lexical interface 3 oct 28, 2015 – day 26 brain & language ling 4110-4890-5110-7960 nsci...

Download LEXICAL INTERFACE 3 OCT 28, 2015 – DAY 26 Brain & Language LING 4110-4890-5110-7960 NSCI 4110-4891-6110 Fall 2015

If you can't read please download the document

Upload: jeffry-conley

Post on 17-Jan-2018

220 views

Category:

Documents


0 download

DESCRIPTION

Grades Q1Q2Q3Q4Q5Q6 MIN AVG MAX 10 10/26/15Brain & Language - Harry Howard - Tulane University 3

TRANSCRIPT

LEXICAL INTERFACE 3 OCT 28, 2015 DAY 26 Brain & Language LING NSCI Fall 2015 Course organization Schedule:topicstopics Today's chapter:Fun with https://www.facebook.com/BrLg15/https://www.facebook.com/BrLg15/ 10/26/15Brain & Language - Harry Howard - Tulane University 2 Grades Q1Q2Q3Q4Q5Q6 MIN AVG MAX 10 10/26/15Brain & Language - Harry Howard - Tulane University 3 THE LEXICAL INTERFACE 2 10/26/15Brain & Language - Harry Howard - Tulane University 4 The lexical interface 10/26/15Brain & Language - Harry Howard - Tulane University 5 Areas ~ hubs ~ effects = sensorimotor semantics 10/26/15Brain & Language - Harry Howard - Tulane University 6 Hypotheses 10/26/15Brain & Language - Harry Howard - Tulane University 7 STS phonological net p(MTG+ITS) lexical interface a(MTG+ITS) combinatorial net 1 aIFG combinatorial net 2 STS phonological net action words, tools motor + somato cortex a(MTG+ITS) combinatorial net 1 ??? aIFG combinatorial net 2 ??? imageable words medial temporal gyrus imageable words medial temporal gyrus Hickok & Poeppel, symbolic? Plvermller, sensorimotor or embodied Some semantic relations synonymy words share the same meaning: violin ~ fiddle antonymy words have opposite meanings: long ~ short hypernymy one word contains the meaning of another in a taxonomy: animal ~ horse hyponymy one word is contained in the meaning of another in a taxonomy: horse ~ animal holonymy one word is a whole for the meaning of another: hand ~ finger meronymy one word is a part for the meaning of another: finger ~ hand metonymy a part of a concept stands for the whole concept: Hollywood ~ American movie industry polysemy multiple meanings 10/26/15Brain & Language - Harry Howard - Tulane University 8 10/26/15Brain & Language - Harry Howard - Tulane University 9 Semantic networks 10/26/15Brain & Language - Harry Howard - Tulane University 10 Ingram goes into great detail on Quillians Teachable Language Comprehender (TLC); I could not find an image, but this illustrates the idea just as well. The linkages in such a network are semantic the relationships of meaning mentioned above, such as hyponymy; these are necessary, in the sense that a robin is by definition a kind of bird. or associative established by the fact that certain words are often used together, such as pig and farm; these are accidental, in the sense that there is nothing in the meaning of pig that requires one to be associated with farms; they are often defined in a free association test, by giving a subject the prime word and asking her to say the first word that comes mind; 10/26/15Brain & Language - Harry Howard - Tulane University 11 LEXICAL SEMANTICS 3 Ingram: III. Lexical semantics, 10. 10/26/15Brain & Language - Harry Howard - Tulane University 12 To prime the pump The facilitatory effect that presentation of an item can have on the response to a subsequent item usually measured in terms of reaction time 10/26/15Brain & Language - Harry Howard - Tulane University 13 Semantic + associative vs. non-associative prime-probe relations Table 10.4, Moss et al. (1995) Semantic relation Category coordination [taxonomy] Function NaturalArtifactInstrumentalScripted Associated cat dogboat shipbow arrowtheater play brother sistercoat hatumbrella rainbeach sand Non- associated aunt nephewairplane trainknife breadparty music pig horseblouse dressstring parcelzoo penguin 10/26/15Brain & Language - Harry Howard - Tulane University 14 Increased priming with respect to control condition in which there is no relationship between prime and probe: unrelated (control, not shown) < semantic + non-associative < semantic + associative Leftovers The modality of presentation has a large influence. Auditory priming fades much more quickly than visual priming. Priming has shown that multiple word meanings are activated before a word is actually recognized. This reminds me of the TRACE model, but semantic networks work like TRACE. 10/26/15Brain & Language - Harry Howard - Tulane University 15 Activation in a semantic network 10/26/15Brain & Language - Harry Howard - Tulane University 16 Semantic feature assignment Table 11.2 manwomanboygirlmarecolt human++++ female+++ mature+++ 11/01/113Brain & Language - Harry Howard - Tulane University 17 manwomanboygirlmarecolt man woman31210 boy3202 girl311 mare31 colt3 Semantic similarity scores Table 11.3 Features as a network 1 excitation 11/01/113Brain & Language - Harry Howard - Tulane University 18 human female mature man woma n boy girl mare colt Activation of man will wind up activating female, which is a contradiction. Features as a network 2 excitation, inhibition 11/01/113Brain & Language - Harry Howard - Tulane University 19 human female mature man woma n boy girl mare colt Activation of man will still wind up activating female, but inhibition will now turn it off. Features as a network 3 excitation, inhibition 11/01/113Brain & Language - Harry Howard - Tulane University 20 human female mature man woma n boy girl mare colt In cortex, long-distance connections are excitatory, while short-distance connections are inhibitory. Activation of man will wind up activating female, but inhibition of woman will turn it off. 11/01/113Brain & Language - Harry Howard - Tulane University 21 Correlated feature theory The way we go from feature representation to neural organization is by hypothesizing that correlation among the features of an object leads to mutually reinforcing activation (co-activation) in the features' neural representation shared properties are inter-correlated and so become strongly activated and less susceptible to damage, distinctive properties are weakly correlated and so become weakly activated and more susceptible to damage. Performance depends on task If the task requires access to the distinctive features of an object, then a deficit for animates will emerge, due to the lesser degree of correlation among their distinctive features. So CFT proposes that category-specific deficits develop from damage to a unitary, distributed semantic system, not from damage to anatomically distinct, content-specific stores Feature network for animates excitation, mutually reinforcing activation (excitation) 11/01/113Brain & Language - Harry Howard - Tulane University 22 head camel crocodile duck penguin zebra torso legs hump eyes bill stripes Inanimate vs. animate, side by side Inanimate few overlapping and inter- correlated features, relatively more distinctive features, and they tend to be more strongly correlated with one another. inanimate concepts are less easy to confuse with one another. Animate many overlapping and inter-correlated features (legs, eyes, teeth), few distinctive features (mane, hump, pouch), and they are only weakly correlated with one another. animate concepts are easy to confuse with one another. 11/01/113Brain & Language - Harry Howard - Tulane University 23 Problem Correlated feature theory cannot account for other patterns of impairment, such as cases in which artifacts are more poorly identified than living things. 11/01/113Brain & Language - Harry Howard - Tulane University 24 Final project Improve a Wikipedia article about any of the topics mentioned in class or any other topic broadly related to neurolinguistics. Write a short essay explaining what you did and why you did it. Print the article before you improve it, highlighting any subtractions. Print the article after you improve it, highlighting your additions. 10/26/15Brain & Language - Harry Howard - Tulane University 25 NEXT TIME More on the lexical interface: word semantics 10/26/15Brain & Language - Harry Howard - Tulane University 26