The second difference concerns the nature of the rules governing interactions among system components.
Two elements relevant to our concerns: (1) what does it mean to be a self-organizing system; (2) what sorts of rules or principles regulate the system.
“Self-organization is a process in which pattern at the global level of a system emerges solely from numerous interactions among the lower level components of the system. Moreover, the rules specifying interactions among the system’s components are executed using only local information, without reference to the global pattern. In short, the pattern is an emergent property.” Caramzine et. al. 2001:8
Patternment and parameters: probabilistic tuning, rather than categorical setting (cf. references to Paul above)
“One speaks of “tuning” a parameter in the system to invoke the onset of a different pattern…By making small adjustments in such parameters, one can induce large changes in the state of the system…the concept of self-organization alerts us to the possibility that strikingly different patterns may result from the same mechanism operating in a different parameter range.” Caramazine et.al. 2001:32 & 38. Powerpoint: O’leary et. al. on tuning in biology9: Powerpoint: Bresnan et. al. on tuning in grammar10
The patterns that emerge can (1) have properties that aren’t simple aggregates of contributing (categorical) parameters and (2) constitute objects which themselves participate in systems of relations at a level different than that which accounts for many of their properties.
9 Based on a lecture by Joan Stiles May 2007. 10 Based on a lecture by Joan Bresnan May 2007.
18
18
“Emergence refers to a process by which a system of interacting subunits acquires qualitatively new properties that cannot be understood as the simple addition of their individual contributions.” Caramzine et. al. 2001:31.
Organization or patternment evident in systems is not driven by teleological considerations, i.e., neither organization nor change within the system reflect conscious efforts to achieve some kind of global or local goal.
The grammar system of language is in many respects a self-organizing system in
the sense that properties of the system serve a canalizing and directive function on the shape of forms and the nature of the distinctions existing at any time period.
This view of a system is reminiscent of Saussure’s speculations concerning the
distinction between diachronic (evolutionary) and synchronic (static) linguistic analysis, where only the latter constitutes an exploration of language as a system:
“The diachronic perspective deals with phenomena that are unrelated to systems although they do condition them.” 1966:85
The Saussurean view is likewise non-teleological: “In contrast to the false notion that we readily fashion for ourselves about it, language is not a mechanism created and arranged with a view to the concepts to be expressed. We see on the contrary that the state which resulted from the change was not destined to signal the meaning with which it was impregnated.” 1966:85
Saussure is referring to the historical change in the system of Anglo-Saxon whereby a formal opposition among forms is required to mark the difference between singular and plural.
An antecedent state when this opposition for the lexeme foot was signalled by fõt:
*fōti eventually becomes fõt: fēt
Neither set of encodings is more suitable for conveying the number opposition and, relatedly, that the later shape of the plural was not designed to specifically express this morphosyntactic category.
Rather, the change occurred and was impressed into the systemic role of expressing
the already existing number distinction.
Of course, changes can also introduce new distinctions.
19
19
Since grammatical systems are clearly complex systems consisting of patterned organization in numerous dimensions of analysis, i.e., phonology, morphology, semantics, syntax, etc., the analogy with biological systems permits us to entertain the notion that their evident complexity emerges from relatively simple interactions among their component parts. As Caramzine et. al., suggest:
“What is intriguing about pattern formation in biological systems and lends excitement to studies of self-organization in animal groups is the recent realization that interactions among system components can be surprisingly simple, even when extremely sophisticated patterns are built…” 2001:13
How does the hypothesis that natural language complexity, like complexity in biological domains, arises from systemic interactions of actually existing elements jibe with MGG nativist and domain-specific perspectives on language?
Whatever eventual contribution to language organization may ultimately and
reliably be attributed to language specific and species specific properties of humans, we are encouraged to seek for clues about language from the “imagination in the system” as expressed below:
“Relatively little needs to be coded at the behavioral level and the information required for action by the individual is often local rather than global. In place of explicitly coding for a pattern by means of a blueprint or recipe, self-organized pattern formation relies on positive feedback, negative feedback, and a dynamic system involving large numbers of actions and interactions. With such self-organization, environmental randomness can act as the `imagination of the system’, the raw material from which structures arise. Fluctuations can act as seeds from which patterns and structures are nucleated and grow. The precise patterns that emerge are often the result of negative feedback provided by these random features of environment and the physical constraints they impose, not by behaviors explicitly coded within the individual’s genome.” Camazine et al. 2001:26
While linguistic theory can profit from results in the analysis of complex systems within biology, both methodologically and conceptually, linguists are singularly positioned to identify the empirical phenomena, i.e. the paradigmatic as well as the syntagmatic organization of language systems, that require explanation.
Linguistics can learn a lot from other fields which focus on the development and
organization of complex systems, but linguists can contribute a lot as well.
Are we willing?
20
20
3. Results in linguistics demand reappraisal of biology: Language is perfect and useless “Any progress toward this goal [showing that language is a “perfect system”] will deepen a problem for the biological sciences that is far from trivial: how can a system such as language arise in the mind/brain, or for that matter, in the organic world, in which one seems not to find anything like the basic properties of human language? That problem has sometimes been posed as a crisis for the cognitive sciences. The concerns are appropriate, but their locus is misplaced; they are primarily a problem for biology and the brain sciences, which, as currently understood, do not provide any basis for what appear to be fairly well established conclusions about language.” (Chomsky, 1995, pp. 1–2). Cited in Piker and Jackendoff 2005:22911
Results in linguistics are incompatible with what it is independently established in other disciplines.
4. Universal grammar: One particular `just so story’ among many
“UG is the true theory of the genetic component that underlies acquisition and use of language….[there is] no coherent alternative to UG.” Noam Chomsky as reported in the New Yorker Magazine on Dan Everett. April 2007.
It is trivially true and long recognized that humans have a unique capacity for Language, but this in no way entails that the MGG conception of UG in terms of tree-theoretic structures and biological assumptions about representational innateness are the only coherent, let alone an actually coherent, explanation of this. (recall speculations based on Mayberry and Elman above.)
In fact, this claim about the reality (either psychological or biological) of underlying tree-theoretic representations and UG should remind linguists of Uriel Weinreich’s rejoinder to Paul Katz concerning changes to an early and far less abstract variant of generative theory and the imperviousness of its basic assumptions to empirical disconfirmation:
“By this mysterious power to change his theory without changing it, Katz seeks to guarantee the perennial correctness of his approach, abstracted from any particular formulation of it.” 1967:286
5. Innate grammar constraints: Do they ever last after scrutiny?
Gordon (1985) and Gordon and Alegre (1996) it has been claimed that English speaking 3 and 5 year old children appear to be aware of the adult-like distinction in acceptability between mice-killers and *rats-killers12: while irregular plurals can occur as left-members in lexical compounds (as in mice-killers), regular plurals cannot (*rats-killers).
11 S. Pinker & R. Jackendoff. 2005. Cognition 95:201-236. 12 See Ramscar (2005) on the (un)reliability of the judgments producing these data – they are claimed to be elicitation procedure artifacts.
21
21
The purported categoricality of the simple split in acceptability in compounds has
led to extraordinary claims concerning the structure of the human language faculty and the nature of language learning (e.g., Pinker 2000, Clahsen and Almazan 2001, Clahsen, et al. 2003).
In particular, the Dual Route Model (Pinker 2000) is designed to preclude the
possibility of regular inflection occurring internal to lexical compounds. This is schematized in Figure 1:
Figure 1 Dual Route Model of Morphology: No regular infection internal to derivation
Since the plural of mouse is the irregular form mice, it is stored in the lexicon: it
cannot be created by the application of a regular rule. Since it is stored, the compounding process in the lexicon has access to it and it
can, as a result, combine with killer to yield mice-killer.
In contrast, since rat is not associated with an irregular plural form, only the singular form is listed in the lexicon and only the singular form is available for lexical compounding.
This precludes creation of the compound rats-killer, while allowing rat-killer to be
formed. The regular plural formation operation can then apply to the compound and yield, if need be, rat-killers.
Putative split and the earliness of behavioral (acquisition) evidence for it has
suggested the possibility of biological constraints on the architecture of grammar, with consequences for synchronic grammar organization and language learning.13
Clahsen 1999:100914 states that the “feeding relationships between plural
inflection and compounding are determined by a grammatical ordering constraint. It is hard to see how children could learn this constraint directly from input data.”
13 This possibility is raised again in a recent series of experiments on English regular and irregular inflection internal to lexical compounds by I. Berent & S. Pinker (in press). They appropriately distinguish between the existence of a constraint and claims about the source of that constraint. They state their goal as follows: “Our interest here is whether the dislike of compounds like rats-eater is due to a constraint against regular plurals in compounds--an issue that is logically distinct from questions concerning the origins of that constraint.” We concur with the independence of these enterprises, but our own interest in this section is in how specific morphological assumptions appear to have encouraged particular beliefs about these origins or source of such a behavior.
22
22
Implication is that since such a constraint cannot be learned, it must be innate, i.e.,
a form of what Elman et al. refer to as representational nativism, the claim that the human species possesses an innate specification of substantive expectations and constraints that could yield the effect of e.g., regular inflection being absent internal to lexical compounding.
If there is a native and early predisposition against regular inflection internal to
derivation one might expect that, all other things being equal, (1) grammars would not contain such constructions, since the (innate) language architecture itself prohibits it and (2) all children at their earliest stages would behave like English-speaking children in using only irregular inflection internal to compounds, whatever their target language.15
6. Regular inflection internal to derivation in synchronic systems16
A. In Sepečides-Romani (Cech 1995/1996:78), plural nominal forms can serve as
bases for denominal verb derivation:
NOUNplural + ndivola → VERBget full of noun
Template for inchoative verb formation:
There are some formations in which a plural interpretation of the nominal base is
transparent:
Singular Plural (1) rukh ‘tree’ rukha ‘trees’ O veš rukhandivola the wood tree.PL.INCHOATIVE ‘The wood gets dense with trees.’ (2) džuv ‘louse’ džuva ‘lice’ O bala džuvandivola the hair louse.PL.INCHOATIVE ‘The hair is getting full of lice.’
In (1) and (2), the suffix –a appears in plural forms of noun: plural form, in turn, appears in the derived verbs meaning ‘dense with trees’ and ‘dense with lice’, where the plural meaning is transparent.
Similar formations apply to loanwords, as long as the semantics of the prospective
predicate is acceptable. For example consider the following loanword from
14 See Scholz and Pullum (2005) for a carefully reasoned critical evaluation of reflexive appeals to innateness of the sort guiding Clahsen’s remarks. 15 It is important to note that we have generalized the presumptive constraint from being specific to regular plurals to encompassing all regular inflection. We are, therefore, supposing that it seems more reasonable to assume a constraint that affects regular inflection irrespective of specific morphosyntactic properties than one which only targets e.g., regular number inflection or regular case inflection. 16 It is generally recognized in the literature that compounding is a type of derivation. Therefore, these data are legitimate challenges, though they are not compounds.
23
23
Turkish: (Cech 1995/1996:78)
(3) kurti ‘worm’ kurtja ‘worms’ kurtjandivola ‘become full of worms’
the word ‘worm’ takes the plural form and this plural in turn used within the derived verb form. On any notion of default, this is the default nominal plural strategy: “Due to the steadily increasing number of loanwords incorporated in the masc. declension class, -a is the most abundant plural type within the noun declension in Sepečides-Romani.” (Cech 1995/1996:79).
B. Tundra Nenets Genitive nominals serve as a base for some types of verb formation in Tundra
Nenets. In particular, this language possesses a process by which a nominal in the genitive plural serves as a base that hosts verbalizing suffixes. The verbs derived in this manner have either the meaning of possession, i.e., to possess X, where X is the entity denoted by the genitive plural, or to use in the capacity of X:17
NOM. SG. GEN. PL INFINITIVAL FORM OF DERIVED VERB cаβа ‘hat’ саби” саби”ць ‘have a hat ~ use in the capacity of a hat’ sava sabyi” sabyi”ts ŋум’ ‘hay’ ŋуβо” ŋуβо”ць ‘use instead of hay’ ŋum’ ŋuvo” ŋuvo”ts Table 1: Tundra Nenets Verbal Derivation
The verbal infinitival marker –ць/ts is suffixed to the GENTIVE PL. form of the nominal.
Thus, cross-linguistic morphological research reveals that synchronic language
systems do exhibit regular inflection internal to derivation, so they had better be learnable!
7. Regular inflection internal to derivation in language acquisition
Diary study and experiments in Finnish language morphological learning (Oulu dialect) by Väntillä and Ackerman 2000.
In Finnish nominal inflection there are 15 case suffixes, two numbers (SG & PL).
Compounds are right-headed (N NH), and non-heads tend to appear in various forms: they are uninflected (= NOMINATIVE18), GENITIVE-SG or GENITIVE-PL, but also can occur in various OBLIQUE cases.
17 Data from Kupriyanova et.al., 1985:139. 18 Uninflected or NOMINATIVE marked non-heads represent approximately 75% of established lexical items. As seen below, however, inflected forms are quite productive.
24
24
These alternatives are illustrated below: they represent the adult target that children must attain.
NOM.SG käsi-∅-kauppa ‘over-the-counter sales’ GEN.SG käde-n-puristus ‘handshake’ GEN.PL käs-ien-hoitaja ‘manicurist’, ELAT.SG käde-stä-ennustaja ‘hand-reader’ ADESS.SG käs-illä-seisonta ‘handstand’
Table 2: Synchronic targets in Finnish
In elicited production experiments for novel compounds 21 of 24 children aged 3;2-7;0 used genitive singular non-heads.19
This is typified by the compound lehmä-n-hoitaja ‘cow-GEN.SG-keeper’, where the
–n marking genitive singular is the sole allomorph for this case ending.
As the only possible marker, of course, it must be interpreted as the regular or default marker. The children’s behavior with GEN.PL is somewhat more complex, but instructive.
The GEN.PL has several allomorphs and therefore developing command over which
form or forms are used with which nominal is more problematic than selecting the –n for GEN.SG.
Despite this, GEN.PL allomorphs also used internal to compounds when the adult
form would contain GEN.PL, even though the child’s form did not correspond to the typical adult form.
This can be seen in Table 3 where a boy 4;10 over-regularized the GEN.PL
allomorph – itten: Child: possu-itten+syöttäjä 'piggy-GEN.PL-feed-agent' Adult: possu-jen+ särkijä Child: pullo-itten+särkijä 'bottle-GEN.PL-break-agent' Adult: pullo-jen+ särkijä
Table 3: Genitive Plural internal to compounds Finnish children’s early departures from the adult norm display forms reflecting
regular case inflection internal to lexical compounds.
This is particularly clear in their use of, e.g., the genitive singular, since the marker for this morphosyntactic category has no allomorphs, appearing only as the suffix –n, although it is also evident in the child’s reliance on a single form to mark GEN.PL as in Table 3.
19 The same children used the same genitive singular form as independent words, i.e., not as members of compounds.
25
25
These markers must be analyzed as the regular, default form and, accordingly,
children had better not be broadly prohibited from acquiring regular inflection internal to derivation.
]
Thus, in Finnish, as in Sepečides-Romani, the adult grammar possesses a pattern
apparently precluded by the relevant nativist constraint, while the evidence suggests that children too are not constrained in their development by the effects of such a presumptive constraint.
8. But, what about the he cost of abandoning representational nativism?
After all, what about the standard scientific strategy of looking at simple systems,
identifying in those systems the minimal atomic elements and combinatoric principles to produce the relevant target, and then scaling up when confronted with greater complexity?
This strategy appears consistent with the goals and practice of such programs as
Distributed Morphology (Embick and Noyer 2005, Embick 2007, among others) and Minimalist Distributed Morphology (Trommer 2003) where increasingly abstract representations have been posited.
9. Some MMG assumptions
Structural Uniformity: An apparently defective or misordered structure is actually a distorted regular form. (Culicover & Jackendoff 2005:46)
Dislocation/Displacement Property: “… the operation Move is… forced on the theory by the facts of language. Natural language suffers from what is called the displacement property (also sometime called the dislocation property) which causes elements in sentences to be moved around, apparently without that being “conceptually necessary.” (P. Seuren 2004:41)
On Move
“The empirical questions that arise are varied and complex, and it is easy enough to come up with apparent counterevidence. I will put these matters aside for now, simply assuming that UG settles the matter for now – hardly an innocuous step, needless to say.” Chomsky 1995:266 cited in Seuren 2004:163)
What counts as real countervidence, if all “apparent counterevidence” can be ignored by trusting in precisely the specific conception of UG that is being called into question by the “apparent counterevidence.”
What are the analyses of the “apparent counterevidence?”
26
26
Interface Uniformity: The syntax-semantics interface is maximally simple, in that meaning maps transparently onto syntactic structures; it is maximally uniform, so that the same meaning always maps onto the same syntactic structure. (Culicover & Jackendoff 2005:46)
Derivational Uniformity: Where possible, the derivations of sentences are maximally uniform.
One consequence of these abstract tree representations is the claim that their psychological implausibility, i.e., their unlearnability with reference to surface stimuli, even more strongly implicates their innate status.
But there are two real problems here. First, there is no guarantee (or even
reasonable expectation) that understanding, say, English morphology in this mode of analysis will shed any light on more complex systems such as Tundra Nenets (see below) or Chinantec (see Stump and Finkal 2007 – see below on word=based implicative morphology), hence problems of scalability and complexity.
(following Keresztes 1999)
What do notions like “optimal” or “economy” mean in relation to these systems? Systemic cost associated with the Mordvin innovation: objective agreement
paradigms in Hungarian, Vogul, Ostyak, and Tundra Nenets are all identical to some case/number paradigm for nominal possessives in those languages, whereas the parallelism between these paradigms is lost in Mordvin.
27
27
How do morphological theories scale up to address the complexities associated with these related agreement systems, where the Mordvin system developed from the much simpler antecedent system in Proto-Uralic?
Zwicky (1986) on two approaches to conceptualizing “the relation between a
general pattern and a special deviation from that pattern.” 20
What role, if any, do these complex systems have in determining our analytic toolkit, i.e., do we keep adapting the preferred toolkit to account for new data sets or do we permit the data sets to determine fundamental aspects of the toolkit?
How are these systems learned and does the postulation of tree-theoretic universal
structures help us explain either the organization or the learning of these systems in all their regularity and messy specificities in a single, unified way?
Second, and perhaps more worrisome, as observed in Braine 1992:79 about the far
less abstract representations imputed to biology in Pinker’s 1984 speculations about syntactic category and syntactic function development in children:
“Within the syntactic theory there is clearly a difficult scientific problem about the origin of syntactic categories: how do we get from genes laid down at conception to syntactic categories manifest two-and-a-half to three years later? Merely labeling the categories as innate does not solve the problem; it just passes the problem to biology without considering how the biologist could ever solve it… While it is certainly not now reasonable to demand anything like a complete theory, it is reasonable to expect a promissory note, and at least a sketch of an argument as to how it might eventually be redeemed.” 10. Some pivotal questions
Will the use of new quantitative and experimental tools simply confirm what MGGs has hypothesized for 50 years, namely, that there is some substantive notion of Universal Grammar (UG) provided by human biology and that, therefore, the psychological implausibility of tree-theoretic representations, i.e., they could not be learned from exposure to stimuli, are grounded in representational innateness?
What is the status of tree-theoretic representations? Are they instrumental descriptive devices or are they intended as realistic representations of the knowledge of language? If the former, aren’t there more simple descriptive devices? If the latter, are they more real than data patterns of varying levels of schematicity? What is the evidence for any belief in this domain?
What is the relevance of observable surface grammatical patterns? Are they only there to tell us about what’s really important, namely, underlying universal structures, or are they there because what grammars share is the systematic relation between surface patterns and meaning, and this systematic relation between surface forms and meanings is in the service of learnability?
20 Discussed in the T. Stewart and G. T. Stump reading on page 414.
28
28
On latter view, words and paradigms and clausal constructions are not epiphenomena derivable from hypothesized invariant abstract representations, but are themselves the very core of grammar, rendering the abstracted tree-theoretic entities unnecessary constructs, i.e. the displacement property of grammar is an epiphenomenon that arises due to theory internal uniformity assumptions?
