decomposing morphologically complex words: across...
Post on 05-Jun-2018
219 Views
Preview:
TRANSCRIPT
Decomposing morphologically complex words: Across contexts and word types II
Robert Fiorentino
Neurolinguistics & Language Processing Lab
Department of Linguistics
University of Kansas
MorphologyFest 2014, Talk 3
Today’s plan
Talk 3: Decomposing morphologically complex words: Across contexts and word types II
Last time, I used masked priming and lexical decision to demonstrate that compounds are processed via constituents, and that transparency affected post-decompositional processing, rather than precluding decomposition
Decomposing morphologically complex words: Across contexts and word types II
Decomposing morphologically complex words: Across contexts and word types II
To summarize where we are so far:1. Evidence from lexical decision/MEG for compound
constituent activation 2. Evidence from masked priming that compounds are
segmented and yield constituent activation regardless of transparency
3. Lexical decision evidence that compound processing is both morpheme-based and affected by transparency
4. Masked, unmasked, and ERP evidence that even novel putatively complex words (including novel compounds) show constituent activation (though not always easy to distinguish from orthographic activation)
Decomposing morphologically complex words: Across contexts and word types II
Taken together, these findings are consistent in suggesting that word recognition involves segmenting or “sampling” the input for minimal morphological units.
Segmentation system:(i) is not necessarily dependent on “affix stripping”(ii) Appears robust to at least some regular morpho-orthotactic changes (McCormick et al., 2008, 2009)(iii) Operates across the board, not strongly constrained by semantic transparency or lexicality(iv) “Exhaustivity” (though recall our discussion of pseudoembedded nonwords like slegrack)
Today’s plan
Talk 3: Decomposing morphologically complex words: Across contexts and word types II
Today, I’d like to extend the discussion to include derivation and inflection, and to engage issues related to regular/irregular distinction and productivity
We will also extend the cross-linguistic coverage of our discussion along the way
Decomposing morphologically complex words: Across contexts and word types II
A major focus in research on derivation has concerned the impact of productivity on complex word recognition
Example: English de-adjectival nominalization
darkness humidity
Derivation
-nessRelatively productive
-ityRelatively unproductive
Does limited productivity lead to non-decompositional/whole-word based storage and processing?
To better understand the locus of productivity effects, cross-method research is called for
We will discuss one study from my lab that takes this approach
Our test case will be de-adjectival nominalization, but in Japanese
We’ll also examine related findings from priming and lexical decision
Derivation
It has been argued that deviations from productivity may engender whole-word storage and access for some derived words (e.g., Clahsen et al., 2003; Hagiwara et al., 1999).
However, it remains unclear whether productivity indeed constrains initial lexical access in this way.
Fiorentino et al. (2010): Productivity and Priming
Hagiwara et al. (1999): Japanese productive de-adjectival nominals (-sa suffixed words) are derived by rule, while unproductive deadjectival nominals (-mi suffixed words) are listed in the lexicon
Evidence from acceptability judgments and aphasiology
Fiorentino et al. (2010): Productivity and Priming
Clahsen et al. (2003): Only productive derivation is aligned with inflection in terms of involvement of morphological combinatorics, in contrast to unproductive derivation/irregular affixation
Evidence from cross-modal priming/lexical decision
(cross-modal priming: auditorily-presented prime word, visually-presented target word)
Fiorentino et al. (2010): Productivity and Priming
Marslen-Wilson et al. (1996): Evidence from cross-modal priming in English, arguing for morphological-level processing of productive but not unproductive affixed words
Consistent with a range of findings from other unmasked paradigms and lexical decision studies suggesting productivity effects (e.g., Ford et al., 2010, Anshen & Aronoff, 1988; Bradley, 1979, Plag & Baayen, 2009)
When do productivity effects come into play? Do they preclude initial segmentation/activation, or operate later?
Fiorentino et al. (2010): Productivity and Priming
Silva & Clahsen (2008): Masked priming experiments on priming of stems of productive English de-adjectival nominals (e.g. bitterness-bitter) and their unproductive counterparts (e.g., humidity-humid)
Robust priming of both word types, tested in separate experiments (no orthographic control condition)
Consistent with a view in which all potentially complex words are initially decomposed during visual word recognition, not initially constrained by productivity
Fiorentino et al. (2010): Productivity and Priming
Fiorentino, Naito-Billen, & Minai (2010):
Utilize masked priming and unmasked, longer-duration priming to test this, using the -sa/-mi contrast in Japanese
Examine new contrast and language
Similarities/differences with the English case
Methodological differences
Fiorentino et al. (2010): Productivity and Priming
Japanese De-adjectival Nominalization:
Productive -sa affix (e.g., haya-sa ‘quick-ness’)
Unproductive -mi affix (e.g., maru-mi ‘round-ness’)
-sa attaches to most all adjectives; -mi limited to ~30 words in Japanese Thus, we are able to test virtually all attested
examples of -mi in our study on this phenomenon
Unlike English -ness vs. -ity, -sa and -mi are matched in length and -mi does not involve stem changes
Fiorentino et al. (2010): Productivity and Priming
Japanese De-adjectival Nominalization:
-sa readily attaches to newly-coined/borrowed words; -migenerally does not
fansii-sa; *fansii-mi ‘fanciness’
It has been argued that –mi takes on more extended meanings (see Clahsen & Ikemoto, 2012 for discussion)
tsuyo-i (“strong”)
tsuyo-sa (“strength”) tsuyo-mi (“virtue”)
Fiorentino et al. (2010): Productivity and Priming
We conducted two priming studies on –sa and –mi in Japanese
I. Masked priming (49 ms SOA) experiment
Will we see similar masked morphological priming patterns with productive/unproductive words (and in Japanese)?