When do we decide that first principles are simply wrong and look elsewhere? 21 (Recall the issue of “external validation” above.)
11. Results in biology demand reappraisal of linguistics: Language is imperfect and useful
“Given the relative rigor and cumulativeness of biology and linguistics, this strikes us as\ somewhat presumptuous (especially since the Minimalist Program is “still just an ‘approach’”, “a conjecture about how language works”).17 There is a simpler resolution of the apparent incompatibility between biology and Minimalism, namely that Chomsky’s recent claims about language have it backwards. Rather than being useless but perfect, language is useful but imperfect, just like other biological systems.” Pinker and Jackendoff 2005:229.
Results in linguistics are compatible with, and can profitably interact, with what is independently established in other disciplines.
12. A construction-theoretic alternative “The “construction-based” view of language that emerges from these considerations, if correct, has consequences for the study of language processing, acquisition, and, most germane to the present discussion, evolution. If a speaker’s knowledge of language embraces all the words, all the constructions, and all the general rules, coded in the same format, then there is no coherent subset of language that “delineate[s] an abstract core of computational operations”, i.e. FHC’s notion of FLN. Thus this alternative theory, which we believe has considerable empirical support, challenges FHC’s program of identifying the language-specific, human-specific part of language with the “core computational operations” and in turn (with the help of MP) with the single operation of recursion.” 2005:22222
Conceptualizations about language are consequential for theory construction, i.e., for identifying relevant empirical phenomena and designing the descriptive and explanatory toolkit.
There are different conceptualizations of construction-theoretic approaches. (An
overview presented below)
21 Elman’s emergentist schema above. 22 R. Jackendoff & S. Pinker 2005. The nature of the language faculty and it implications for the evolution of language. Cognition 97:211-225.
29
29
4. TAXONOMY OF CONSTRUCTION-THEORETIC APPROACHES (ADAPTING GOLDBERG 2006) 1. Comparison between syntactocentric and correspondence based approaches
The dimensions of differentiation most important for us concern (1) the way in which syntactocentric construction theoretic approaches are distinguishable from correspondence based proposals (Culicover and Jackendoff 2006, Kaplan and Bresnan, Bresnan 2001), (2) how different variants of the latter are classifiable into distinctive subtypes, and, most relevantly, (3) where the present proposal fits within the taxonomy of the latter approaches.
The first pass at distinguishing distinct approaches, following Goldberg (2006), is
the divide represented by (MGG)23 where constructions are essentially reifications of (innately specified) syntactic representations versus those who question representational innateness assumptions and are less committed to specific representations, i.e. that humans achieve language particular linguistic representations via experience.
Hale & Keyser24, Borer
2006, Marantz (as representative of Distributed Morphology)
Construction Grammar, Cognitive Construction Grammar, Unification-based Consturction Grammar
Status of Constructions
Representational innateness: tree-theoretic templates reflect innate specified representations
Innately guided learning: learned pairings of form and meaning/function: no priority given to particular representational assumptions.
Role of constructions
Central in the sense that contextual specification of information via tree structure representations is crucial for the derivation of non-idiosyncratic properties of wordforms and phrasal combinations
Central in the sense that wholes are construed as recombinant gestalts, i.e., wholes consist of systematic and systemic reuse of parts where configurations of parts are associated with meaning, rather than meaning being (necessarily) associated with parts themselves.
Derivational Yes - Global parsimony in the sense that all grammar domains are analyzable in terms of the same tree-theoretic assumptions.
No - Local parsimony in the sense that each grammar domain is analyzable in terms of elements and operations most appropriate for that domain.
23 See Culicover and Jackendoff 2005 for a substantive discussion of this characterization of this position in contrast to standard construction theoretic approaches. 24 Hale’s views in particular will play a role in Chapter 6.
30
30
Relatedness among constructions
Constructions are not located within networks of constructions where parts of constructions can be transparently related to (parts of) other constructions.
Constructions are located within networks of constructions where parts of constructions can be transparently related to (parts of) other constructions via default inheritance.
The syntactocentric approach can be characterized as syntagmatic, compositional,
and vertical maintaining some of the basic assumptions associated with the Post-Bloomfieldian analytic tradition:25
syntagmaticicity, i.e., linear (and hierarchical) tree-structure representations, and
compositionality, where the meaning of complex structures are assumed to be (straightforward) compositions of their component elements, with claims about scopal relations among nodes, and
verticality, where surface expressions are derived asymmetrically from either underlying representations or from a single surface representation (Albright 2001).
The correspondence approach can be characterized as paradigmatic, configurative
and horizontal:
paradigmaticity, i.e., it construes whole words and syntactic constructions as participating in networks of content theoretic alternatives, e.g., the nominative case form of a word contrasts with the genitive case form of a word, while a passive construction contrasts the active, etc., and
configurative, where the meaning of a construction is systematically associated with the ensemble of elements comprising it, i.e., it is not calculated by knowing small meaningful pieces and how they are combined, but rather by knowing that certain configurations of elements are associated with a particular meaning and that the same small pieces take on different values in different configurations.
Horizontality, where the networks of symmetrical relations yield implicative relations among surface forms.
25 See Matthews 1993 for discussion. As will be seen below, some variants also adopt a Bloomfieldian conception of the lexicon according to which it is the repository of the irreducibly idiosyncratic in language. This precludes the sort of generative lexicon/morphological component assumed in lexicalist approaches and, hence, an anti-lexicalist stance on the part of its proponents.
31
31
2. Neo-constructionist approaches26
Departing from the conventional syntactocentric view that constructions are epiphenomenal, Borer (2005) and Marantz (1997) and various proponents of Distributed Morphology (Harley 2005, Embick 2007, among others) develop a representational perspective on constructions.
Highly articulated phrasal structures – hierarchically organized cascades of binary
branching functional heads and their complements – provide the constructional context in which categorially and morphosyntactically unspecified lexical items (roots construed as encyclopedic entities with idiosyncratic meanings) acquire information responsible for their full surface appearance.
Following the general disparagement of the lexicon characteristic of post-
Bloomfieldian structuralists and syntactocentric main stream grammarians, conceives of the lexicon as the repository of what is idiosyncratic and unprincipled in language.
Relegates all productive and principled grammatical phenomena to syntactic
configurations.
Reliance on syntactic representations for e.g., lexical and morphological phenomena, explicitly designed to avoid positing the type of productive and rich lexical component central to lexicalist theories of grammar such as Lexical Functional Grammar and Head Driven Phrase Structure Grammar27; in lexicalist frameworks fully derived and inflected wordforms (of varying degrees of regularity) are composed in a lexical/morphological component which is representationally distinct from the syntactic component responsible for grouping lexical items into phrases of various types.
Latter yields a correspondence perspective on the relation between independent
components of grammar: lexicalist theories and their attendant principled analyses based on a productive lexical/morphological often ignored in standard anti-lexicalist literature.
The trend toward the elimination of lexical rules in the service of a grammar view
where lexically idiosyncratic roots become categorically and morphosyntactically specified in syntactic context cogently summarized in Harley et. al. 2005:33
26 I take this term from Borer (2005). 27 It is curious that the research tradition in lexicalist theories represented by these frameworks is generally ignored when “lexicalism” is allegedly repudiated: typical of this strategy is Marantz (1997) which declares lexicalism dead, while failing to provide a characterization of this tradition that several generations of lexicalists might identify as familiar and failing to cite any research in LFG or HPSG.
32
32
“Beginning with Baker, Johnson and Roberts (BJR) (1989), and realized most fully in the work of Hale and Keyser (1992, 1993 and subsequent work.), however, a sustained effort has been made to eliminate lexical rules and generate all argument structure alternations in the syntax, greatly simplifying the model of the lexicon. In such “constructionalist” theories, the verb is inserted into a particular complex syntactic structure, which determines the location and interpretation of each of the arguments in the verb phrase. Argument structure alternations then become a matter to be treated in the syntax, rather than in the lexicon.”28 3. Confronting Tundra Nenets
Many-to-many relations between morphosyntactic property sets and wordforms29, with the same formal pieces used for different functions in different wordforms (homonymy/syncretism).
The same members of a suffix set can be used with different lexical categories,
sometimes serving essentially the same function, and sometimes serving different functions
N V
Suffix set I Predicative Subjective
Suffix set II Possessive Objective
Table 4: Suffix homonymy in Tundra Nenets
Markers from Suffix Set I can appear on both nouns and verbs, and the inflected word functions as the predicate of the clause: set I markers reflect person and number properties of the clausal subject.
Markers from Suffix Set II similarly occur with either nouns of verbs, their
function differs within each class: they reflect person/number properties of the possessor when they appear with N, but number properties of clausal objects when they appear with (transitive) verbs.
There is a configurational dynamic whereby the same elements in different
combinations are associated with different meanings.
28 Harley, H. et. al. 2005. Determinants of event type in Persian complex predicates. Lingua, 115.10: 1365-1401 29 See Trosterud (2004/to appear) for an insightful exploration of this issue within Uralic agreement systems from a word-based morphological perspective, especially focusing on why syncretism occurs and why it occurs where it does.
33
33
Inflected words, accordingly, are best construed as recombinant gestalts, rather than simple (or even complex) combinations of bi-unique content-form mappings (i.e., morphemes)
There are isolable pieces of complex words, but it is the configurations in which
these pieces occur and the relation of these configuration to other similar configurations that are the loci of meanings relevant in morphology.
CONJUGATION NUMBER OF OBJECT MORPHOLOGICAL SUBSTEM SUFFIX SET
subjective I
sg
general finite stem
(modal substem) II
du dual object (modal) substem objective
pl III
reflexive
special finite stem
(special modal substem) IV
Table 5: Exponence of Tundra Nenets verbal forms as a function of conjugation (from Salminen 1997)
There is little one-to-one correspondence between any cells across columns: is, if
one considers the column containing the general finite stem, exemplified in (A), we see that the stem serves as base for both subjective (a) and objective conjugation (b), as well as for the singular number for objects as marked by suffix set II.
Likewise, as exemplified in (B), the dual object (modal) substem (a) hosts members
of suffix set III, but this set also serves to mark plural objects with the special finite stem (b).
Finally, the special finite stem is not restricted to plural object conjugation, since,
as shown in (c), it is associated with the reflexive conjugation and this conjugation’s characteristic distinctive use of suffix set IV.
General finite stem:
A. a. subjective: tontaø-d0m cover.I (=1sg.) ‘I cover (something)’ b. objective sg.: tontaø-w0
cover.II (=1sg/sg) ‘I cover it’
34
34
Dual obj. Stem: B a. objective du. tonta-gax0yu-n0
cover.dual.III (=1sg/du.) ‘I cover them (two).’ Special finite stem: b. objective pl. tonteyø-n0
cover.III (=1sg/pl.) ‘I cover them (plural)’ c. reflexive tonteyø-w0q cover.IV (=1sg) ‘I got covered’
Tundra Nenets has three distinct conjugation paradigms for predicates Conjugation I is for intransitives that reflect person/number agreement with their
subject argument: тухд тy’пя eрцы tuxd ty’pya yertsy fire.ablat. firewood stick out.CONJ I.3SG `The firewood is sticking out of the fire.” (Tereschenko 1965:107)
Conjugation II for transitives that reflect agreement with the person/number of the
subject and the number of the object: хора янда янмдм’ ϐаделабтаϐэда xora yanda yanmdm’ vadyelabtavyeda reindeer-bull fellow.GENPX3SG antler.ACC break.PERF.CONJII.3SG `The bull is breaking the antler’s of its companion.” (Tereschenko 1965:33)
Conjugation III for (roughly, for semantically reflexive/inchoative) intransitives that reflect agreement with the person/number of their subject.30
пи’ сырада ха”мы” pyi” syrada xa”my” night snow.GEN fall.CONJ III.3SG `Snow came down during the night.” (Tereschenko 1965:763)
Standard analyses of Tundra Nenets (Tereshchenko 1956, Kotvely 2005 ) lexically specify the assignment of a verb to these classes, with some verbs belonging to more than one class, and some belonging to all three.
30 Examples cited in Kortvely 2005.
35
35
Roots belonging to Class I and II: Class 1 (intransitive) Class II (transitive) 31
мальёсь `to break’ мальесь `to break’ malyos malyes сидёсь `to awake’ сидесь `to waken’ syidyos’ syidyes’ нердесь `to be ahead’ нертесь `to pass ahead’ nyerdyes’ nyertyes’
Roots belonging only to Class III32: ϐанебтесь `to smile’ vanyebtyes’ малесь `to glut oneself’ malyes’ нярамзь ` to redden’ nyaramz хонась `to lie down and sleep’ xonas’
Roots participating in two classes with different meanings: Class II Class III падась “to write” “to register” padas илась “to raise” “to rise” yilas малась “to collect” “to assemble” malas
Markers from paradigms for Class II versus Class III serve to distinguish specific lexical meaning e.g., “write” from “register”, i.e., markers conventionally analyzed as inflectional, are also correlated in some instances with meaning changes.
31 From Tereshchenko 1956:113 32 From Tereshchenko 1956:402-405.
36
36
4. Syntactocentric strategy for analyzing Tundra Nenets
Stipulate abstract syntactic constructions which would determine the lexical category of underspecified roots, inflect them with appropriate class markers, specify which derived verbs combine with which inflectional markers, and, when necessary, endeavor to account for meaning changes correlative with the presence of particular markers
Does this do more than redescribe the relevant distributions in terms of abstract
representations?
What does an appeal to hidden, highly articulated structure accomplish that is not accomplished by the traditional straightforward specification of lexical category and conjugation class, without the mediation of underlying structure?
Since conjugation class specification is largely, though not entirely, dependent on
the (class of) the lexeme, how is specification done without avoiding circularity? If verb conjugation class specification is associated with a categorially unspecified root, why should this be? While consistent with the assumption that idiosyncrasy is restricted to the lexicon, wouldn’t this simply presuppose that the root is really a verb, since as a noun it would not select for conjugations II and III (see below for discussion of conjugation I). Selection cannot be associated with the root after it becomes a V, since being a verb underdetermines conjugation class membership. Does becoming a verb potentiate the intrinsic conjugational options of the root?
What is the DM theory of selectional specification?
“There are some clarifications to be made about the “Root-specific” (selectional) effects identified in this classification. By assuming an approach in which Roots are category-neutral, we are not making the prediction that every Root should appear felicitously in every possible environment, e.g. in every different “lexical category” (cf. Borer 2004 for another conception). While a theory of Root/functional head combinations must be part of a comprehensive theory of competence, we cannot provide such a theory here.” Embick and Marantz 2006:7 footnote 2.
Prediction of lexicalist approach: roots are not associated with conjugation class selection nor case government, verbs are: it is not sufficient to be a verb to determine either conjugation class selection or case government – being a (particular class of) of verb determines this.
These are not problems presented by Tundra Nenets, they are artifactual problems
created by theoretical assumptions.
Tree theoretic derivation of lexical category and attendant selection of conjugation class raises more questions than it resolves in Tundra Nenets.
37
37
5. Clausal negation in Tundra Nenets
Lexical underspecification of conjugation class becomes particularly unilluminating in clausal negation.
Negative polarity in Tundra Nenets is periphrastic, consisting of an inflected
negative verb followed by a special (connegative) form of the lexical verb.33 Present/Immediate past: (Almazova 1961:183) Мань нидм’ ту’ Man nyidm’ ty’ I.NOM not.1SG come.CONNEG `I am not coming/did not come’ Past: Мань нидaмзь ту’ Man nyidamz ty’ I.NOM not.1SG come.CONNEG `I did not come’ (a while ago) Future: Мань нидм’ тут’ Man nyidm’ tyt’ I.NOM not.1SG come.CONNEG.FUT `I will not come’
The negative verb co-occurring with the connegative lexical verb can be inflected with forms from any of the three conjugation classes.
The inflected form of the negative verb is determined by the lexical verb.
For example, if the lexical verb is one which can only inflect for conjugation III,
then the negative verb can only be inflected with forms from this conjugation.
Mitchell’s (2006:234) proposal for negation with negative verbs for a subset of the Uralic languages for which the negative verb hosts agreement and tense.34
33 This is generally syncretic with the 2nd singular imperative form of the lexical verb. 34
38
38
Tundra Nenets is not treated by Mitchell and this is reasonable since she provides a discrete classification of Uralic languages into two basic types: one in which Neg is below Tense, as illustrated above, and one in which Neg is above Tense.
For Tundra Nenets Tense {past} must be above Neg, i.e., Tense apppears on the
negative verb, while Tense {future} must be below Neg, since Future tense is marked on the connegative verb.
Though there is no dedicated marker for Tense {present}, the tree-theoretic
approach requires an arbitrary designated position, rather than simply assuming a paradigmatic gap, i.e., the absence of a marker is meaningful in the context of overt markers for other tense values.
In affirmative polarity Tense {past} follows Agr, while Tense {future} precedes
Agr, and there is no special marker for Tense {present}.
In negative polarity Tense {past} follows Agr, as in the affirmative, but Agr also appears on negative in Tense {future}, even though the future marker appears on the connegative.
Powerpoint: Tundra Nenets tree-diagrams
The postulation of two different structures in order to characterize different languages as well as the need to adapt it to different tenses in the same language, namely, Tundra Nenets reinforces the status of such representations as taxonomic descriptive devices.
After category specification for both the negative verb and the connegative, the
negative verb will move to the AgrO position in order to be inflected.
39
39
However, what particular marker will be spelled out after movement, i.e markers from which of the three classes will appear and what principles determine this?
Selection of conjugation class is determined by the requirement of the connegative
verb, despite the fact that the host of this marker will be the negative verb.
This again, points to selection being determined by lexical specification for the verb (either idiosyncratically or as a member of a (sub)class), as expected on lexicalist accounts.
From a construction-theoretic perspective the distribution of tense and agreement
can be analyzed as distributed exponence associated with the lexeme for the verbal stem. (see presentation of Realization-based Lexicalism below)
Roughly speaking, a single lexeme is associated with a lexical category,
specification of conjugation class(es), and expressed either periphrastically or synthetically depending on polarity. This is, of course, to be expected if verbs are lexically specified for polarity as well as conjugation class, rather than these being specified in syntactic contexts.
Thus, participation in specific (sets of) conjugation class as well as specification of
polarity (among other properties) are associated with lexical constructions independent of syntactic context in which the lexical constructions receive surface syntactic expression.
Proposal is facilitated by the assumption that e.g., verbal morphosyntactic
information is stateable independent of syntactic configuration, but this is the guiding assumption of lexically oriented construction theoretic proposals. The present cursory proposal, then, follows from ordinary lexicalist assumptions.
In clauses with negative polarity (marked relative to affirmative polarity) there is
periphrastic exponence: negation is expressed by a verb which hosts present and past tense as well as agreement.
These three tenses and three classes of agreement markers are all properties
expressible by verbal lexemes.
When polarity is affirmative, the lexeme is expressed synthetically; when polairyt is negative this information is distributed among the pieces of the periphrastic expression of the verbal lexeme.
Also applicable for Mordvin negative polarity and periphrastic future with
transitive verbs:
40
40
Thus, the Mordvin SUBJ/OBJ agreement distributed over the pieces of the negatives
future predicate, reflects distinctions concerning SUBJ/OBJ agreement found with all other paradigmatically contrasting polarity and tense values.
6. Predicate nominals/adjectives and clausal negation
Tundra Nenets predicate nominals/adjectives host C(lass) I markers: (A. V. Alamazova 1961:62
Present Tense: Predicate Nominal: Man tyuku školoxana toxodannadm’
I.NOM this school.LOC student.CI.1SG `I am a student in this school.’
Verb: Man savavna toxodanadm’
I.NOM well study.CI.1AG `I study well.
41
41
Obvious agreement generalization: all clausal predicates participate in “agreement” and there are different Class membership options for different lexical categories.
In Ackerman & Webelhuth 1998, this is directly addressed by a positing a universal
primitive PREDICATE with variable surface expression: nominal/adjectival predicates select for Class 1 markers.
Nominal/adjectival predicates do not undergo lexical category conversion:35
a) they display defective tense expression, i.e., they are synthetic for present and past
tense, but are periphrastic for future tense, co-occurring with a future copula.36 Past Tense: Predicate nominal: Man tyuku školoxana toxodannadamz’
I.NOM this school.LOC student.C1.1SG.PAST `I was a student in this school.’ Verb: Man savavna toxodanadamz’
I.NOM well study.C1.1SG.PAST `I studied well.’ Future Tense: Predicate Nominal: Man tyuku školoxana toxodannadm’ ŋeŋudm’
I.NOM this school.LOC student.CI.1SG COP.FUT.CI.SG `I will be a student in this school.’
Verb: Man savavna toxodanaŋudm’
I.NOM well study.FUT.CI.1AG `I will study well.’
b) nominals can be modified by adjectives, displaying multiple agreement marking,
i.e., the nominal must bear a Class 1 marker and Tense {present, past}, while the adjective can also bear a Class 1 marker and tense marker.
c) in clausal negation they display different linear order, e.g., N/A NEG COP, and
different agreement pattern, e.g., N/A –AGR NEG-AGR COP than clausal negation with lexical verbs. (Kupryanova et. al. 1985:225) Мань хaнeнaдм’ нидм’ ӈa” Man xanyenadm’ nyidm’ nga” I.NOM hunter.1SG not.1SG COP `I am not a hunter.’ 35 Salminen (199?) and Ackerman and Webelhuth (1998) 36 Examples Tereshchenko 1959:100 Tereshchenko 1965:918
42
42
In sum, all of these phenomena suggest that the operative notion of
construction in syntactocentric approaches create multiple descriptive and explanatory problems in their effort to provide universal structural scaffolds:
Rhetorical claims concerning the superfluity of a rich, productive lexicon in
contrast to the necessity and sufficiency of syntactic derivation are simply illusory.
All of the construction theoretic proposals, whatever their differences,
essentially, adopt the so-called Strong Lexicalist Hypothesis, which, for present purposes, can be interpreted as follows: derived and inflected words are not created by syntactic configurations and are created by principles distinct from syntactic principles within a lexical/morphological component explicitly designed to address degrees of regularity.
More importantly, the syntactocentric approach as represented by DM, doesn’t
address the most elementary issues that a good, detailed description demands and that are conventionally addressed in lexicalist/constuction theoretic proposals.
You can’t be explanatory, if you can’t (perspicuously) describe the facts.
7. Competing correpondence based approaches CG, CCxg, RCxg Present Proposal (as a subtype
of unification-based construction grammar)
Status of Constructions
Leaned pairings of form and function
Learned pairings of form and function
Role of Constructions
Central Central
Non-derivational
Yes Yes
Inheritance Default Default Usage based Yes Yes Formalism Notation developed for
ease of exposition only Heuristic use of unification-based formalism and probabilistic analysis
Role of “motivation”
Central Central
Emphasis on Psychological plausability Psychological plausibility, viewing Language as a complex system on analogy with biological systems.
43
43
8. A characterization of constructions Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist. Goldberg 2006:5
This general characterization has led to the postulation of constructions in all domains of linguistic analysis, i.e., morphemes, words, phrases etc.
Goldberg claims that morphemes, e.g., pre-, -ing, words, e.g., avocado, and
complex words, e.g., daredevil and N-s (for plurals) are all constructions: she adopts a morpheme-based perspective on morphology, rather than the word and paradigm based perspective most compatible with construction-theoretic assumptions.
In the present latter approach the word is the construction: since cross-
linguistically there is no need to posit morphemes, morphemes or elements smaller than a word cannot be constructions.
The whole word participates in paradigms (systemic patterns) and its meaning is
configuratively determined: the meaning of a word is systematically associated with specific co-occurrences of its elements, while the same elements can take on different values in different combinations.
This view is cogently expressed in Gurevitch 2006:44 with respect to her
comprehensive analysis of Georgian morphology:37 “The meaning of the whole word licenses the exponents to be used, but there is no precondition that the meanings of the exponents have to combine to comprise the meaning of the whole. Compositionality may, indeed, emerge, but as a side product rather than a central principle, or perhaps as an effective learning strategy. The whole itself may contribute meaning to the meanings of the parts, or may override the meanings of the parts.” Gurevitch 2006:44
We assume the word-as-sign interpretation of morphology attributed to Saussure by Carstairs 2005.
37 See also Bochner 1993, Booij 2002, 2005, Harris 2007, among others). Goldberg’s position is defensible in the following way: if the morpheme is a sign and constructions are signs, then it would seem to follow that the morpheme is a construction in the relevant sense. But as argued in the WP tradition, if the markers often referred to as morphemes are not signs, then they are not constructions.
44
44
9. The role of motivation for constructions
Goldberg 2006:217 proposes the following:
“Motivation aims to explain why it is at least possible and at best natural that this particular form-meaning correspondence should exist in any given language. It simply explains why the construction “makes sense” or is natural. “
Modern restatement of Saussure’s insight concerning the distinction between
absolute and relative arbitrariness, where the latter notion can conversely be construed in terms partial motivation. He writes 1966: 133
“Everything that relates to language as a system must be approached…from this viewpoint…: the limiting of arbitrariness. This is the best possible basis for approaching the study of language as a system. In fact, the whole system of language is basd on the irrational principle of the arbitrariness of the sign, which would lead to the worst sort of complication if applied without restriction. But the mind contrives to introduce a principle of order and regularity into certain parts of themass of signs, and this is the role of relative motivation.
Specific notion of motivation Goldberg proposes is as follows: If construction A is related to construction B syntactically, then the system of construction A is motivated to the degree that it is related to construction B semantically. 1995:67
Appears to suggest that the sole or primary motivation guiding the shape of constructions is semantic.
Effort to relate language to organization in complex biological systems suggests
that this is too limited a source of motivation: constructions are motivated along many more dimensions than just semantics.
Multidimensional/multifactor view of constructions challenges proposals based on
all reductionist assumptions whether these posit conceptualist or universal syntactic configurational principles designed to characterize, in the case of the former, or constrain, in the case of the latter, classes of language structures and phenomena.
Reductionist approaches, however popular, ignore a crucial observation by Darwin
concerning empirical phenomena:38 “Our ignorance of the laws of variation is profound.”
38 See Juliette Blevins 200? for discussion.
45
45
A construction-based treatment should provide an adequate and perspicuous descriptive apparatus for analyzing the synchronic state of a grammatical phenomenon, while doing so in a manner that makes its historical and psychological motivation, at least in principle, amenable to transparent explanation.39
Foundational intuition of constructional approaches is that grammars can be viewed
as complex networks of both compositional and apparently non-compositional mappings between content, i.e., lexical information, morphosyntactic properties, and pragmatic information, and their surface expression or exponence, i.e., wordforms and syntactic phrases.
Within recent morphological theory, this view has been referred to as inferential-
realizational in the taxonomy of approaches delineated by Stump (2001) and has been profitably employed in the analysis of various types of periphrastic expressions (class presentation of Ackerman and Stump 2004, Ackerman, Stump, and Webelhuth to appear).
Most broadly conceived, both whole words as well as smaller and larger
combinations of words (i.e., syntactic constructions) are viewed as richly associated with meaning. Schematically, contentive information, C, is realized, R, by some formal expression E: R(C) = E.
A grammar consists of the rules/sehemata for specifying such realizations as well
as principles relating (parts of) constructions to one another. 40
Theoretical toolboxes are designed to scale-up from the most obviously regular and rule-governed, the more narrowly circumscribed “core” within the Chomskyan tradition of research, to phenomena exhibiting variable degrees of deviation from regularity.
In this respect we adopt the following working assumption from Culicover and
Jackendoff 2005:26:
“Peripheral phenomena are inextricably interwoven with the “core”.
39 See Harris and Campell (999?), Joseph (19??) for views of diachrony compatible with the present perspective and Tomasello (2002) for a constructional, developmental perspective on language acquisition.
40 Explore possible differences between lexical v. phrasal constructions and their relation to paradigmaticity more deeply: contrast this with the sweeping simplification in Goldberg 2006.
46
46
10. What is X doing Y as discussed in Kay and Fillmore (1999)
What’s a fly doing in my soup?
Kay and Fillmore argue that this apparently idiosyncratic and “peripheral” construction is actually quite productive and amenable to principled analysis when viewed as a composite of properties found in various more central or core constructions in the English grammar.
They conclude (1999: 30):
“The picture that emerges from the consideration of special Constructions such as WXDY is of a grammar in which the particular and the general are knit together seamlessly. The architecture of valence sets and the principles of valence satisfaction together with the inheritance of more abstract constructions by less abstract ones provide for the expression of the relevant linguistic generalizations. Constructions do not whither away, but many of them can be expressed as inheriting, and thus providing restricted instances of, more general constructions.” 11. Mulitple inheritance hierarchies capture similarities and differences
Constructionist theoreticians appeal to explicitly formulated lexical and morphological representations as well as to phrasal structures all organized according to the design features of multiple-inheritance networks used in knowledge representation systems
The essential idea underlying the relevant architecture is that recurrent properties of
entities are factored into independent dimensions of classification, while multiple dimensions can be appealed to in order to both classify and cross-classify entities.
Following Hajdu (1978: 126) we can identify certain morphosyntactic properties
that occur or do not occur within members of the Uralic family in such a manner as to define shared and distinctive properties among these related languages.41
41 Dual refers to the existence of a dual number distinction; Genitive refers to the existence of a distinct genitive case; objective Vx refers to the existence of a verbal conjugation paradigm that reflects (primarily number) properties of objects; NEG V refers to the use of negative verb forms with inflectional paradigms, rather than negative particles for clausal negation; morphotactics refers to the surface order of case (Cx) and possessive (Px ) markers.
47
47
Number Case Verbal agreement Polarity
Dual Genitive Objective Vx Neg. V Morphotactics
Px-Cx Cx -Px
Tundra Nenets W. Ostyak Finnish Mari
Figure 2
48
48
5. BLOCKING EFFECTS: EVERYTHING IS SYNTAX: (EMBICK 2007) 1. Approaches to blocking effects: (Boyd 2007 – see section 7 below)
2. The data a. Synthesis - -er: smarter *more smart42 b. Periphrasis (Analytic) - more: *intelligenter more intelligent 3. A descriptive generalization Blocking (Abstract formulation): A case in which the existence of one form prevents the appearance of another form whose existence would otherwise be expected (all other things being equal). (Embick 2007:5) e.g., glory blocks gloriousness. 4. Hypothesis: There is no morphological blocking principle – its effects derive from principles of (non-probabilistic) syntactic word formation 5. Lexicalist theories “Lexicalist theories are theories in which (at least some) derivation of complex objects, and in particular at least some word formation, takes place in the Lexicon, defined for this discussion as a generative system distinct from the syntax. “Lexicalist theories posit an architectural difference between the creation of words and the creation of phrases. For this reason, such approaches appear to face problems in cases where analytic forms, which are supposed to be constructed by syntactic derivations, alternate with synthetic forms, which are supposed to be created by rules in Lexicon. If smarter blocks more smart, as Poser (1992) and others following him have proposed, then blocking must be extended out of the lexicon and into the syntax. Since interactions between these distinct components of the grammar are not permitted in (at least typical versions of) the Lexicalist architecture, the grammar must be set up so that words can sometimes take precedence over phrases. The proposal advanced in Poser (1992) and subsequent work attempts to implement this proposal by allowing limited interaction between Lexicon and syntax—enough to allow words to block phrases under certain conditions. One could ask whether allowing even limited interaction between the ‘word’ and phrase systems is compatible with the Lexicalist program. “ 2007:2 42 Of course, more smart is an acceptable “word” in English (see Boyd 2007):
49
49
Rhetorical question that presupposes enormous flexibility in the formulation of tree-thoeretic assumptions and representations, but unalterable fixity in the assumptions of lexicalist proposals. (There has been 10 years of active research exploring exactly this issue - Realization-based lexicalism – above and below)
6. Non-lexicalist theories “On the other side of this basic division, Non-Lexicalist theories are those in which all derivation of complex forms takes place in the syntax; there is no generative Lexicon. “ 2007:2 “While motivating some sort of change in perspective in Lexicalist frameworks, alternations between ‘words’ and larger syntactic objects (phrases) seem to be directly compatible with a syntactic approach to morphology; i.e., with the Non-Lexicalist view. In such approaches, a single system is responsible for the generation of all objects, whether they surface as ‘one word’ or ‘two words’.” 2007:3
7. “Direct compatibility” has costs for anyone - Non-lexicalist global parsimony
Informationally complex morphological entities are represented by default in terms of (binary) branching syntactic structures, where each isolable grammatical meaning is identified with a syntactic head position
looking past implementations, all complex morphology is viewed as periphrastic,
since separate information is parceled out among independent syntactic pieces.
Deviations from surface syntactic independence are explicable in terms of special operations (displacement or dislocation property43) e.g., fusion or incorporation (in the simplest cases) of functional and lexical heads in syntactic trees in order to create surface (complex) synthetic wordforms.
Thus, the standard objects of traditional morphological frameworks, i.e., complex
morphophonologically integrated wordforms, are treated as special products of special hypotheses applied to syntactic representations
In contrast, demonstrable periphrastic expressions can be used to illustrate the
empirical motivation and well-foundedness of such a syntactic reduction, since such entities do, after all, consist of syntactically independent elements.
8. Costs for everyone: Lexicalist local parsimony
Strong lexicalist hypothesis requires synthetic wordforms to be fully derived and inflected within the lexical/morphological component, precluding them from being produced by the sorts of syntactic operations standard in syntactocentric proposals.
43 This property is characterized in Seuren 2004:162 as “the fact that surface structures must be taken to differ in non-trivial ways from their underlying semantic analyses.”
50
50
Roughly speaking, such theories/frameworks only permit synthetic wordforms within their compass, while periphrastic or multi-word expressions are assumed to be produced within the domain of syntax, since they are made up of syntactically independent elements.
Thus lexicalists, who prohibit syntactic word-formation for synthetic forms,
generally adopt some sort of syntactic composition operation when confronted with periphrastic expressions, despite the fact that periphrastic formations express the same information content as synthetic ones.
Each element of the periphrastic expression must be provided by the
lexical/morphological component and provision must be made to coordinate the appearance of the relevant pieces so that they co-occur and contribute their specific information to the whole composite construction. ( = Poser’s problem as cited by Embick above.)
9. Paradox for both approaches
On neither the non-lexicalist nor lexicalist approach does language really behave as expected, given basic assumptions and w/o augmentation (recall the notion of “auxiliary hypotheses” from above).
All expressions would be periphrastic w/o dislocation and all lexical compositions
would be synthetic w/o a story about syntactic co-heads.
Having seen the cost of a non-lexicalist uniform account, what assumptions would have to be made for a uniform account lexical account.
19. Words don’t block words “There is no sense in which two forms are competing with one another for existence; complex forms exist only as the outputs of derivations, and not on lists that are accessed in the course of derivations.” 2007:8 “(23) Components of Aronoff’s Blockin a. PARADIGMATICITY: The blocking effect arises because each “lexical item” has associated with it a set of cells expressing different meanings for that lexical item. Each cell may be occupied by (at most) one phonological form. b. LEXICAL RELATEDNESS: The competition that results in blocking is between words that share the same Root. c. IRREGULARITY: Irregularity is crucial to blocking. Only elements that are irregular in some respect must be listed in the lexicon, i.e. must be recorded in the “paradigm slots” (“The words that must be listed are blocked, and those which must not be listed are not blocked” (1976:45)). Therefore blocking effects may obtain only between formations each of which is “irregular” or “unproductive”. d. WORDHOOD: The objects that are entered into paradigm slots– and which thus compete with one another by virute of blocking one another– are words. “Embick and Marantz 2006:13
51
51
Words are only “outputs of derivations” – Words are not stored and, therefore, don’t compete.
Not only are words epiphenomenal, possessing less reality than the elements and
operations used to produce them, but they are really only interesting inasmuch as they produce insights into the elements and operations used to to produce. Similarly, paradigms, i.e., the patterns systems of word-relatedness are epiphenomenal (Bobaljik refs), since, after all, words don’t exist anyway.
This leads to issues concerning abstractness and the relevance of demonstrable
surface forms (either words of phrases) to the tool-kit employed in theories. 11. A tree-theoretic account of English comparative formation
52
52
12. Evidence for the abstract structure 1: Adverbial modification “… words which differ in their external syntax so so in ways which correlate with their content, not their form… In the morpheme-based… the possibility is left open that words’ morphological structure might correlate with differences in their external syntax that are not simply predictable from differences in their morphosyntactic content.” T. Stewart and G. T. Stump to appear 385.
(18)a. Mary is the mo-st amazingly smart person. b. *Mary is the amazingly smart-est person. (19) *Mary is the amazingly most intelligent person. (20) John is amazingly smarter than Bill. =The degree to which John is smarter than Bill is amazing “That is, the scope here is [amazingly [Deg smart]], not [Deg [amazingly smart]] as it is in (18).” Embick 2007:1444 13. Alternative Lexicalist Hypothesis
Since there are two readings, a simple lexicalist alternative proposal might be:
A: -er is a word affix and only scope over the word, hence [amazingly [smart DEG]] B: the second reading, [DEG [amazingly smart]], is a comparative over a phrase, hence the inability to use –er, given A, and the need to use more.
These distributions could be argued to follow from standard lexicalist assumptions
concerning the expected scope of affixes as being restricted to the domain of their host word and not having scope over a syntactic phrase.
14. Evidence for the abstract structure 2: Metalinguistic Comparatives a. John is more sad than tired. b. John is sad more than tired. c. *John is sadder than tired. Embick 2007:17 (34) John is more lazy than stupid. Embick 2007:19 (35) Assertion: The term lazy is more appropriately applied to John than the term stupid is. Embick 2007:19
44 Let’s face it. Mary is the more amazingly quicker of the two.
53
53
I assume that metalinguistic comparatives involve a silent adverbial element—given here as κ heading κP—that provides the semantic properties of “appropriateness” that are found in comparatives of this type. DegP is attached to κP in metalinguistic comparatives, and to AP in normal comparatives. Embick 2007:22 Interface Uniformity (IU) The syntax-semantics interface is maximally simple, in that meaning maps transparently into syntactic sttucture; and it is maximally uniform, so that the same meaning always maps onto the same syntactic structure (Culicover and Jackendoff 2005:6) 15. Structure of the metalinguistic comparative
What independent evidence motivates the postulation of a κP that renders it more than an ad hoc device to account for the presumed distributions?
16. What is the empirical basis of the data being modeled?
How many native speakers were consulted and what was the distribution of their judgments?
Given the prediction of categoricality that follows from the analysis, how well is
the predicted categoricality observed in (minimally) different contexts? Given the necessity for the data to be categorical, there should be experimental results demonstrating this.
54
54
17. Some relevant data a. John is lazy more than (he is) stupid. b. John is lazier, than he is stupid b. ?John is lazier more than (he is) stupid. c. Being lazy more than stupid, he eventually succeeded. d. Being more lazy than stupid, he eventually succeeded. e. Being clearly lazier than stupid, he eventually succeeded. Being clearly more lazy than stupid, he eventually succeeded. Being clearly lazier more than (he was) stupid, he eventually succeeded.
What if the generalization is that the default marking for the metalinguistic comparative construction contains “more”, hence, the acceptability of Adj more than Adj, as in (a)?
This might lead to a preference for “more” pre-adjectivally, since it is a crucial
element of the crucial and is the default in ordinary comparatives. Thus, all of surface distributions are sources of explanation.
Moreover, we have hypothesized in (16) that -er ordinarily has scope of over
words, not phrases, and since metalinguistic comparative has scope over phrases, we might expect it to be dispreferred.