II. Unmasked priming (252 ms SOA) experiment
Will productivity effects emerge with this longer duration, conscious presentation of the primes?
Fiorentino et al. (2010): Productivity and Priming
Fiorentino et al. (2010): Productivity and Priming
Productive Condition: 30 i-adjective primes, –sa suffixed de-adjectival nouns as targets
Unproductive Condition: 30 i-adjective prime, –mi suffixed de-adjectival nouns as targets
Orthographic-Overlap Condition: 30 prime-target pairs matching in orthography except at the end of the word, mirrors proportion/position of prime-target orthographic overlap in the other conditions
Fiorentino et al. (2010): Productivity and Priming
In this study, we presented the –sa and –mi words as targets
Thus, in this study, we will be able to see the effect of pre-activating the root, on how each word type is processed
Fiorentino et al. (2010): Productivity and Priming
The primes for the –sa and –mi words are also complex (they carry an –i adjectival ending)
Fiorentino et al. (2010): Productivity and Priming
Primes/targets: Matched on frequency, length, proportion of overlap
Fillers: 30 additional unrelated prime-target pairs to reduce relatedness proportion to 37.5%
120 real-word prime, nonword target pairs were also included as fillers (1:1 word:nonword ratio)
Semantic relatedness: assessed for each item in pencil-and-paper rating study (N=74 native Japanese speakers)
I. Masked priming (49 ms SOA) experimentII. Unmasked priming (252 ms SOA) experiment
Trial structure:
Mask (hash marks) for 500 ms (12pt Times New Roman)
Presentation of prime (11pt PMingLiu font) for 49 or 252 ms
Presentation of target (12pt MS PMincho font) until lexical decision or 3000 ms timeout
Fiorentino et al. (2010): Productivity and Priming
Results: Masked Priming
Fiorentino et al. (2010): Productivity and Priming
Robust, equivalent priming (facilitation) for productive and unproductive conditions
Marginal inhibition for orthographic condition Effects not modulated by semantic relatedness
Results: Unmasked Priming
Fiorentino et al. (2010): Productivity and Priming
Robust priming (facilitation) for productive and unproductive conditions, but significantly reduced for unproductive
Significant inhibition for orthographic condition Effects not modulated by semantic relatedness
Results: Unmasked Priming
Fiorentino et al. (2010): Productivity and Priming
Evidence of morpheme-based processing of productive and unproductive complex word targets; pre-activation of the stem facilitates processing of both productive and unproductive target words
Evidence suggests initial access to morphemes of both productive and unproductive de-adjectival nominals, dissociable from semantic and orthographic effects
Cross-task, cross-linguistic converging evidence suggesting that productivity may constrain processes subsequent to initial morphological analysis, rather than precluding a morphological parse
Fiorentino et al. (2010): Productivity and Priming
Converging evidence:
Clahsen & Ikemoto (2012): Masked priming (-sa and –mi words were primes in this study)
Fiorentino et al. (2010): Productivity and Priming
Clahsen & Ikemoto (2012): equivalent masked priming for –sa and –mi words
Fiorentino et al. (2010): Productivity and Priming
They also tested a version with Kana/Kanji, and find a similar pattern
No orthographic condition but argued effects are not orthographic via analysis of fillers
Clahsen & Ikemoto (2012): equivalent masked priming for –sa and –mi words
Fiorentino et al. (2010): Productivity and Priming
They also tested a version with Kana/Kanji, and find a similar pattern
No orthographic condition but argued effects are not orthographic via analysis of fillers
Clahsen & Ikemoto (2012): equivalent masked priming for –sa and –mi words
Fiorentino et al. (2010): Productivity and Priming
They also tested a version with Kana/Kanji, and find a similar pattern
No orthographic condition but argued effects are not orthographic via analysis of fillers
Clahsen & Ikemoto (2012): Whole-word frequency effect found in unprimed lexical decision in another experiment for both word types
Fiorentino et al. (2010): Productivity and Priming
Clahsen & Ikemoto (2012): Whole-word frequency effect found in lexical decisions in another experiment for both word types
Fiorentino et al. (2010): Productivity and Priming
Clahsen & Ikemoto (2012): Whole-word frequency effect found in lexical decisions in another experiment for both word types
Fiorentino et al. (2010): Productivity and Priming
Clahsen & Ikemoto (2012): Whole-word frequency effect found in lexical decisions in another experiment for both word types