18. Theoretical reification of the “fictitious agglutinating analog” (Lounsbury 1953):
“…I believe that the syntactic approach to morphology is on the right track, since it captures directly what is typical and regular in word formation – namely, affixation. To put it differently, the syntactic approach gives a correct perspective on the regularities and the irregularities of word formation in that it pays more attention to what normally happens and is not led astray by the existence of other patterns. For such aberrant cases as Germanic strong verb conjugation, one would have to assume that some additional operation applies – for example, that To and Vo are fused, so that one lexical element can spell out To and Vo simultaneously….” Julien 202:15
19. Summary Assumptions of the syntactic representation-based model
a. Morpheme-based morphological assumptions. b. Simple assumptions concerning the relation between word structure and meaning,
since semantic scope is provided by the hierarchical representation of words. c. Representations must be quite abstract, since simple assumptions concerning word
structure and meaning must constrain surface orders of morphs. d. Operations on tree structure representations are required to achieve surface true
forms of words.
55
55
6. SYNTHESIS VERSUS PERIPHRASIS – MORPHOLOGY IS MORPHOLOGY AND SYNTAX IS SYNTAX (STUMP 2002) 1. Claim: There is no morphological blocking principle - Its effects derive from a general morphological principle, i.e., Paninian determinism (more specific rules/constraints/patterns apply before less specific ones). “We propose that morphological blocking is better interpreted as “lexical blocking” and that viewed this way there is evidence that synthetic lexical items block analytic lexical items: this limits the domain of relevant blocking to the lexicon.” (Ackerman & Webelhuth 1998:78) 2. Numeral classifiers in Malto (North Dravidian)
56
56
3. The basic constraint: A noun phrase consists of a noun and a numeral phrase and the numeral phrase must exhibit a classifier
4. The periphrastic pattern:
5. But, no periphrasis with 1 and 2:
6. A synthetic pattern:
57
57
7. Another synthetic pattern: 1 or 2 humans
8. Representations for the three patterns:
9. Summary of data
58
58
10. The theoretical challenge: Can one define a single mechanism/principle that can relate different types/degrees of synthetic expression with periphrastic expression? 11. Different morphological modes of surface expression: Synthesis and Periphrasis
12. A realizational approach to “blocking”: Panini’s principle
A cascade of rules, differing in terms of their degrees of specificity, is responsible for the surface realization of words. So, the blocking effect is simply the result of how rules operated in the morphology.
The default realization of a lexeme and its morphosyntactic property set is
periphrasis and this occurs unless there is a “rule” (pattern) which specifies that some value(s) for some morphosyntactic property must be realized in a particular way and where the content cells with the most highly specified values “block” the type of realization associated with a less specified cell.
59
59
13. Words don’t “block” syntactic phrases: vaporize v. to turn to vapor
Contrary to initial hypotheses of limited interaction between the lexicon and syntax for purposes of blocking and, therefore, of the necessarily ad hoc nature of lexicalist approaches to blocking phenomena, the evidence suggests that the blocking effects are limited to elements within the same component, i.e., synthetic morphological objects block periphrastic morphological objects as constrained by Paninian determinism.
14. Summary
The right sort of lexicalism provides a principled distinction between morphological synthesis and periphrasis versus syntactic phrases. As in Embick, there is no “blocking principle”, there is simply the Paninian principle applied to various expression types in the lexical/morphological component.
The “blocking” effect crucially refers to whole words and within paradigms, i.e.,
systems of whole words.
It is words as (complex) wholes, but not as made up of morphemes, that is crucial, and it is the whole pattern that words participate in that is crucial.
16. An evaluation of both uniform accounts
Both proposals based on categorical distribution of data. Neither addresses optionality nor, consequently, provides a uniform treatment for
blocking and optionality.
5. COMPETITION AMONG SURFACE EXPRESSIONS: A UNIFIED TREATMENT OF BLOCKING AND OPTIONALITY (BOYD 2007) 1. General characterization of the phenomenon
Two factors: identify the grammatical expressions options and the factors determining choice among these options.
Given a set of morphological expressions E containing (e1…en), where the
exponents (e1…en), can be synthetic, periphrastic, supragmental, etc, and where E realizes a meaning µ, for each lexeme L which can express µ there is a probabilistic relation P concerning how L(µ) maps into E.
For a meaning µ that is mapped onto a set of competing expressions E = {e1,
e2…en}, the probability that one expression (en) will be used over the others is given by P(e1) + P(e2) + P(e3) + … + P(en) = 1
60
60
General statement that P(en) is inversely related to the sum of P(ei) values. As other expressions become more probable, P(en) becomes less probable; as other
expressions become less probable, P(en) becomes more probable.
The Blocking Hypothesis and the Optionality Hypothesis offer different interpretations of the values that P(ei) can take in Equation 1, with the understanding that the sum of the different P(ei) values can never exceed 1.
P(ei) = {0, 1} (Blocking) P(ei) = [0, 1] (Optionality)
Under the Blocking Hypothesis, this entails that P(e3) will be zero if either P(e1) or P(e2) are 1, and that P(e3) will be one if both P(e1) and P(e2) are zero.
Under the Optionality Hypothesis, since P(ei) can take on any value from zero to
one, inclusive, the relationship between the three competing expressions is not one in which one expression blocks the other two. Instead, it is possible for all three to be mapped onto M on different occasions.
To the degree that languages tend to limit optionality, there should be an
explanation for this, but we don’t want our theories to preclude optionality. The best account will be the one that treats categoricality and optionality uniformly,
but also tells a plausible story about why there is a tendency for grammars to exhibit categorical effects.
61
61
2. Optionality in morphology
Bulgarian: Manova, S. Paradigm linkage and Bulgarian verb inflection. Ms.; Spencer, A. Periphrastic paradigms in Bulgarian’. In Uwe Junghanns & Luka Szucsich (eds.) Syntactic Structures and Morphological Information. Berlin: Mouton de Gruyter, 249—282, 2003.
62
62
Erza Mordvin definite singular declension for moda `land, country’45: R. Bartens 1999. Mordvalaiskielten rakenne ja kehitys. Suomalais-Ugrilainen Seura.
Nom. modaś Gen/Acc. modańť All/Ill. modańťeń Abl. modadońť El. modastońť ~ modańť ejste In. modasońť ~ modańť ejse Pro. modavańť ~ modańť ezga
English comparatives: British National Corpus Adjective Identified From BNC P(er) angry BNC 0.578 complete BNC 0.014 costly Mondorf (2003) 0.146 cozy Mondorf (2003) 0.375 crazy Mondorf (2003) 0.818 cruel Mondorf (2003) 0.522 deadly BNC 0.222 deep BNC 0.989 empty BNC 0.579 fit Mondorf (2003) 0.928 free Mondorf (2003) 0.852 friendly BNC 0.426 gloomy BNC 0.786 hazy BNC 0.5 hungry Mondorf (2003) 0.879 lucky Mondorf (2003) 0.5 mellow BNC 0.5 proud Mondorf (2003) 0.867 risky Mondorf (2003) 0.494 salty BNC 0.5 sleepy BNC 0.455 sober BNC 0.021 sorry BNC 0.531 thorny BNC 0.8 unhappy BNC 0.414 warm BNC 0.993
45 Mokša Mordvin displays only periphrastic expression for ablative, inessive, and prolative.
63
63
Distribution of adjectives by comparative frequency and probability of –er inflection:
“When I was a kid, one time I had an old-maid teacher that used to tell me, “Buzz, you’re the thickest-headed dunce the school.” But I noticed that she told this a whole lot oftener than she used to tell the other kids how smart they were…” Sinclair Lewis It can’t happen here. 1935:83.
What the comparative forms found in derivatives such as thickest-headed and how might this relate to independent forms of comparatives?
Adjective-N-ed
Standard Comparative Comparative dim-witted dimmer * dimmer – witted more dim-witted light-hearted lighter * lighter-hearted. more light-hearted. tight-fisted tighter * tighter-fisted more tight-fisted cautious-minded more cautious more cautious-minded
64
64
Hungarian
Similar distributions of comparative marker when the superlative prefix leg- is
used?
There’s robust optionality in English comparative formation, but this is entirely ignored in proposals based on categorical judgments.
Adequate characterization of native competence of English comparatives requires
modeling of optionality.
This suggests the need for probabilistic models of word-based morphology. 3. Generalizing what we’ve seen
The previous proposals exemplify contrasting classes of grammar analysis.
65
65
6. SYNTAGMATIC VERSUS PARADIGMATIC APPROACHES TO GRAMMAR 2. The basic contrast
“The syntagmatic relation is in praesentia. It is based on two or more terms that occur in a an effective series. Against this, the associative ( = paradigmatic) relation unites terms in absentia in a potential mnemonic series. From the associative and syntagmatic viewpoint a linguistic unit is like a fixed part of a building, e.g., a column. On the one hand, the column has certain relation to the architrave that it supports; the arrangement of the two units in space suggest the syntagmatic relation. On the other hand, if the column is Doric, it suggests a mental comparison of this style with others (Ionic, Corinthian, etc.) although none of these elements is present in space; the relation is associative.” Saussure 1966:123. 3. Complex forms “A unit like painful decomposes into two subunits (pain-ful), buth these units are not two independent parts that are simply lumped together (pain + ful), The unit is a product, a combination of two interdependent elements that acquire value only through their reciprocal action in a higher unit (pain x ful). The suffix is non-existent when considered independently; what gives it a place in the language is series of common terms like delight-ful, fright-ful, etc….The whole has value only through its parts, and the parts have value by virtue of their place in the whole.” Saussure 1966:128
It is possible identify pieces and specify their arrangement without necessarily attributing meaning to each of the pieces or assuming that the whole is simply the sum of such pieces (cf. below for the relation between syntagmatics, morphemes, and compositionality.)
Of course, in some language(s) pieces of words can be construed as meaningful,
especially if there is a bi-unique relation between content and form. (See below for examples.)
66
66
4. Paradigmatic and syntagmatic analysis in morphology: Saussure 1966:129
Examples: un-do; un-stick; move an object; un-pick Examples: make; redo; counterfeit
Saussure’s analysis has a syntagmatic aspect, but without morphemes. Powerpoint: J. Hay 2003:30 & 155 - on degrees of coalescence between affixes and stems. 5. Paradigmatic and syntagmatic analysis in syntax: Paul 1970/1890 “In the the process of naturally mastering one’s own mother-tongue no rule, as such, is given, but only a number of examples. We hear gradually a number of sentences which are connected together in some way, and which hence associate themselves into one group…It is precisely because no abstract rule is laid down that no single example suffices, but only a group of examples whose special content (sentence specific words – FA) appear a matter of indifference.” Paul 1979:98 6. Compare two general hypotheses concerning grammar analysis:
H1: syntagmatic/compositional /vertical Mainstream generative grammar
H2: paradigmatic/configurative/horizontal Construction-theoretic
Each perspective encourages certain questions, while ignoring others, i.e., the creation of (classes of) wordforms/sentences versus wordforms/sentences as wholes in relation other entities of the same sort.
67
67
7. SYNTAGMATIC/COMPOSITIONAL/VERTICAL MORPHEME-BASED APPROACHES 1. Some “traditional” observations By the morphology of a language we mean the constructions in which bound forms appear among the constituents. By definition, the resultant forms are either bound forms or words, but never phrases. Accordingly, we may say that morphology includes the constructions of words and parts of words, while syntax includes the construction of phrases. Bloomfield 1933:207. The object of study in morphology is the structure of words, and the ways in which their structure reflects their relation to other words – relations both within some larger construction such as a sentence and across the total vocabulary of a language. Anderson 1992:7
The analysis of discontentedness (following Anderson 1992:10) a. [N [A [N dis [N [A content ] ∅ ] ] ed ] ness] b. [[[ OPPOSITE-OF[[SATISFIED]STATE]]CHARACTERIZED-BY]STATE-OF-BEING]
Simple agglutination seems reasonable: Hungarian
Simple agglutination seems less reasonable: Estonian (Baayen & Hay 2005, Blevins 2006, Hughes and Ackerman 2003)
Powerpoint: Estonian nominals
Simple agglutination seems unreasonable: Georgian (Gurevitch 2005) Powerpoint: Georgian nominals
While e.g., Hungarian reflects one possible, particularly transparent, strategy for encoding morphosyntactic notions, neither Estonian nor Georgian provide evidence for such transparency.
Should grammatical theory privilege one surface encoding over any other? If so,
which? And, why?
If not, what relates languages to one another?
68
68
Every language possesses patterned ensembles of forms that systematically and reliably correspond to meanings/functions, i.e. given this pervasive and unifying pattern-based principle, there is no need to privilege one surface strategy over another. (See Sapir below)
2. The agglutination belief prevails
Current assumptions originating from a simpler time “[In Chomsky and Halle 1968] tense is not seen as a feature which locates a word in a paradigm. Instead the verb is said to contain a formative `past’ just like the morpheme`past’ of Syntactic Structure. This the terminal element of one branch of a phrase structure tree, just like any other formative Both lexical and grammatical formatives `will…be represented as feature matrices of an abstract sort’…In general, we are told, there are discrepancies between a surface structure which is the optimal output of the syntactic component and one which serve as an optimal input to the phonological component. We, therefore, need an intermediate set of rules, called `readjustment rules’, to remove them…`The readjustment rules would replace `past’ by d’, we are told, except in strong verbs: for `d’ we are to read `the distinctive feature matrix representing [d].’ The implication is that `past’ itself is no more than a complex of syntactic features, perhaps a single feature [+past]. The formative which is the input to the readjustment rules is therefore, in this instance, essentially the morpheme of the Post-Bloomfieldians. It is a unit of form, located at a specific position in a sequence of forms; but in itself without phonetic content.” Matthews 1993:93
Against some indeterminate variant of lexicalism that repedicates all lexicalist proposals (remember the difficulty of theory comparison!)
“However, there exist complex word forms whose syntactic properties cannot be made to follow from a lexicalist analysis. The most comprehensive treatment to date of such word forms is Baker (1988a); further examples are found in many places in the literature. These forms could be seen as compelling evidence in favor of a syntactic approach to morphology. It can further be argued that even in cases where the syntactic analysis is not forced by the raw facts, such an analysis yields the simplest and most principled account of the word forms in question. Hence, the syntactic approach is preferable because it is superior when it comes to explanatory adequacy. Then, once the view is taken that complex words can be formed in syntax, it follows that the elements that words are made up of must be allowed to have their lexical and syntactic representation. Consequently, word formation must be basically concatenative… The model of word formation that will be defended in this work is based on two fundamental assumptions: first, that all morphemes are listed separately in the lexicon, and second, that they are inserted separately into syntactic structures.” Julien 2002:12 3. Morpheme-based morphology
Morphology, adapting the tradition of American structuralists and certain ideas of Europeans such as Badouin de Courtenay (Anderson 1985, Hockett 1987, Matthews 2001, among others), focuses on the small meaningful pieces which can be composed into complex words: (morpheme-based morphology)
69
69
Syntagmatic: it emphasizes the linear (and hierarchical) combination of constitutive elements
Compositional: it endeavors to derive the meaning of the whole word from the meanings of its isolable parts.
Vertical: it is derivational or proof-theoretic, positing an invariant underlying or single surface structure form from which a target structure can be asymmetrically derived.46
4. The nature of language (here morphology, but, by hypothesis, syntax as well) on the “morpheme to utterance view”
Familiar efforts to identify small meaningful pieces (morphemes) as well as the rules (morphotactic and phonological) that yield the legal combinations of these entities evident as surface wordforms.
Global parsimony concerns, i.e., Uniformity assumptions, regarding the minimal elements (either underlying or surface-based) and operations required to construct or build wordforms.47
Neither surface wordforms nor the systematic patterns of surface alternations that whole words participate in are construed as basic units of grammatical organization. Rather, the surface words of particular languages are useful to the degree that they provide insight into underlyingly invariant atoms of analysis and combinatoric operations that can account for (variable) surface expressions.
Surface patterns of both words and networks of words (i.e., relations between surface alternants) are not considered to be proper objects of linguistic analysis, while the abstract elements and operations responsible for constructing these ephemera are.
Questions concerning the psychological or biological basis of these constitutive elements and operations.48
46 Seuren 1998 and Itkonen 2005:89 on vertical v. horizontal approaches. 47 It should be noted that in practice these parsimony considerations have mostly been restricted to establishing the set of primitive representations, e.g., binary branching, and operations on tree-theoretic representation with uniform phrasal expansions. Reliance on this small inventory of representations and operations has led to increasingly abstract and some would suggest unparsimonious representations of rather simple surface expressions. 48 See S. R. Anderson’s characterization of competing approaches to phonological analysis below.
70
70
8. SUSPICIONS ABOUT THE SYNTAGMATIC/COMPOSITIONAL AND MORPHEME-BASED APPROACH 1. The great agglutinative fraud ( Hockett 1987) “Typical of alternation in Potawatomi is the difference between the bare noun /msən’əkən/ `paper’, and the allocated form /nməsnə’kən/ `my paper’. The latter not only adds a prefix /n-/ but also switches the vowels around to different positions relative to the consonants… But suppose we write, morphophonemically | məsənə’əkən | and | nəməsənə’əkən |, and formulate a rule for the deletion of some of these vowels in actual speech: the vowel before a final consonant stays; otherwise, starting from the beginning of the word, the odd numbered vowels drop and the even ones stay. Now, we can chop the whole words, in morphonemic transcription, into recurring meaningful pieces of INVARIANT shape; a prefix |nə-| `my’, a root |məsənə-|, and a noun forming suffix |-‘əkən|.” Hockett 1987:81 “In 1953 Floyd Lounsbury tried to tell us what we were doing with our clever morphophonemic techniques. We were providing alternations by devising an “agglutinative analog” of the language and formulating rules that would convert expressions in that analog into the shapes in which they are actually uttered. Of course, even such an agglutinative analog , with its accompanying conversion rules, could be interpreted merely as a descriptive device. But it was not in general taken that way; instead, it was taken as a direct reflection of reality. We seemed to be convinced that, whatever might superficially appear to be the case, every language is “really” agglutinative. The extreme of this tendency, for me its reduction ad absurdum, was a proposed way of handling the English plural men. To recognize this as the plural of man is, of course, correct… But the proposal went far beyond that: in order to make it conform to the agglutinative ideal we said that it is really a sequence of two morphs, the first one, /men/ being an allomorph of the morpheme {man}, while the second, of phonemic shape zero, is an allomorph of the noun-plural morpheme.” Hockett 1987:84 2. Chomsky on the superiority of Word and Paradigm models "In a traditional grammar, a particular occurrence of a Noun would be described in terms of its place in a system of paradigms defined by certain inflectional categories, namely, categories of gender, number, case, and declensional type. Each of the categories constitutes an independent `dimension' of the paradigm, and each word has a particular `value' along each of these dimensions. Thus the word Bruder would be characterized as Masculine, Plural, Genitive, and belonging to a certain declensional class with Vater, Mutter, etc. Chomsky 1965:17
71
71
3. Paradigm-based representation of German nominals:
I(Fem) II(Masc) III(Neut) IV(Masc)
Sing Plur Sing Plur Sing Plur Sing Plural
Nom Frau Frauen Vater Väter Schaf Schafe Mensch Menschen Acc Frau Frauen Vater Väter Schaf Schafe Menschen Menschen Dat Frau Frauen Vater Vätern Schaf Schafen Menschen Menschen Gen Frau Frauen Vaters Väter Schafes Schafe Menschen Menschen
‘woman’ ‘father’ ‘sheep’ ‘person’ Table 1: German Duden 2005: 197
4. The Word and Paradigm (WP) approach:
R1 - 1. [C..V..C] [C..V.. C] / DC1… PLURAL [+F] Rule: Front stem V when the word has the features [DCα] & [PLURAL]. (The absence of a realizational rule accounts for ∅ realization.)
"More generally, the often suppletive character of inflectional systems, as well as the fact that (as in the present example) the effect of inflectional categories may be partially or even totally internal, causes cumbersome and inelegant formulation of rules when the representations to which they apply are in the form of (31). However, suppletion and internal modification cause no special difficulty at all in the paradigmatic formulation…" 5. Morpheme-based approach: "The characteristic method of analysis of modern linguists is rather different from the traditional approach… In place of the traditional categories (our features), this approach would substitute morphemes. Thus Brüder in (30) would be represented in the manner of (31), in a completely “item-and-arrangement” grammar.”
Morpheme-based representation of German nominals: Chomsky 1965:173 (= ex. 31) Bruder – D(ECLENSIONAL) C(LASS) - MASCULINE - PLURAL – GENITIVE In contexts such as that defined in (31), all 4 morphemes are realized as ∅
(= 30) N
[GENDER] [NUMBER] [CASE] [DECL. CLASS] …
72
72
6. Evaluation of morpheme-based analysis: The cost of null elements " Representations such as (31) are clumsy for a grammar based on transformations… For one thing, many of these `morphemes' are not phonetically realized and must therefore be regarded, in particular contexts, as zero elements. In each such case a specific context sensitive rule must be given stating that the morpheme in question is phonetically null. But this entire set of rules is superfluous and can simply be omitted under the alternative paradigmatic analysis." Chomsky 1965:173
Morpheme-based proposal entails the need to posit phonetically null elements.
7. Another cost: The need to establish morpheme order (morphotactics) "Finally, notice that the order of morphemes is often quite arbitrary, whereas this arbitrariness is avoided in the paradigmatic treatment, the features being unordered." Chomsky 1965:174
What is the theory independent evidence that posited underlying orders are anything but aribitrary, given that there is enormous variability among languages and even within languages?
Is morphotactics constrained by Baker’s Mirror Principle a part of universal
grammar design or does it reflect Bybee’s semantico-functional tendency of relevance?
8. Types of variant orders I. Different ordering of the “same” morphosyntactic/derivational markers:49
Multiple prefixation in Chichewa50 (Bantu): Hyman 2003 a. main (root) clause: NEG- SUBJ- TNS- ASP- OBJ- stem si- ti- dzá- ngo- mú- ményá b. subordinate (non- ti- sa- dzá- ngo- mú- ményá root) clause: SUBJ- NEG- TNS- ASP- OBJ- stem ‘we will not just hit him’
49 See T. Stewart and G. T. Stump 12.7.1 for a formal means to address variable marker ordering: they do not address optional ordering. 50 -dzá- ‘future’, -ngo- ‘just’
73
73
Moro (Nilo-Kordofanian): Gibbard 2006
2nd person pl. OBJ after verb root in PERF, but before in IMPERF and SUBORD: the –o marker of PERF elided when OBJ follows root.51
1st person pl. incl. OBJ after verb root in PERF, but part of it before in IMPERF:
51 Note glossing error in 2b: 2pl actually precedes root.
74
74
3rd person pl. OBJ always after the verbal root:
Chintang: B. Bickel. et. al. 2006
(1) Free prefix ordering in Chintang52
a. u-kha-ma-cop-yokt-e 3NS.A-1NS.P-NEG-see-NEG-PST
b. u-ma-kha-cop-yokt-e 3NS.A-NEG-1NS.P-see-NEG-PST
c. kha-u-ma-cop-yokt-e 1NS.P-3NS.A-NEG-see-NEG-PST
d. ma-u-kha-cop-yokt-e NEG-3NS.A-1NS.P-see-NEG-PST
e. kha-ma-u-cop-yokt-e 1NS.P-NEG-3NS.A-see-NEG-PST
f. ma-kha-u-cop-yokt-e NEG-1NS.P-3NS.A-see-NEG-PST All: ‘They didn’t see us.’
Nhanda (Juliette Blevins, 2001), Kusunda (Watters, 2006), and Madurese (Stevens, 1971) Rice (2000) in various Athabaskan languages; Muysken (1986) in Quechua. 52 NS = nonsingular, 1 = first person, 3 = 3rd person, A = actor, P = primary object, Neg = negative, Pst = past.
75
75
Filomeno Mata Totonac (Totonaco-Tepehua) (McFarland 2007)
Affix order zones:
Fixed Free Scopal ROOT Fixed Free Fixed
Fixed order zones--mostly inflectional morphemes (TAM and person agreement), also deictics and the reflexive marker.
Middle freely ordered zone--prefixes include applicatives, causative, others;
suffixes are primarily quasi-inflectional aspectual morphemes, also benefactive and progressive
Inner zones -- first five prefix positions ordered on scopal principles, while the first
two suffix positions occur in fixed order.
76
76
II. Language internal differences in orders keyed to morphosyntactic property sets
Permian (Uralic) suffix order for possessive person/number markers (PNM) and case markers (CM): Udmurt: (Kel’makov & Hännikäinen 1999:77, Suihkonen 1995) – 15 nominal cases Pattern 1: STEM - PNM - CM, where CM = { dative, genitive, ablative, abessive/caritive, adverbial, approximative.}
[BOY; CM:dat; PNM: α] pi, where α = PERS:{1,2,3}, NUM = {sg,pl}, ⇒ PF53 [BOY; CM:dat; PNM:1sg] (pi) ⇒ pielị `to my boy' PF [BOY; CM:dat; PNM:2sg] (pi) ⇒ piedlị `to your boy' PF [BOY; CM:dat; PNM:3sg] (pi) ⇒ piezlị `to his/her boy' PF [BOY; CM:dat; PNM:1pl] (pi) ⇒ pimịlị `to our boy' PF [BOY; CM:dat; PNM:2pl] (pi) ⇒ pidịli `to your boy' PF [BOY; CM:dat; PNM:3pl] (pi) ⇒ pizịli `to their boy'
53 PF, as in Stewart and Stump, stands for paradigm function and here states the full feature set associated with a grammatical word. See also their section 12.7.1.
77
77
Pattern 2: STEM - CM- PNM, where CM = {instrumental, inessive, illative, elative, translative, egressive, prolative}
[BOY; CM:instr; PM: α] (pi), where α = PERS:{1,2,3}, NUM = {sg,pl}, ⇒ X PF [BOY; CM:instr; PNM:1sg] (pi) ⇒ pienịm `with my boy' PF [BOY; CM:instr; PNM:2sg] (pi) ⇒ pienịd `with your boy' PF [BOY; CM:instr; PNM:3sg] (pi) ⇒ pienịz `with his/her boy' PF [BOY; CM:instr; PNM:1pl] (pi) ⇒ pienịmị `with our boy' PF [BOY; CM:instr; PNM:2pl] (pi) ⇒ pienịdị `with your boy' PF [BOY; CM:instr; PNM:2pl] (pi) ⇒ pienịzị `with their boy'
Pattern 3: Either order, where nominal is [CM: term; NUM: sg].54 (Csucs 1990:39)
[FIELD; CM:term; PM: α] (busι⁄), where α = PERS:{1,2,3}, NUM = {sg,pl}, ⇒ X
PNM-CM ~ CM-PNM PF [FIELD; CM:term; PNM:1sg] (busị) ⇒ busịjeoź ~ busịoźam`up to my field' PF [FIELD; CM:term; PNM:2sg] (busị) ⇒ busịjedoź ~ busịoźad `up to your field' PF [FIELD; CM:term; PNM:3sg] (busị) ⇒ busịjezoź ~ busịoźaz`up to his/her field' PF [FIELD; CM:term; PNM:1pl] (busị) ⇒ busịmioź ~ busịoźamị `up to our field' PF [FIELD; CM:term; PNM:2pl] (busị) ⇒ busịdioź ~ busịoźadị`up to your field' PF [FIELD; CM:term; PNM:3pl] (busị) ⇒ busịzioź ~ busịoźazị`up to their field' Volgaic (Uralic) suffix orders: Mari (Alhoniemi 1985:76, Luutonen 1997): 9 case distinctions in Meadow(-Eastern) Mari.
A. Suffix order for person markers (PM) - possessive paradigm - and case markers (CM) Pattern 1: STEM - PNM - CM, where CM = { genitive, accusative}
[EGG; CM:gen; PNM: α] muno, where α = PERS:{1,2,3}, NUM = {sg,pl}, ⇒ PF [EGG; CM:gen; PNM:3sg] (muno) ⇒ munәžn `of his/her egg'
Pattern 2: STEM - CM- PNM, where CM = { inessive, illative, lative, comitative}
PF [EGG; CM:inessive; PNM:3sg] (muno) ⇒ munštәžo `in his/her egg'
54 According to K&H 1999:167, the terminative is used only in the singular in the literary language (see below.)
78
78
B. “Free” affix order: 55 CM-PNM v. PNM-CM; PL-PNM v. PNM-PL56
No evidence for semantic difference correlated with morpheme order variability (see
Pattern 3: Variable order, where nominal is [CM: {dat, modal57}; NUM: pl; PM: α], where α= PERS:{1,2,3}.
A. Orders for CM & PNM:
[EGG; CM:dat; PNM: α] muno, where α = PERS:{1,2,3}, NUM = {sg,pl} ⇒ X or Y
PNM/CM ~ CM/PNM PF [EGG; CM:dat; PNM:3sg] (muno) ⇒ munәžlan ~ munәlanže `to his/her egg'
B. Orders for PL & PNM: [APPLE; CM:nom; NUM; pl; PNM: α] muno, where α = PERS:{1,2,3}, NUM = {sg,pl} ⇒ X or Y PNM/PL ~ PL/PNM PF [APPLE; CM:nom; NUM; pl; PNM:1sg] (olma) ⇒ olmamβlak ~ olmaβlakem `my apples' III. Fixed affix ordering and scope58
Morphosemantic mismatches “Inflections may be morphologically realized `inside of’ an outermost layer of category preserving derivation/compounding, but are logically `outside of’ this layer.” Stump 1991:686 55 The most detailed study of this phenomenon is Luutonen 1997 with conclusions and results based on both text frequencies and psycholinguistic experiments on native speakers in different dialects of Mari. I ignore here the results based on combinations of all three affix types. Despite the preferences identified above in context-free forced choice experiments and according to other criteria, it is important to note the following: "…the degree of inconsistency in the responses of the individual subjects was remarkable. In general, a subject very often reacted differently to the structurally similar test case distributed on different pages of the questionnaires. -Luutonen 1997:135. This, of course, suggests that variability of affix order is both a prevalent feature of Mari and psychologically real. 56 Judgments based on context-free preference tasks with native speakers indicate that morphotactic variation is stronger in Meadow Mari than in Eastern Mari. There is more variation in the order between PL & PNM than between CM & PNM and order for CM & PNM indicates variability and significant preference of CM-PNM order only for PNM: { 1pl, 2pl} and virtual fixity of PNM-CM order for all other values of PNM. 57 There is some indication that there is more limited variability in certain dialects with the accusative and genitive case-markers. Luutonen 1997:129. 58 See Stewart and Stump 12.7.2.
79
79
Study of 32 Bantu languages with respect to the ordering of causative, applicative, reciprocal,
“linear position must be stipulated independent of their semantic scope.” Good 2005:6.59
Periphrastic expression:
Scope of causative (Ackerman 2004) Hungarian: PV … V stem- ( CAUSATIVE) -(POTENTIAL)-TNS-AGR
Synthetic expression:
Scope of causative (Hewitt 1979) Abkhaz: SUBJ/OBJabs-IO-((PV)-(POTENTIAL)-(NEGATIVE)-(SUBJerg)-(CAUSATIVE)-Vstem
59 Jeff Good 2005. Reconstructing morpheme order in Bantu. Diachronica 22.1:3-57. Also Hyman 2003, …
80
80
Scope of tense (Stump 2001) Sanskrit: PV-TNS-Vstem
Scope of aspectual marker
Slave: (Rice 2003)
Morphotatic problem: the order of markers from simple verbs is the same as the order of markers for the complex verbs.
Semantic scope problem: the semantic scope of markers always over semantic unit
expressed by the verb, irrespective of whether the expression the preverb/prefix are contiguous with the verb stem.
81
81
IV. There are no obvious classical morphemes, although there is an obvious order.
Dhaasanac (Galab). Sasse, H.J. 1974. Notes on the structure of Galab. Bulletin of the School of Oriental and African Studies. Vol.37.2:407-438. See also Petsova 2006:10.
Two basic forms (A & B) distributed throughout the paradigm: small number of “selectors” indicating person/tense/mood used in conjunction with basic forms to provide determinate combinations.
82
82
(Adapted from the presentation in Gibbard 2007) 9. Reversal: Ful (Baerman Language 2007)60 Here we find that all nouns belonging ro the personal class form the plural by changing the initial g, i, d, b, h, ch, and, p to y (or w), y, r, w, h, s, and f respectively; e.g., jim-o `companion’ yimm’-be “companions"; pio-o ‘beater," fio’-be "beaters." Curiously enough, nouns that belong to the class of things form their singular and plural in exactly the reverse fashion, e.g., yola-re “grass-grown place,', jola-je “grass-grown places"; fitan-du soul” pital-I "souls." 1921:75
Beyond reversal, the phenomenon of gender class marking in languages with large inventories of genders is essentially paradigmatic, since marking patterns are keyed to pairs of e.g., singular v. plural.
10. The utility of the paradigm
Whereas the arbitrariness of assuming a universal order is avoided in the paradigmatic proposal, fixed order of markers, as well as the absence of classical morphemes, is compatible with a paradigm-based approach, while it is also predicted that (1) alternative orders could be characteristic for subparadigms, hence, the correlation of different morphosyntactic property sets and surface realization and (2) ensembles of properties associated with words, rather than classical morphemes, could be reliably distinctive.
11. A paradox from the perspective of history "I know of no compensating advantage for the modern descriptive reanalysis of traditional paradigmatic formulations in terms of morpheme sequences. This [= morphemic analysis - FA] seems, therefore to be an ill-advised theoretical innovation." -Chomsky 1965:174 60 See Baerman Language 2007 for a recent investigation into reversals.
83
83
"It seems that in inflectional systems, the paradigmatic analysis has many advantages and is to be preferred … It is difficult to say anything more definite, since there have been so few attempts to give precise and principled description of inflectional systems in a way that would have some bearing on the theoretical issues involved here." Chomsky 1965:174. 12. Ignoring Chomsky’s assessment
There have been numerous attempts to "give precise and principled descriptions of inflectional systems" within Chomskyan syntactocentric frameworks over the past few decades and none of them have avoided the pitfalls cited by Chomsky himself.
Neither the syntactocentric61 approaches which adopt the "Mirror Principle" nor
Distributed Morphology solve the morphotactic ordering problems or the abstractness problems, i.e., the need to identify underlying order, identified by Chomsky.
Instead, they have embraced abstractness both in the postulation of "morphemes"
that are ordered and in the readjustment rules required to achieve surface exponence for them (see Marantz 1997, Embick and Noyer 1999, McFadden 2003, Embick and Marantz 2006, Embick 2007, among others, for particularly imaginative moves) and have, in several instances, argued that these are virtues whose very unlearnability from surface evidence suggests the innateness of the proposed constructs.62
That the abstractness of positing highly specified underlying representations with
phonetically null elements doesn’t appear to be worrisome today doesn’t mean that it shouldn’t be a concern.
Nor, do they follow the theoretical ramification implied in Chomsky’s conclusions,
namely, that morphology is quite distinct from syntax (see P.H. Matthews, Robbins 1959, Bach 1996, among others)
Instead, morphology is construed as a kind of syntax with a universal ordering of
functional categories and the (considerable number of) deviations from it are "explained" by conditions of trees and facilitated by appeal to the sort of unconstrained abstractness eschewed by Chomsky.
61 I take this term from Jackendoff 1997:15 where is used to refer to those theories in which "the fundamental generative component of the computational system is the syntactic component…" 62 Of course, arguments from the mere subtlety of phenonema, either syntactic or syntacticized morphology, to their probable source in some innate principle are quite questionable given the evident success of humans to learn the subtleties and complexities of such phenomena as word meaning and our evident talent for identifying transitional probabilities (Saffrin et. al. 1996), patterns (Elman et.al. 1996) and algebraic learning (Marcus G. et. al. 1996), Tomasello 2003.
84
84
Where some have posited tree-theoretic universality of linear orders, leading to such claims as the Mirror Principle and consequent categoricality in the relation between meanings and the elements that encode them, others have suggested that observable orderings represent a tendency based on functional/semantic motivations and diachonic changes (Bybee 1985, Stump 2001, among others).
What are the elements that are ordered? Are they discrete and atomic?
Linguistically relevant notions amenable to discrete categorizations and this
categorization is cross-linguistically valid, even if there may be questions as to its universal or language particular encoding: thus, discrete thematic roles permit the formulation of UTAH or LFG’s thematic role hierarchy.
But, this is called into question by such approaches as Dowty 1991; Ackerman and
Moore 1999, 2001; Primus 1999; Beaver 2006 on proto-properties. 13. Probabilistic tendencies “The most that we can say is that a language tends to express similar functions in either the one or the other manner. If a certain verb expresses a certain tense by suffixing, the probability is strong that it expresses its other tenses in analogous fashion and that, indeed, all verbs have suffixed tense elements. Similarly, we normally expect to find the pronominal elements, so far as they are included in the verb at all, either consistently prefixed or suffixed. But these rules are far from absolute.” Sapir 1921:69
33. Taking Chomsky’s assessment seriously
Acknowledging that the syntactic reduction of morphology was predictably problematic, entailing an agglutinative idealization interpretable in terms of highly articulated tree theoretic structures and movements within such structures, we should adopt whatever substantive empirical and theoretical results that have been achieved, but look elsewhere in order to develop a formally precise, empirically responsible, and psychologically supportable theory of morphology.
85
85
9. PARADIGMATIC/CONFIGURATIVE/HORIZONTAL WORD-BASED APPROACHES 1. H2: Paradigmatic/configurative63/horizontal64
Morphological tradition that focuses on words and the ways in which related
surface wordforms cohere into networks of wordforms. Paradigmatic: identifies (sets of) patterns that whole words participate in, i.e., the contrast sets that wholes participate in. Configurative: while the meaning of a wordform is not necessarily construed as a straightforward composition of individually meaningful parts, the meaning of the whole is associated with reliable arrangements of its constitutive elements, i.e. it is the arrangement of coocurring properties of word that are associated with the whole word meaning, rather than the meaning being derived from its constituent parts. Horiziontal: Implicative relations among surface forms (and the schematic patterns abstracted from them). 2. Consequences of word-based morphology:
Wordforms are not insightfully reduced to simple combinations of constitutive pieces but are better viewed as recombinant gestalts or configurations of recurrent partials (segmental or suprasegmental) that get distributed in principled ways among members of paradigms.
From this perspective the focus in morphology is shifted as in a Necker cube:
instead of morphology being (solely) about the composition of complex wordforms from smaller pieces, it is about complex surface wordforms as representing types of configurations of elements and whole surface wordforms as elements in a network of related wordforms.
“words are not merely wholes made up of parts, but are themselves construable as parts with respect to systems of forms in which they participate.” P.H. Matthews 1991:204
Emphasis on surface patterns of different sorts leads to a different set of research issues and questions. it becomes crucial to identify how complex words are organized into meaningful wholes without necessarily attributing meanings to identifiable parts, and how wordforms are organized into structured networks of conjugation and declension classes within inflectional and derivational families.
63 This terms is adopted from Whorf for the same purpose to which he put it. 64 Neither LFG nor HPSG are construed as horizontal though they are not vertical: while they do not derive one level of representation from another, they also do not explore the networks/paradigms that surface wordforms and sentences participate in.
86
86
It becomes natural to ask why the systems of organization cohere in the ways that
they do, how such organization is learned, and whether the nature of the organization reflects learnability constraints, either specific to language or relevant in other learned domains as well.
The paradigmatic/configurational perspective takes surface patterns seriously as
entities that may facilitate learning, so the patterns represented by complex words and the patterns of organization among related words are, therefore, not the epiphenomenal result of representations and operations designed to produce individual words, as standardly assumed in syntagmatic and compositional approaches.
There is no need to assume that no language has classic morphemes: languages
that do are the simplest cases of patterns, so a patterned-based perspective scales to agglutination, but the morpheme-based perspective can only address apparent deviations by assuming underlying agglutinative identity, i.e., it doesn’t scale, and denying in effect the relevance of surfaces being what they are. It’s a mystery why one gets such incredible surface diversity and why such deviations are so long-lived and robust.
If patterns are reliably and systematically associated with meanings, then the issue
not how to reduce all wordforms or clauses to small concatenative bits, but to identify the configurations reliably and systematically can be associated with meanings. One strategy for doing this is certainly a straightforward composition of bi-unique form-function elements, but many other arrangements can suffice to serve this purpose.