Fiorentino et al. (2010): Productivity and Priming
Taken together, these findings suggest initial, across-the-board access to morphological constituents regardless of productivity, and subsequent effects of morpheme combination that are modulated by productivity (and in Clahsen & Ikemoto, 2012, frequency)
Significant equivalent masked priming for -sa/-mi: (i) When -sa/-mi words were targets (Fiorentino et al.,
2010) in Kanji/Kana(ii) When -sa/-mi words were primes (Clahsen &
Ikemoto, 2012) in Kanji/Kana or all Kana
Fiorentino et al. (2010): Productivity and Priming
Taken together, these findings suggest initial, across-the-board access to morphological constituents regardless of productivity, and subsequent effects of morpheme combination that are modulated by productivity (and in Clahsen & Ikemoto, 2012, frequency)
Productivity effect emerges for -sa/-mi:
(i) In longer-lag, unmasked priming (Fiorentino et al., 2010), with morpheme activation also still evident
Fiorentino et al. (2010): Productivity and Priming
Taken together, these findings suggest initial, across-the-board access to morphological constituents regardless of productivity, and subsequent effects of morpheme combination that are modulated by productivity (and in Clahsen & Ikemoto, 2012, frequency)
Productivity effect emerges for both -sa/-mi:
(i) In unprimed lexical decision (Clahsen & Ikemoto, 2012)
Fiorentino et al. (2010): Productivity and Priming
Decomposing morphologically complex words: Across contexts and word types II: Interim summary
Segmentation - Activation - Composition
Constituent priming; M350;
N400; RT (w/caution)
Not limited to affix stripping; exhaustivity;
not constrained by lexicality, transparency,
productivity
Constraints on morpheme
combination:Assembly of
meanings from parts? Transparency effects
Productivity
Priming data allow inferences regarding
segmentation
..and motivate questions regarding the (existence and) nature of post-
decompositional processes
Decomposing morphologically complex words: Across contexts and word types II
Masked morphological priming: how about discontinuous/interleaved morphology?
e.g., Boudelaa & Marslen-Wilson, 2005, 2011; Deutsch et al., 1998; Frost et al., 1997, among others
Let’s take a look at Boudelaa & Marslen-Wilson (2005)
Decomposing morphologically complex words: Across contexts and word types II
Arabic morphology: discontinuous/interleaved
Root: tri-consonantal (associated with semantic content)
Word pattern: typically vowels/also can be consonants (convey syntactic information)
[xatama] “seal”
Root: {xtm} conveys ‘sealing’Word pattern: {faʕala} conveys active, perfective.
(Boudelaa & Marslen-Wilson, 2005)
Decomposing morphologically complex words: Across contexts and word types II
Let’s look at Boudelaa & Marslen-Wilson (2005, Ex. 2):
Masked priming in Arabic verbs, testing prime-target pairs including:
(1) Those sharing a word pattern(2) Those sharing a root (also semantically related)(3) Those sharing a root (not semantically related)(4) Purely orthographically related conditions(5) Purely semantically related conditions
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Shared word
pattern
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Ortho overlap to mimic WP
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Shared root
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Shared root,
opaque
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Ortho condition to mimic roots
Boudelaa & Marslen-Wilson (2005): Ex. 2 (masked priming in Arabic verbs)
Only semanticrelation
Decomposing morphologically complex words: Across contexts and word types II
Boudelaa & Marslen-Wilson (2005, Ex. 2): Results
Root priming: robust regardless of semantics, across stimulus-onset-asynchronies (SOA)
Word pattern priming: at 48 ms SOA
Semantic priming: absent
Orthographic priming: largely absent (save at the longest SOA)
Boudelaa & Marslen-Wilson (2005): Results (Ex. 2, verbs)
Boudelaa & Marslen-Wilson (2005): Results (Ex. 2, verbs)
Boudelaa & Marslen-Wilson (2005): Results (Ex. 2, verbs)
Boudelaa & Marslen-Wilson (2005): Results (Ex. 2, verbs)
Decomposing morphologically complex words: Across contexts and word types II
Boudelaa & Marslen-Wilson (2004): similar effects when testing strong and weak roots
Boudelaa & Marslen-Wilson (2010): priming effects for word pattern only in context of productive roots (i.e., roots with relatively large type frequency/morphological family)
Thank you!
fiorentino@ku.edu
Neurolinguistics & Language Processing Lab:www.neuroling.ku.edu
top related