“…we cannot but conclude that linguistic form may and should be studied as types of patterning, apart from the associated functions. We are the more justified in this procedure as all languages evince a curious instinct for the development of one or more particular grammatical_processes at the expense of others, tending always to lose sight of any explicit functional value that the process may have had in the first instance, delighting, it would seem, in the sheer play of its means of expression.” Sapir 1921:60 “This feeling for form as such, freely expanding along predetermined lines and greatly inhibited in certain directions by the lack of controlling types of patterning, should be more clearly understood than it seems to be. A general survey of many diverse types of languages is needed to give us the proper perspective on this point.” Sapir 1921:61
Since different configurations/patterns convey different meanings, this diminishes the need for bi-unique relations between forms and functions, rendering their syntagmatic arrangement and composition less essential to the morphological enterprise.
87
87
The primary focus in the paradigmatic/configurational approach on surface words and their systematic alternants rather than on the identification of invariants responsible for deriving them recalls S. R. Anderson’s insightful overview of Badouin de Courtney’s and Kruszewski’s theory of alternations in phonology and morphology. Describing the evolving conception of the phoneme in the works of these two linguists, he argues 1985:68 that:
It is worthwhile to notice, however, that the issue of such an invariant element arises most directly as a consequence of the need to deal with the systematic variance represented by the alternations. It is this systematic variation, with its fundamentally relational character, that language presents us most directly. One way to organize this variation is to hypothesize underlying invariant units – indeed, judging from the history of the discipline, this is the most natural way for linguists to conceptualize such relations – but it should be borne in mind that this is not the only way to do, or even the most transparent… for example, Saussure seems to have held a view of the phenomenon of variance and alternation that was much closer to an immediate account of the relations in question than to an account in terms of another kind of representation for linguistic forms, on given in terms of hypostatized invariants.
Crucially, the surface alternants were interpreted as participating in associative or paradigmatic networks of relations, requiring recognition of (networks of) whole words, hence, this approach is exemplary of the paradigmatic/configurational perspective advanced here.
This perspective on grammar analysis is an essentially paradigmatic one: an object considered as a whole is interpreted in terms of its relations to other similar entities within a system form-meaning oppositions.
Basic strategy reflects in linguistics the various aspects of what Mayr 2004
characterizes as the difference between reductionism and analysis in biology. “Analysis is continued downward only as long as it yields useful new information and does not claim that the “smallest parts” give all of the answers… “ 2004:70
“Analysis differs from reduction by not claiming that the components revealed by analysis, provide complete information on all of the properties of a system, because analysis does not supply a full description of the interactions among the components of the system.” 2004:71
Similarly, Corning 2006 identifies current convergent trends within several
scientific disciplines in which the analysis of entities into parts and wholes and the relations between them play an necessary role:
“What sets the present era apart is the fact that the scientific enterprise seems to be in the process of bridging the theoretical chasm between holism and reductionism; there seems to be a growing appreciation of the inextricable relationships between (and within) wholes and parts, and between various “levels” of organization, relationships which necessitate multi-leveled, multi-disciplinary, “interactional” analyses.” Corning 2005:48
“Synergy, broadly defined, refers to combined or “co-operative” effects-literally, the effects produced by things that “operate together” (parts, elements or individuals). The term is frequently associated with the slogan “the whole is greater than the sum of its parts” (which traces back to Aristotle in The Metaphysics) or “2+2=5,” but, as we shall see, this is actually a caricature, a narrow and perhaps even misleading definition of a multi- faceted concept. We prefer to say that the effects produced by wholes are different from what the parts can produce alone.” Corning 2000:50
88
88
On usual reductive syntagmatic assumptions, all that morphology is concerned with
is the construction of (lexical) wordforms and all that syntax is concerned with is the construction of phrasal combinations: oddly, what is ignored is the system of inflectional and derivational relations that wordforms participate in and likewise the system of syntactic constructions that a single well-formed syntactic structure participates in. Neither wordforms nor clauses are interpreted as parts of a larger whole which itself may possess principles of organization for its parts.
89
89
10. REALIZATION-BASED LEXICALISM: ACKERMAN AND STUMP 2004, ACKERMAN, STUMP, WEBELHUTH (TO APPEAR) 1. A syntactic conception of periphrastic expression (with reference to lexicalist frameworks
Russian verbal lexeme EARN; because its perfective present cell remains unfilled, this is an instance of what Spencer (to appear: 8) calls an UNDEREXHAUSTIVE PARADIGM.
Since imperfective future is expressed periphrastically, it is assumed to be fundamentally different from the perfective future-- synthetic forms of the perfective future are defined by the morphology of Russian, while the periphrastic expressions of the imperfective future are composed in phrase structure.
LFG, like other lexicalist and non-lexicalist generativist frameworks, subscribes to a
morpheme-based conception of morphology.
On the LFG account of Holloway-King (1997:227), for example, the imperfective future consists of two c-structure heads which pool their separate lexical information into a single f-structure:
Powerpoint: Holloway-King categorical and functional structure representations
Basic unification apparatus, used for canonical monotonic relations between words in syntax (e.g. the relation of subject-predicate agreement) as well as in the definition of a synthetic word form’s properties, extended to the composition of morphosyntactic property sets expressed by periphrastic combinations of word forms.
Lexical information associated with the individual words occupying terminal nodes
in the V′ combines to determine the information associated with the f-structure of the V′, hence that associated with the f-structure of the entire sentence.
90
90
Syntactic approach to periphrasis transports the assumptions of lexical-incremental morphology into the domain of syntax: while the morphosyntactic information contained in a predicate’s f-structure representation is, in the case of morphologically synthetic forms, projected from a single word’s lexical representation via c-structure, it is, in instances of periphrasis, an amalgam of morphosyntactic information distributed among two or more syntactically atomic co-heads in c-structure.
Nonexistence of periphrastic perfective futures such as *budet zarabotat’
attributed to “morphological blocking” (recall previous discussion)—periphrastic forms are blocked by synthetic expressions that express the same content.
Blocking is a special mechanism that checks whether the morphosyntactic
information contributed by co-heads is the same as the information that could have been contributed by a synthetic expression.
If so, then periphrastic expression is blocked. (“morphology competes with
syntax”)
Assume a default expectation: all verbal lexemes will be associated all relevant morphosyntactic properties.
Morphology here must be prevented from associating certain morphosyntactic
property sets (in the present case, imperfective future property sets) with verbal lexemes, and syntactic operations provide just these missing associations.
Ad hoc and unexpected restrictions on the morphosyntactic property sets available
for morphological expression, i.e., the sets that cannot be expressed synthetically, are necessarily accompanied by compensatory syntactic idiosyncrasies (such as the need for co-heads and the stipulated blocking of syntax by functionally equivalent morphology).
3. Realization-based lexicalist alternative
Three considerations favor such an approach: a) it allows one to dispense with the otherwise unmotivated device of co-heads; b) it allows one to dispense with the need to stipulate the blocking of syntax by functionally equivalent morphology; c) it allows one to maintain a basic assumption of realizational morphology--that every well-formed morphosyntactic property set is available for morphological realization.
91
91
Requires the following hypothesis:
The Periphrastic Realization Hypothesis:
Inflectional rules that deduce a lexeme’s realizations include rules defining periphrastic combinations as well as rules defining synthetic forms.
“Constructions of these sorts will be referred to as analytic predicates .. . they will be interpreted as entities created by morpholexical rules . .. a portion of the analytic predicate, specifically the aux element, functions as a constituent structure verb, i.e., it is the categorical or structural head of the lexical composition, while the infinitival ...contributes the lexical meaning of the derived grammaticalword' i.e., it serves as the functional head. 'In other words' morpholexical rules produce a grammatical word with discrete structural and functional heads and this analytic composition is associated with a lexical form.” (Ackerman 1987:329) 4. Content-paradigms and their realization
Every lexeme L of category C has an associated CONTENT-PARADIGM: a set of cells each of which consists of the pairing of L with a complete set σ of compatible morphosyntactic properties appropriate to lexemes of category C.
(18) Past-tense cells in the content-paradigm of the Russian perfective verbal lexeme
ZARABOTAT’ ‘earn’ a. 〈 ZARABOTAT’, {singular masculine perfective past}〉 b. 〈 ZARABOTAT’, {singular feminine perfective past}〉 c. 〈 ZARABOTAT’, {singular neuter perfective past}〉 d. 〈 ZARABOTAT’, {plural perfective past}〉
Each cell in a lexeme’s content paradigm has a realization PARADIGM FUNCTION PF: for any cell 〈L,σ〉 in a content paradigm, PF(〈L,σ〉) = X iff X is the realization of 〈L,σ〉 5. Exponence-based approach
Where 〈L,σ〉 is one of the content-cells in (18), the value of PF(〈L,σ〉) might be defined as the result of applying all of the applicable rules in (19a,b) to the root of ZARABOTAT’ (yielding the realizations zarabotal, zarabotala, zarabotalo, and zarabotali for the respective cells in (18)).
(19) a. Realize {past …} through the suffixation of -l b. Realize {singular feminine past …} through the suffixation of -a Realize {singular neuter past …} through the suffixation of -o Realize {plural past …} through the suffixation of -i
92
92
6. Implicative approach
PRINCIPAL PARTS of a lexeme L are a set of cells in L’s content paradigm whose realization suffices to determine the realization of all of the remaining cells in L’s content paradigm
Exponence-based and implicative approaches are not theoretically opposed to one another, since they are suited for distinct purposes: the exponence-based approach is better suited for analyses of morphological exponence, while the implicative approach is better suited for analyses of realizational predictability. (see below on construction-theoretic morphology)
7. Unary expression as one kind of expression among several (11)a. Synthetic Realization Principle ( = Unary Expression)):
Where the realization w of 〈L,σ〉 is a synthetic member of category X, w may be inserted as the head of XP.
Russian verbs as synthetic expressions:
Information lexically associated with zarabotaet ‘s/he will earn’ (↑PRED) = ‘earn < (↑SUBJ)(↑OBJ) >‘ (↑TENSE) = future (↑ASPECT) = perfective
(11)b. Periphrastic Realization Principle:
Where the realization of w1 w2 of 〈L,σ〉 is periphrastic and w1 and w2 belong to the respective categories X and Y, w1 and w2 may be inserted as the heads of the respective phrases XP and YP.
93
93
Periphrastic Realization Principle makes no claims about the surface constituency relations among the elements of periphrastic constructions; thus, our initial assumption is that the structural relationship between XP and YP in (11b) is determined by RULES OF PERIPHRASTIC SYNTAX at least some of which may be language-specific.
The syntactic configuration of morphologically defined periphrases such as [ budet
zarabatyvat’ ] is defined by rules which we here assume to be language-specific; in the case of any periphrase [Y Z] defined for Russian we assume that the verb form
Y is the head of [Y Z] and that the verb form Z appears in c-structure as the head of Y’s complement, as in (34).
Structures such as (34) highlight an important issue: do the parts of a periphrase appearing in some c-structure participate independently (compositionally) in determining the corresponding f-structure i.e., the representation of clausal properties (see 35), or is the f-structure determined by a single lexical representation expressed by two pieces?
Suppose that the f-structure of (34) is (35)
94
94
Is (35) the result of combining budet and zarabatyvat’ (i.e. the f-structures
associated with the content-cells in (36)) or as incorporating the f-structure associated with the content cell in (37) (whose realization is [ budet zarabatyvat’ ]).
Though some compound tenses in the languages of the world can be analyzed as compositional, i.e., in terms of (36) above, many compound tense constructions in are not amenable to a simple compositional analysis, unless the pieces are opportunistically supplied with precisely the information known to be relevant for
the composite construction.
How can one treat compositional and non-compositional constructions in a unified way?
Are there criteria to distinguish between syntactic phrases and morphological
periphrases?
95
95
6. The lexicality of Mari and Udmurt predicates
M. M. Gukhman (1963:199) provide a cogent statement of the basic task:65
The need to establish criteria for the differentiation of analytic verbal constructions from other types of word combinations, such as the combination of two or several full words or the combination of an auxiliary with a full word, is connected with the question of whether these constructions are considered members of a paradigmatic series, that is, whether they are units of the morphological level.
Three sufficient criteria for the status of lexicality 7. Noncompositional periphrases Criterion I: If the morphosyntactic property set associated with an analytic combination C is not the composition of the property sets associated with its parts, then C is a periphrase.
M. M. Gukhman (1955:343) concludes with respect to German periphrastic verbal constructions:
“…the grammatical meaning of analytic constructions in German is never equal to the sum of the grammatical meanings of its component parts, but appears as the meaning of an nondecomposable whole.”
Udmurt imperfective past tense:
Compound tense used to describe “a protracted or repeated activity occurring in the ... distant past” (Csúcs 1990:51).
Realized by the periphrastic combination of a future-tense form (inflected for
subject agreement) with the invariant past form val of the copula, as in Table 3; compare the future-tense forms in Table 4.
65 See also Zhirmunskij (1963:24).
96
96
Neither part of an imperfective past-tense periphrase such as [mïno val] carries any
exponent of an aspectual property such as durativity or habituality; yet, such a property is associated with the verb phrase [VP mïno val ] as a whole.
The finite head of [mïno val] is marked for future tense, while the periphrase as a whole expresses the distant past tense.
Departure from pure compositionality is determined by the morphology of
Udmurt: the temporal and aspectual properties of the verb phrase [VP mïno val ] aren’t deducible from the individual content-cells in (38), but from the periphrastically realized content-cell in (39).
According to this constraint, the f-description of the Udmurt sentence Ton mïnod val ‘you (sg.) used to go’ must be determined by the content-cell in (39) rather by those in (38).
Eastern Mari second past tense: Negative predicates are a periprhastic combination of KOL’s affirmative gerundial
stem kolen with a negative present-tense form of the copula UL (cf. Table 11); the latter form itself a compound of a present-tense form of the negative auxiliary OK (cf. Table 12) with the stem əl of the copula UL.
97
97
None of the parts of a negative second-past verbal periphrase expresses the second-past tense; the exponents of tense carried by the finite head of such a periphrase are expressions of the present tense.
Consequently, though the verb phrase [VP kolen oməl ] ‘I didn’t die’ is associated
with the property set {POL:neg, TNS:2nd past, AGR:{PER:1, NUM:sg}}, this association cannot be seen as an effect of ordinary property unification.
The analytic combinations in Table 10 are therefore periphrases by criterion (30).
98
98
13. Sanskrit periphrastic future: 1st & 2nd person for all 3 numbers are periphrastic and based on the “agentive form of the verb plus the appropriate present indicative form of “be”; 3rd person for all numbers is syncretic with the nominative case form of the “agentive” nominal.
Note two configurative or construction-theoretic properties of these forms: P1: The combination of the “agentive” form and present indicative is associated with the future tense, despite neither form on their own being independently attested as bearing this morphosyntactic property. This is a periphrastic reuse of grammatical elements in which the configuration is systematically associated with a particular morphosyntactic property. P2: In 3rd person the “agentive” form on its own is associated with future tense; it gets this value by occupying a particular place in the paradigm of future tense expressions. This is a reuse of elements where their grammatical value arises from construing them as parts within the complete paradigm for future tense.
99
99
8. The paradigmatic opposition of periphrasis to synthesis
Periphrases commonly stand in paradigmatic opposition to synthetic realizations; that is, they realize contrasting values for the same morphosyntactic features but are otherwise identical in their lexicosemantic content.
Western Mari: Present desiderative, first-past, and second-past66 realizations of the verb KOL ‘die’
in Western Mari are given in Tables 5-7.67 The second-past realizations in Table 7 are uniformly synthetic, for both the
affirmative-polarity and the negative-polarity portions of the paradigm.
The desiderative and first-past realizations in Tables 5 and 6, by contrast, are synthetic in the affirmative but periphrastic in the negative; in particular, each of the negative realizations in Tables 5 and 6 involves a finite form of the negative verb AK (the relevant forms of which are isolated in Table 8) and an essentially invariant form of KOL itself.
66 According to Kangasmaa-Minn (1998:229), “[t]he first past refers especially to states and events which the speaker has personally witnessed, while the second past is more or less a record of what has been or happened without any emphasis on the speaker’s attitude towards the truth value of the utterance”. 67 The segmentation of Mari formatives assumed here and throughout follows the analysis of Eastern Mari verbs proposed by Sebeok & Ingemann (1961).
100
100
Note that the Tense and Person/Number suffixes in the affirmative are essentially the same as the negative verb in the negative.
101
101
Distribution of periphrasis in Tables 5-7 is FEATURALLY INTERSECTIVE; there is no one morphosyntactic property among those expressed by the realizations in Tables 5-7 that is always expressed periphrastically rather than synthetically.
Not all negative realizations are periphrastic, nor are all desiderative or first-past
realizations synthetic; instead, it is the intersection of negative polarity with the desiderative mood or with the first-past tense which is expressed periphrastically in Western Mari.
(12) Criterion II: If an analytic combination C has a featurally intersective distribution,
then C is a periphrase.
In syntactic approach to periphrasis in which all synthetic realizations are defined by morphological realization rules and all analytic combinations are defined by ordinary principles of syntax, the absence of one-word realizations in the negative desiderative or negative first past (whether these be synthetically inflected realizations or, through poverty of exponence, simple uninflected forms) would have to be attributed to the following ad hoc stipulation:
(13) Contrary to expectation, property sets specified for negative polarity and either
desiderative mood or first-past tense are not available to Mari verbal lexemes.
In a realization-based lexical approach to periphrasis morphological rules of synthesis and periphrasis participate competitively--as alternatives--in the realizational definition of a lexeme’s forms.
Competition not between morphology and syntax, as in morphological blocking
proposals, but between the varieties of exponence employed in realizing the cells in a lexeme’s content-paradigm.
The fact that some morphosyntactic property sets lack single-word realizations is
therefore attributed to (a) the lack of any rules of synthetic exponence realizing those property sets and (b) the existence of a general default rule of periphrastic exponence realizing those sets.
102
102
No ad hoc stipulation ( = 13) is needed: property sets specified for negative
polarity and either desiderative mood or first past tense are, as expected, available for realization in the inflection of Western Mari verbal lexemes, but are realized by a default rule of periphrasis.68
Thus, the inferential-realizational approach is--unlike the purely syntactic
approach--fully compatible with the restrictive hypothesis that any property set that is legal in c-structure can also legally drive morphological realization.
Synthesis excluding periphrasis in the negative second past follows from the
assumption that the default rule of periphrasis defining negative realizations is (in accordance with Pànini’s principle, i.e., the more specific rule/pattern applies before the less specific) overridden by a narrower rule of synthesis defining negative second-past realizations; thus, the inferential-realizational approach also avoids appealing to an ad hoc principle of morphological blocking to account for the exclusion of periphrasis in such instances.
There is no need to posit a special blocking principle to regulate the relation
between morphology and syntax: as a general, independently motivated constraint within morphology, Pànini’s principle suffices to account for the relevant data if periphrasis is defined realizationally (Stump to appear).
9. Distributed exponence and configurative effects in periphrases
DISTRIBUTED EXPONENCE: Tendency for each of the morphosyntactic properties
realized by an inflected word form to have no more than a single exponent in that form’s morphology.
Fullest expression in agglutinating languages
Swahili verb form ha-tu-ta-taka ‘we will not want’ has one affixal exponent for
each of the morphosyntactic properties it expresses: ha- expresses the property POL:neg; tu-, the property AGR:{PER:1, NUM:pl}; and ta-, the property TNS:future.
Criterion III: If the morphosyntactic property set associated with an analytic combination
C has its exponents distributed among C’s parts, then C is a periphrase.
In a syntactic approach to periphrasis, there is no particular reason to expect that periphrases should exhibit a comparable tendency toward distributed exponence; word combinations in syntax aresometimes highly redundant in their expression of shared morphosyntactic properties (as in the Swahili sentence ki-kapu ki-kubwa ki-moja ki-lianguka ‘one large basket fell’, every one of whose words carries an exponent of the subject’s gender and number).
68 The default appeal to periphrastic realization parallels the status of periphrastic negative expressions as the unmarked encoding within Uralic.
103
103
If the economy of inflectional exponence exhibited by heavily agglutinating languages is seen as a property of morphological rules, then the assumption that periphrases are morphological entails that periphrases should be no less likely to exhibit this same economy.
Udmurt: 1st conjugation verb mïnï - ‘go’: (the data follow Csucs 1998:290, but see Csucs 1990:51 and Serebrennikov 1963 for alternative transcriptions)
Distinction between 2nd person singular and plural is indicated by identical person marking on the negative verb and - ï for singular and -e for plural on the connegative form of the verb.
The form ug is not determinate for person (except possibly that it is -2nd and -1st PL);
while a form such as mïnïškï is determinate for singular number, it is not determinate for person (except possibly for -3rd).
The combination of forms ug mïnïškï cannot uniquely determine the feature set
NEGATIVE 1ST SING PRESENT simply as a function of composing independently motivated information associated with it pieces.
Rather, ug mïnïškï seems a (motivatable) realization of this feature set within the
context of paradigmatic morphosyntactic information sets and their exponents as found in the NEGATIVE PRESENT TENSE paradigm. – it is a systemic reuse of pieces.
In the negative future tense paradigm, there is distinctive marking for the 2nd and 3rd persons and distinctive markings for 1st singular and 1st plural.
104
104
In contrast to the number marking found in the negative present tense, number marking in the negative future tense is indicated by the same singular connegative forms for all persons and the same plural connegative forms for all persons.
The regular distinctions reflect the regular distinctions for person and number found
in the synthetically expressed affirmative paradigms, in conjunction with the expression of number also typical of SUBJ agreement in this language.
TABLE 16. Negative imperfective past-tense realizations of Udmurt MÏNÏ ‘go’
[Suihkonen 1995:302] SG. 1 ug mïnï val 2 ud mïnï val
3 uz mïnï val ‘s/he didn’t used to go (long time ago)’
PL. 1 um mïne(le) val 2 ud mïne(le) val
3 uz mïne(le) val
Forms for the negative future are used for the negative imperfective past-tense, just as affirmative future are used in the affirmative imperfective past-tense (see Table 3 above.)
10. Prediction 1: Periphrasis and language change
Since periphrasis is just one type of morphological exponence, the development of synthetic morphology from periphrasis is therefore not different, in principle, from the development of fusional morphology from agglutination: both sorts of developments involve an increasing degree of fusion in the inflectional realization of a paradigm’s cells.
Predicts that just as one can observe different degrees of progress in the development from agglutination to fusionality, one should likewise find different degrees of progress in the development of synthesis from periphrasis; Mari provides compelling evidence of this sort of gradation.
The synthetic negative second-past realizations in Western Mari
105
105
Descend historically from periphrases--in particular, from periphrastic
combinations of the negative gerund with affirmative present-tense forms of the copula UL; compare the free forms of this copula in Table 17.
The affirmative second-past realizations of TOL ‘go’ from Northwest dialect in
Table 18.
Plural realizations consist of an uninflected affirmative gerund with an independent copula inflected for present tense and subject agreement.
Copular construction has become synthetic in the first- and second-person singular
realizations
106
106
Like the negative second-past realizations in Western Mari, the singular affirmative
realizations involve a gerundial stem whose absolute form appears in the third-person singular and whose conjunct form (suffixed with l) otherwise appears with the appropriate person/number marker.
Development from periphrasis to synthesis has progressed partway across the
affirmative second-past paradigm.
Gradually, the copulative verb used in the periphrastic expression of the affirmative second past has become enclitic, then reanalyzed as synthetic morphology.
Similar change in progress documented in Eastern Mari, where the plural
realizations of the copula UL in the negative present tense appear sometimes as single-word forms (as in Table 11), but sometimes as periphrases (as in Table 19); presumably the periphrastic realizations are losing ground among innovative speakers.
A theory of morphology should explain the fact of gradience within synthesis as well between synthesis and periphrasis and how this reflects diachronic development.
11. Prediction 2: Periphrasis and morphological markedness
Types of exponence correlate with the degree of markedness of the morphosyntactic property sets which those forms realize.
Exponents of more highly marked morphosyntactic properties tend to be less
fusional (cf. Greenberg 1966, Mayerthaler 1988):
107
107
in Sanskrit, the nominative singular is often expressed purely by stem gradation (e.g. PITAR ‘father’, nom. sg. pità, RĀJAN ‘king’, nom. sg. ràjà), while the dative singular is always expressed suffixally (pitr-e ‘to the father’, ràj¤-e ‘among the kings’); in Swahili, negation and first-person singular subject agreement are expressed by a portmanteau prefix si- (e.g. si-ta-taka ‘I will not want’) while negation and first-person plural subject agreement are expressed separately by the respective suffixes ha- and tu- (ha-tu-ta-taka ‘we will not want’). On the hypothesis that periphrasis is a kind of morphological exponence,
periphrastic exponence would be expected to participate in this same correlation. Since periphrasis is by definition nonfusional, one would expect that in paradigms
in which synthesis and periphrasis exist side by side, the incidence of periphrasis will be associated with more highly marked morphosyntactic properties.
In the Northwest dialect realizations in Table 18, it is those whose property sets
include the marked number specification ‘plural’ which are periphrastic; similarly, periphrasis is the default expression of the marked polarity specification ‘negative’ in Western Mari.
Tundra Nenets:
TABLE 20. Declension of ti ‘reindeer’ in Tundra Nenets
[Salminen 1997] Singular Dual Plural
Nominative ti texºh tiq Accusative tim texºh tí
Grammatical cases
Genitive tih texºh tíq Dative tenºh texºh nyah texºq Locative texºna texºh nyana texºqna Ablative texødº texºh nyadº texøtº
Local cases
Prosecutive tewºna texºh nyamna
teqmºna
The marked number, DUAL, and the marked cases, LOCAL, are periphrastically expressed.
A theory whose notion of exponence encompasses synthetic but not periphrastic markings affords no coherent articulation of the overarching generalizaion which such cases embody; in the grammatical ontology of such a theory, the phenomena which ought to be subsumed under this generalization - phenomena such as fusion, agglutination, periphrasis - fail to constitute any kind of natural class'
108
108
11. CONSTRUCTION-THEORETIC MORPHOLOGY: WORD-BASED IMPLICATIVE MORPHOLOGY 1. Noun paradigms in Uralic
Noun declensions in Uralic exhibit intricate patterns of interpredictability. These patterns cannot in general be reduced to interderivability.
The class of a noun can be identified by between one and three principal parts.
Certain forms, or sets of forms, imply the shape of other forms in a paradigm.
Patterns of syncretism refer to paradigm cells not to properties of stem forms.
2. Saami
Saami declensions are based on two principal parts: the genitive and nominative singular.
Nouns can be assigned to three classes, based on syllable count and stem
alternations.
In the first declension, the genitive singular has an even number of syllables.
The nominative and genitive singular exhibit a strong-weak or weak-strong grade contrast.
109
109
3. First declension nouns exhibit three patterns of cell-based stem syncretism
(i) The nominative singular predicts the illative singular and essive. A strong nominative singular, such as bihttá, implies a strong illative singular and essive. Conversely, a weak nominative singular, such as bargu, implies a weak illative singular and essive.
(ii) The genitive singular predicts the locative and comitative singular and all plurals.
(iii) The comitative singular is identical to the locative plural (and in possessive paradigms).
4. Irrelevance of independent stem properties
First declension nouns contain distinguishable weak and strong (geminated) stems.
Yet no patterns of stem syncretism refer to stem grade directly.
No first declension forms are consistently based on the strong or on the weak stem: forms are based on the genitive or nominative singular and share its grade.
5. Prediction and referral
Stem syncretisms relate the forms that are realized by distinct paradigm cells, and can be described by the exponence-based rules of referral in Zwicky 1985 or Stump 2001.
The same patterns can be described in terms of implicational relations over a
lexicon of full word forms, as in the traditional models of Paul 1880 and Kuryłowicz 1949: The nominative singular predicts the essive form: <[Nom Sg], X> → <[Ess], Xn> Referral relations then involve interpredictability: <[Com Sg], X> ↔ <[Loc Pl], X>
6. Paradigm cell filling problem: First pass
What licenses reliable inferences about the surface wordforms for the inflectional (and derivational) families of wordforms associated with (classes of) lexemes?
“Don’t you see that neither you nor anybody else has ever heard all of the nouns of the paradigm fa’il or maf’ul? You have heard some forms and then you have proceeded by analogy to produce others.” Langhade 1985:111. Cited in Itkonen 2005:89
110
110
a. How are such complex systems organized? b. What role might this organization play with respect to licensing inferences
concerning paradigm cell filling? c. What relation does this organization and the possibility for inferences based on
surface patterns have concerning the learnability of complex systems? 7. A (relatively) simple example It will be useful to here to consider a (relatively) simple example. Finnish (Uralic) nouns are marked for case (NOM, GEN, PART, …) and number (SG, PL). A representative sample of Finnish declension classes appears in Table 7:
Class NOM.SG GEN.SG PART.SG PART.PL INESS.PL
4 lasi lasin lasia laseja laseissa ‘glass’
9 nalle nallen nallea nalleja nalleissa ‘teddy’
8 ovi oven ovea ovia ovissa ‘door’
32 kuusi kuusen kuusta kuusia kuusissa ‘six’
10 kuusi kuuden kuutta kuusia kuusissa ‘spruce’
Shaded cells are diagnostic: they uniquely resolve class assignment for a lexeme.
a. Predictive wordforms identify correct class assignment
Stimulus: tuohta birchbark.part.sg:
kuusta : tuohta :: kuusi : TUOHI Correct assignment to class 32 b. Neutralized wordform: underdetermines correct class assignment
Stimulus: nuken puppet.gen.sg nallen : nuken :: nalle : NUKKE oven : nuken :: ovi : NUKKI Correct class assignment to 9 indeterminate with respect to 9 v. 8
111
111
c. Stimulus: nukkeja puppet.part.pl
nalleja : nukkeja :: nalle : NUKKE laseja : nukkeja :: lasi : NUKKI Correct class assignment to 9 indeterminate with respect to 9 v. 4 Stimuli: nuken puppet.gen.sg & nukkeja puppet.part.pl Conjunction of wordforms is predictive – correct class assignment to 9
8. Basic strategy for pattern assignment Paradigmatic patterns: The wordform in a specific cell or wordforms in several separate cells (i.e., patterns of cells) are diagnostic of declension class membership 9. Taxonomy of strategies (G. T. Stump and R. Finkal 2006, to appear)
Stump and Finkel’s notion of dynamic principal parts, contrasting with static and adaptive analyses.
d. In fact, there are many equally good alternative sets of principal parts for Finnish, and many more solutions that are almost as good.
e. We speculate that this is a common feature of complex morphological systems (cf. resilience in biological systems)
10. General hypothesis of (sub)paradigm organization
Identifiable patterns of relatedness between wordforms in paradigms facilitate paradigm cell filling. Related wordforms are partitioned into (sub)paradigms with their own small systems of relatedness among forms.
What Finnish (sub)paradigms share are recurring formal elements , e.g., lasi occurs in nom. sg., gen. sg. & part. sg. while lase occurs in part. pl. and iness. pl
11. Paradigm Cell Filling problem: General formulation
Given a lexeme L associated with a set of morphosyntactic properties (=morphosyntactic or grammatical word) and expressed by a surface wordform (exponent), what are the surface wordforms for all other possible morphosyntactic property sets of L, i.e. what is the complete paradigm of surface wordforms for L?
112
112
So, paradigm cell filling concerns the licensing of reliable inferences about the
surface wordforms for the inflectional (and derivational) families of wordforms associated with (classes of) lexemes, i.e., given a novel inflected word form, what are all the other wordforms in its inflectional (and derivational) families?
12. The importance of surface words Q: What forms the basis for prediction? A: Surface words and patterns of relatedness among surface words 13. Patterns in the word system
Patterns of relatedness between wordforms partition morphosyntactic feature combinations into (sub)paradigms which cohere with respect to the recurrence of “formatives” constitutive of wordforms, i.e., configurations of “recurrent partials” such as segments, tones, etc. For any given language: What are the (sub)patterns of (inter)predictability and what
are the elements relevant to (inter)predictability? 14. Present task: (partial) Tundra Nenets (Samoyed) nominal inflection
Given any Tundra Nenets inflected noun wordform, what are the remaining 209 forms of this lexeme for the morphosynactic feature property combinations CASE { nom, acc, gen, dat, loc, abl, pro}, NUMBER {singular, dual, plural}, POSSESSOR {3 pers. x 3 num.}? (7 x 3) + (7 x 3 x 3 x 3) = 210
Identifying patterns of (inter)predictability for a subset of Tundra Nenets nominal
declensions within and across (sub)paradigms
given exposure to a stimulus such as that in (1a), the nominal nganuqmana ‘boat (plural prosecutive)’, what leads to the inference that its nominative singular form is the target ngano?
If confronted with the plural prosecutive of the nominal wíngoqmana ‘tundra
(plural prosecutive)’ in (1b), what leads to the inference that its nominative singular is the target wíh?
(1) a. Stimulus: Target vs. b) Stimulus Target nganuqmana ngano wíngoqmana wíh boat.PL.PROS boat.SG.NOM tundra.PL.PROS tundra.SG.NOM
113
113
15. Patternment within and across stem types – Absolute declension (=Non-possessive)
Lexical categories are divisible into the following gross stem type classification (ignoring the relevance of syllabicity, see Salminen (1997, 1998) for careful of exposition of Types and see VI below for use of these classes:
Type 1 (T1): ending in C (except glottal) or V: Type 2: (i): stem ends in a nsal/voicing glottal stop ( = h) (ii): stems ends in non-nasalizing/devocing glottal stop69 ( = q) Type 1: polysyllabic vowel stem: ngano ‘boat’
Singular Plural Dual
Nominative ngano nganoq nganoxoh Accusative nganomh nganu nganoxoh Genitive nganoh nganuq nganoxoh
Dative-Directional nganonh nganoxoq nganoxoh nyah Locative-Instrumental nganoxona nganoxoqna nganoxoh nyana
Ablative nganoxod nganoxot nganoxoh nyad Prolative nganowna nganuqmana nganoxoh nyamna
Type 2i: non-nasalizing glottal stem: wíh/wíng ‘tundra’
Singular Plural Dual
Nominative wíh wíq wíngh
Accusative wímh wíngo wíngh Genitive wí0h wíngoq wíngh Dative-Directional wíndh wíngq wíngh nyah Locative-Instrumental wíngana wíngaqna wíngh nyana
Ablative wíngad wíngat wíngh nyad Prolative wímna wíngoqmana wíngh nyamna
69 See Salminen 1997, 1998 for a careful taxonomy of stem types in Tundra Nenets which forms the basis for the analysis below. The orthographic conventions in the table represent an admixture of Latinized traditional Cyrillic orthography and Salminen’s phonological representation. This is intended to make the representations transparent without going into more phonological detail than the use of Salminen’s phonological representations would require. There are inevitably, as a consequence, certain aspects of the representations which are misleading. In contrast to the utilitarian motivations guiding the representations in these table, all of the statistical calculations are based on Salminen’s phonological transcriptions of words.
114
114
Type 2(ii):
Singular Plural Dual
Nominative myaq myadq myakh
Accusative myadmh myado myakh Genitive myadh myadoq myakh Dative-Directional myat myakh myakh nyah Locative-Instrumental myakana myakaqna myakh nyana
Ablative myakad myakat myakh nyad Prolative myaqmna myadoqmana myakh nyamna
16. Basic observation about Tundra Nenets
Nominal paradigms for all stem classes are partitioned into subparadigms each of which is defined by the presence of a characteristic and recurring stem, e.g., ngano, nganu, nganoxo ( = sectors in Stump and Finkal to appear)
17. Hypothesis about the organization of Tundra Nenets paradigms
Subparadigms are domains of interpredictability among wordforms, rather than of derivability from a privileged base
Following Trosterud (2004:54), we will assume that:
Morphology is essentially about relations between whole words (paradigmatics), not about pieces that make up single words (syntagmatics): “wordforms are signs, parts-of-wordforms are not.
18. Surface-based and Symmetric
No form need necessarily serve as a privileged base form among different surface expression of single lexeme
Regardless of whether a stem exists as an independent word, all these systems share the property that they have clusters of related forms where it is at least somewhat arbitrary to take any one form as basic. This is what I take to be defining characteristic of a paradigm. Thus, we need a way to relate to the various members of paradigm directly to each other without singling out any one of them as a base for the others.” (Bochner 1993:122)
115
115
19. Surface-based and asymmetric
There is a single most informative surface form within any paradigm from which all other forms are derived.
This single base form is the same for all conjugation or declension classes.
If the partitive singular is identified as the base for declension class 1, then the
partitive singular form must likewise serve as the base for all other declension classes
“learners are restricted to selecting just one form as the base within the paradigm, and all other forms must be derived from the same base.”70 (Albright 2002:118)
70 This resembles one of the analytic options for determining (underlying) base forms argued against in Kenstowicz and Kissebirth 1979:201: “The alternant selected as the UR must occur in the same morphological category for all morphemes of a given morphological class (verb, noun, particle, etc.).”
116
116
The postulation of a base entails an asymmetric derivational relation between it and derived forms
20. Symmetric or asymmetric ?
There is no single answer: Language particular evidence determines 21. Tundra Nenets: A Pilot study
Hyp: Tundra Nenets nominal paradigms are organized into several subparadigms, and the domains in which recurrent forms bases play role constitute domains of interpredictability
Tundra Nenets nominal paradigms are organized into several subparadigms, and
the domains in which recurrent forms bases play role constitute domains of interpredictability
22. Comparing (a)symmetric approaches
Explore whether the most challenging and problematic instance of relatedness between two wordforms is reliably asymmetric and based on the same morphosyntactic cell across declension classes, and so to test Albright’s Single Surface Base Hypothesis.
If the complexity of analyzing the two least transparently related words within a paradigm can be simply stated by referring to a single constant base form across conjugation classes, then this will stand as an argument for the Single Base Hypothesis.
117
117
Conversely, under the symmetric analysis, we expect that no single form can be compellingly demonstrated to be the base, but that a principled account can be stated in terms of patterns of interpredictability within alliances of forms.
23. Tundra Nenets Nom. Sg. and Acc. Pl. Nom sg. Acc. pl. Gloss ngøno ngønu ‘boat’ lyabtu lyabtu ‘harnessed deer’ ngum nguwo ‘grass’ xa xawo ‘ear’ nyum nyubye ‘name’ yí yíbye ‘wit’ myir myirye ‘ware’ wíh wíngo ‘tundra’ weh weno ‘dog’ nguda ngudyi ‘hand’ xoba xobo ‘fur’ sawonye sawonyi ‘magpie’ tyírtya tyírtya ‘bird’
Table 1: Tundra Nenets inflected nominals
Indeterminacy or uncertainty with respect to predictability in both directions, i.e., ACC.PL of both boat and harnessed deer end in the vowel –u, their NOM.SG forms end in –o and –u respectively
-yø
-yo
-yu
-yi
-ye
-ya
-yø
-yi
-yu
-ye
-e
-yo
Nom.Sg. Acc.Pl.
.
.
.
.
.
.
Figure 1: NOM. SG. and ACC. PL. pairings in Tundra Nenets
118
118
Given exposure to one form, how can one predict the other? This is the Paradigm
Cell Filling problem.
From the perspective of a single base hypothesis, the question is whether one is derivable from the other (or whether some 3rd form can serve as a base for both).
24. Data
Corpus of 4,334 nominals extracted from Salminen’s dictionary of 16,403 entries, based on Tereshchenko (1965).
The dictionary specifies meaning, frequency, as well as the stem class assignment
for lexemes. 25. Measure of predictability: Entropy
Quantify "prediction" as a change in uncertainty, or 'information entropy' (Shannon 1948).
Intuitively, the more choices there are, and the more evenly distributed their probabilities, the greater the uncertainty (entropy) that a particular choice will be made.
Conversely, random variables with only a few possible outcomes, or with one or two highly probable outcomes and lots of rare exceptions, have a low entropy.
Given a random variable X which can take on one of a set of discrete choices and their probabilities P(X), we can calculate the entropy H(X) of the variable.
The more choices there are in a given domain and the more evenly distributed the
probability of each particular occurrence, the greater the uncertainty or surprise there is (on average) that a particular choice will be made among competitors and, hence, the greater the entropy.
Conversely, choices with only a few possible outcomes or with one or two highly
probable outcomes and lots of rare exceptions have a low entropy.
For example, the entropy of a coin flip as resulting in either heads or tails is 1 bit; there is equal probability for an outcome of either heads and tails.
26. Calculating predictability
Measure predictability using conditional entropy H(Y|X), i.e., the uncertainty in the value of Y given that we already know the value of X.
The smaller H(Y|X) is, the more predictable Y is on the basis of X, i.e., the less
surprised one is that Y is selected.
119
119
27. Different hypotheses, different predictions
Given our competing hypotheses, we would expect the following. Under the asymmetry hypothesis, we should see a greater degree of predictability (and a lower conditional entropy) in one direction than in the other, meaning that either NOM. SG. or ACC.PL should be a good predictor of the other.71
In contrast, the absence of a global directionality of predictability is compatible
with the symmetric approach, though additional evidence would be required to address its claims of alliances of interpredictability.
28. Nom. sg. Acc. pl ( = H(acc.pl.|nom.sg.) vs. Acc. pl. Nom. sg. ( = H(nom.sg.|acc.pl.)
Consider H(acc.pl.|nom.sg.), i.e., the uncertainty in the acc. pl. given the nom. sg.
In some cases, knowing the nom. sg. of a word uniquely identifies its acc. pl., e.g. the nom. sg. ending in –ye is always –ye in the acc. pl.
With this type of nom.sg. word, there is no uncertainty in the acc.pl. and the conditional entropy H(acc.pl.|-ye)=0.
In other cases, however, knowing the nom.sg. narrows down the choices for the acc.pl. but does not uniquely identify it.
For example, words whose nom. sg. ends in -ø might have an accusative plural in -ø, -o, -yø, or -yo.
71 Of course, if it turns out that only some forms reliably predict other form unidirectionally, i.e., there is low conditional probability going in one direction for select number of instances, but not for others, one might argue that the offending pairs are lexically listed exceptions and, consequently, uninteresting. We believe that this strategy runs the risk of ignoring small regularities by overwhelming them with possibly less representative more transparent pairs of relations.
120
120
On average, across the whole (sample) lexicon, the uncertainty in the acc.pl. given
the nom.sg. is 0.77 bits. In other words, the nom.sg. "predicts" all but 0.77 of the 3.21 bits known to be
needed to encode the acc.pl. Going in the other direction, from acc. pl. to nom. sg., the conditional entropy
H(nom.sg.|acc.pl.)= 0.80. In other words, the acc. pl. “predicts” all but 0.80 of the 3.24 bits in the nom. sg.
29. Interpretation of results
Nom.sg. is slightly more helpful for predicting the acc.pl. than vice versa, i.e., the value is closer to 0 in the former than the latter.
But, the real conclusion is that neither is especially useful for predicting the other, since there’s still plenty of surprise in the outcome.
30. Type versus token frequencies
The above calculations are based on "type" frequencies: each distinct noun is counted once.
We can also calculate conditional entropy on the basis of "token" frequencies: all occurrences of nouns in the corpus, i.e. the same type can be represented by several tokens.
Using token counts from Salminen’s corpus, we get:
121
121
H(nom.sg.) = 3.61 bits H(acc.pl.) = 3.75 bits H(acc.pl.|nom.sg.)=0.99 bits H(nom.sg.|acc.pl.)=0.85 bits
The results here are reversed, i.e., the acc. pl. is a slightly better predictor of the
nom. sg. than the other way around. But the previous point remains true: neither form predicts the other very well.
31. Relatively low likelihood of encountering the acc. pl. Frequency distributions of absolute declension for all case and number encodings of the 13,497 noun tokens in Salminen’s sample sentence corpus :72 sg du pl nom 4656 8 878 gen 3348 7 422 acc 1223 5 402 dat 835 0 102 loc 805 0 124 abl 327 0 60 pros 427 0 45 where, nom. sg. represents 34.4% and acc. pl. 3.0%.
If the system were organized around the need to encounter the acc. pl., paradigm completion would be hampered, given the low probability of encountering this form.
In fact, this is likely true for all wordforms, except the nom. sg. and gen.sg. 32. Higher likelihood of encountering the partials associated with acc. pl. if we posit subparadigms
Frequency distributions for all absolute and possessive forms for all case, number, and possessor encodings
Partial 1 14,519 Partial 2 1,889 Partial 3 2,145
Cues for wordshapes can be gotten even for lower frequency forms if we assume alliances of related forms within subparadigms, i.e., any form within a subparadigm predicts the others and may provide clues for forms in other subparadigms.
72 This corpus contains 9,993 sentences consisting of 39,417 words.
122
122
Note that even if a derivability relation had been identifiable, this would not have accounted for the evident subpatterns of shared forms as described above, rendering such subpatterns epiphenomenal, rather than central to organization. (we have no explanation for why the alliances consist of the forms and feature sets they do.)
33. Summary of results
Both type and token calculations suggest that, for the comparison of nom. sg. and acc. pl., neither reliably serves as the single base from which the other is predicted.
These equivocal results with respect to directionality of prediction, contrast with
the overwhelming likelihood of encountering nom. sg. versus acc. pl. on the basis of frequency distributions.
Positing subparadigms reveals that partials appear with much higher frequency than
any given wordform, so that there is no need to encounter a specific form in order to predict allied forms. What’s important that the aggregate frequency of partials be high enough to be useful.
34. Bochner's symmetrical pattern sets and Albright's asymmetric local bases are both used to paradigm structure.
The two models make very different predictions when considered in the light of the paradigm completion problem:
In Albright's model, derived forms should be predictable from bases, but there is no
reason to expect bases to be predictable from derived forms or derived forms to be predictable from each other.
Bochner's model, on the other hand, allows for potentially complex interrelations
between forms in the same paradigm or subparadigm. (Sub)paradigms are organized in terms of patterns of whole word relatedness with
members of (sub)paradigms exhibiting interpredictability: this facilitates solving the paradigm completion problem, i.e., to reliably predict an inflected form of a word given any other inflected form in languages like Tundra Nenets.
35. Ramifications: a developmental issue concerning learnability
How do children go about identifying the relevant dimensions of morphosyntactic properties and how do they isolate the appropriate patterns of surface exponence?
123
123
The whole word hypothesis is consistent with a learning strategy that begins with
what children are exposed to concretely and then discovers/develops relatedness schemata of increasing abstractness which license inferences about novel wordforms (Tomasello 2003; Gentner: Pinker 1984; MacWhinney 1978, among others.)
So, hypotheses concerning the paradigm completion problem have consequences
for how we formulate acquisition questions, among other sorts of external evidence bearing on linguistic analysis.
124
124
14. CONSTRUCTION THEORETIC SYNTAX – COOPERATING
CONSTRUCTIONS AND EURASIAN RELATIVE CLAUSES
1. Big issues affected by little language specificities
Grammatical systems are clearly complex systems consisting of patterned organizationnumerous dimensions of analysis, i.e., phonology, morphology, semantics, syntax
Recent proposals within Evolutionary Phonology (Blevins 2005, 2006, 2007,
Wedel to appear) represent an explicit exploration within phonology of how differences in methodology, argumentation, and analytical assumptions extend to linguistic phenomena, and how the role of language development over historical time can be accommodated within a biologically oriented approach to natural language phenomena.
In the context of the present research this view concerning sound patterns can be
interpreted as follows in connection with morphosyntax:
Recurrent synchronic grammatical constructions (whether widely attested or somewhat restricted in distribution) have their origins in both diachronic development and systemically motivated constructional analogy.
In fact, Blevins 2003 herself develops her phonological proposal more broadly in
order to apply across various domains of grammars. She suggests that:\ “The evolutionary approach to phonology, morphology, syntax is not a theory of what synchronic grammars must encode, but rather what they need not encode as properties of universal grammar. Any cross-linguistic tendency which has a straightforward historical explanation should be excised from statements of universal grammar, unless it can be independently motivated. Within synchronic systems, an adequate grammatical description needs to distinguish productive from non-productive processes, and be able to express the categorical knowledge available to speakers. However, apart from these basic distinctions, the form of synchronic grammars remains highly underdetermined.” Blevins 2003:344
Adopting an evolutionary orientation, we propose that the pronominal relative clause type found in numerous Eurasian languages:
(a) is best (perhaps, only) understood within the system of grammatical phenomena
(both morphological and clause level constructions) in which it occurs and (b) has the probabilistic effect of specific configurations of grammatical properties for
potentiating the development of these relative clause constructions
125
125
2. The Empirical Phenomenon: Little language specificities
Many languages contain externally headed prenominal relatives containing non-subject gaps ( VMC = verbal mixed category, e.g., participle)
And, many languages express pronominal subjects of relatives via person/number
markers ( = PNM) and optionally co-occurring independent pronouns:
The clause headed by the verbal mixed category functions as a nominal modifier. The PNM functions as the pronominal subject argument associated with the verbal
mixed category (to which it is affixed).
The relativized nominal appears in DOM2 and is external to the verbal mixed category’s clausal domain, specifically, DOM1.
The PNM’s distribution is local, appearing within DOM1 as defined by the VMC.
3. Locality “‘locality’. . . refers to the proximity of the agreeing element within the clause structure; a local agreement relation is one which holds between elements of the same simple clause, while a non-local agreement relation is one which may hold between elements of different clauses. (Bresnan and Mchombo) 1987:752)
Locality obtains in Type 1 since the PNM is internal to domain 1, the domain of the verbal mixed category.
Both grammatical agreement and pronominal inflection/pronoun incorporation are
local.
126
126
“...anaphoric agreement as distinct from grammatical agreement is predicted to be non-local, i.e. it can carry over between elements of distinct simple clauses. The reasoning supporting this claim is that the arguments must be expressed within the phrase structures headed by the predicate, be marked on the predicate itself, or otherwise remain unexpressed and be anaphorically controlled by non-local elements (Bresnan and Mchombo 1987, 752). Verbs agree grammatically only with their governable arguments, so grammatical agreement must be structurally local to the verb. In contrast, the antecedent of the anaphoric (non-reflexive) relationship is not related to the argument position by government, therefore there is no requirement for it to be structurally local.” Nikolaeva on Northern Ostyak 1999. 4. Empirical focus of the presentation
The clause headed by the verbal mixed category functions as a nominal modifier. (just as in Type 1)
The PNM functions as a pronominal SUBJ argument associated with the verbal
mixed category (just as in Type 1).
Given evidence that the NP head is external to the clausal domain defined by the mixed category, the pronominal subject, represented by the PNM, is (from a descriptive perspective) non-local.
Non-local here means that the PNM is external to domain 1, appearing on the head
NP in domain 2.
The PNM cannot have a possessor reading, despite displaying the surface istribution of possessive marking (see below).
5. Summary
Possessive relatives are very similar to the ‘ordinary’ Type 1 prenominal relatives that are found in closely related languages. (Type 1: Eastern Ostyak, Eastern Armenian vs. Type 2: Northern Ostyak, Western Armenian)
Yet, they appear to violate locality expectations adopted by all linguistic theories
and, consequently, their existence is not predicted by any linguistic theory.
127
127
6. Goals of the presentation
Big theoretical challenge
(a) Can we do more than adapt and adjust familiar theoretical toolboxes so that they redescribe the data using favorite assumptions?
(b) Can we both describe and explain the odd distributions of PNMs using uniform and
independently motivated linguistic assumptions which are also plausible psycholinguistically and diachronically? (see Bresnan and Kaplan 1983 on Strong Linguistic Competence)
Suggest that only analyses which view possessive relatives in constructional
terms can:
(a) explain their systemic cross-language motivation as the result of an ensemble of properties from contributing constructions, specifically, Modifier Constructions, Possessive Constructions, and Non-finite clause constructions.
(b) account with the same theoretical tools for the language particular (and sometimes
idiosyncratic) encodings, i.e., develop a proposal that address gradience in grammar, degrees of relatedness between constructions, and degrees of regularity within constructions.
7. Desiderata for any adequate theory
Particular set of properties jointly characterize the possessive construction wherever it occurs in our Eurasian corpus.
Desideratum 1: Address the reliable identities in form and function for both pronominals and lexical NPs exhibited by nominal possessive and possessive relative constructions.
128
128
Desideratum 2: Address the nature of the shared semantic relation evident in nominal possessive and possessive relative constructions.
Desideratum 3: Address the parallelism between the pronominal subject use of PNMs in possessive relatives and other non-finite clauses, as well as the differences in distribution PNMs evident in possessive relatives versus other non-finite constructions.
Desideratum 4: Address the semantic and syntactic parallelisms between simple adjectives in modificational constructions and the verbal mixed categories heading prenominal relative constructions, as well as those properties shared by these verbal modifiers and other derived nominal modifiers. Desideratum 5: Address the unifying generalizations beyond language specific differences concerning the surface encoding of morphosyntactic properties such as case, number, and person/number, i.e., with respect to linear orders of separate markers versus cumulative exponence, as well as beyond those differences concerning affixal, clitic, or periphrastic realization status. Desideratum 6: Address the variety of resolutions attested for the simultaneous expression of subject and possessor that expresses its systemic status within particular grammars, whether this can be related in a straightforward way to independent (sub)patterns in the language or must be stipulated as a construction specific strategy.
8. Theoretical assumptions: A construction-theoretic approach
The semantic and systemic motivation guiding the function and surface form of possessive relative construction wherever it occurs is captured by highly articulated lexical representations in tandem with multiple-inheritance type hierarchies of the sort found in Construction Grammar (Fillmore et. al., 1988,Goldberg 1995, 2006,O’Connor, Kay, Fillmore 1999, Fillmore, Kay, Michaelis, Sag to appear, among others) and Head Driven Phrase Structure Grammar ( Sag 1997,Ackerman & Webelhuth 1998,Malouf 2002 2003, Ginzburg & Sag 2001,Jackendoff 2002, Goldberg+Jackendoff 2004, Culicover & Jackendoff 2005}.
Construction-theoretic morphological and syntactic assumptions grounded in
language development such as representational redescription Karmiloff-Smith 1995 and analogy-based pattern learning Tomasello 2003.
9. Paradoxical result: Possessive relatives are nothing special
Analyzed construction-theoretically, possessive relative constructions are the consequence of a general cross-linguistic tendency for relative constructions to be analogically related nominal possessive constructions and modifier constructions. See Newmeyer 2006.
129
129
When a language's nominal possessive construction is head-marked, i.e., the possessor is marked on the possessed, then this analogical tendency yields the appearance of non-locality. In other words, possessive relatives result from the convergence of a frequent correlation between relatives and possessives and the fact that possesssive constructions in some languages are head-marked. If we treat possessive relatives as reflecting some sort of special syntax associated with raising etc. we lose the fact that these are really ordinary relative construcctions that happen to occur in languages with head-marked nominal possessive constructions.
10. Some simple properties of contributing constructions
The Nominal Possessive Construction: Morphological encoding within Tundra Nenets73
11. Possessive lexeme formation
Given a common noun lexeme, the Possessive Lexical Rule produces a new lexeme which has two semantic arguments, i.e., a ``possessor'' and a ``possessee'', with a genitive specifier corresponding to a lexical nominal possessor and nominative for a pronominal possessor.
The meaning of the new lexeme is an underspecified two place relation � which
obtains between the meaning associated with the argument identified with the specifier and the meaning of the original lexeme.
73 The paradigm used below is also used to mark two-place semantic relations for other lexical categories; i.e., verbs and postpositions. It is the paradigm for transitive verbs used to mark the person/number of the subject and the number of the object, and the paradigm used to mark the pronominal object of adpositions. There are different paradigms used to encode the different case and number of the possessed.
130
130
A Possessive Realization Rule produces a fully inflected possessed noun form from a possessed noun lexeme.
12. The Possessive Realization Rule (in the spirit of Word and Paradigm morphology)
Produces a fully inflected possessed noun form from a possessed noun lexeme.
The actual morphological realization of a possessed noun in Tundra Nenets depends on phonological and morphological properties of the lexemic stem and the indices of the noun and the specifier.
A sample realization: Lexical realization for third person singular nominative possessed of the lexeme \emph{ti} `reindeer'
131
131
13. Semantics of possessive constructions: A general associative relation
In Tundra Nenets, as elsewhere in Eurasia, ``possessive'' constructions, are used to express a wide range of relations between two nominals.
“a nominal in the genitive case used in the expression of adnominal determination designates not only possession in the substantive sense of this word, but also a relation (relevance), concerning a characteristic of one entity with respect to another entity.” Tereshchenko 1956:64
Possessive constructions'' represent a vague two place associative relation between
a `possessor' and a `possessed' argument.
132
132
In extrinsic possession (Barker 1995, Jackendoff 1977, Partee 1997,Partee & Borschev 2003) the precise nature of the associative relation is determined pragmatically or contextually:
where the construction provides a relation designated by R which is contextually specified.
This contrasts with lexical possession where the semantic relation between nominals is determined by the semantics of the relational head nominal; e.g., in child's mother the head mother itself specifies the nature of the relation.
14. Hypothesis for the motivation of possessive relatives
How is this accomplished? It exploits how modification operates syntactically and semantically in the NP.
15. Modifier/head construction
Modifiers obligatorily agree with the head for number (singular/dual/plural)
Modifiers optionally agree with the case of the head and the person/number of the possessor.
16. Syntax and semantics of modification (a) Formally, we can say that the number and case of the head and the index of the
possessor are part of the head noun's concord value
133
133
134
134
Representation for (15a):
17. A possessive relative with optional person/number agreement
135
135
18. The Prenominal Relative Lexical Rule Produces a participle lexeme which selects for one fewer arguments than the original participle and projects a non-finite clause. \item The `missing' argument is instead identified with the common noun the participle modifies. \item This relativized noun must be a possessive word (i.e., select for an \textsc{np} specifier), and the meaning of the participle is unified with the underspecified relation $\Re$ in the meaning of the noun (not the same semantic relation as we find with intersective modifiers, but still consistent with the Head/Modifier Construction). \item This has the effect of co-indexing the noun's possessor with the participle's subject.
136
136
19. Representation for 17
20. An obvious question}
Hey, wait a minute! Since possessive relatives have the form of nominal possessive constructions, what happens when the relativized head really is itself possessed?
21. Language particular resolutions
Unlike the cross-linguistic motivation that leads to uniformity in the expression of possessive relatives, there is significant variation (both between and within languages) in how they resolve the problem created by interpreting prenominal relatives as a instances of the possessive construction.
In Nenets, the mixed category hosts the pronominal subject argument (as in Type 1 constructions):
Otherwise, this language does not use the Type 1 strategy (VMC-inflected relatives).
In Western Ostyak:
Otherwise, this language is always pro-drop with respect to subject pronominals.
137
137
In Western Armenian, preliminary investigation suggests that speakers use (at least) three strategies:
22. A paradoxical result
A paradoxical result of the correct analysis is that by assimilating possessive relatives to the class of constructions analogically based on nominal possessive constructions and modification constructions, it both predicts that this may be a more widespread process than attested in Eurasian possessive relatives.
If so, then it renders possessive relatives even less odd, since their non-locality
effects are simply the consequence of this more broadly attested analogy as realized in a language with head-marking possessive constructions.
138
138
Wolof (Torrence 2005)
Mandarin (Li & Thompson)
23. Conclusions
Possessive relative constructions (Type 2 relative strategy) look superficially strange, particularly with respect to semantic scope and the locality of surface exponents.
However, they are straightforwardly explained as arising from the interaction of
independent constructions:
139
139
Opportunistic use of the inherent vagueness of `possessive' semantics associated. with possessive constructions
Construction specific variation on the syntax and semantics of modification.
Morphological similarity between PNM markers found in the possessive paradigm
and non-finite clause constructions (Ackerman, Nikolaeva, Malouf to appear).