Lecture 15: Lexical semantics IIProfessor Robert C. [email protected]/9.611J SP11The Menu Bar• Administrivia• How can we learn knowledge about words?• What the seb?• How to gorp• The truth does not set you free6.863J/9.611J SP11“I saw one car flattened down to about one foothigh, and my mechanics friend told me that thedriver who got out of that cab that was squasheddown by accident got out by a narrow escape”People can learn a lot of language fromjust a little input6.863J/9.611J SP11David Hume (1711-1776)6.863J/9.611J SP11How do children learn words?6.863J/9.611J SP116.863J/9.611J SP11Word learning patterns reflect informationchange rather than (or “more than”) conceptualchange• The word learning process reflects gradual recruitment andweighting of different information sources.• Referential cues play the largest roles early because theyrequire no prior linguistic knowledge. This informationalavailability artifactually promotes the learning of concretenouns.6.863J/9.611J SP11Information change• The requirement to learn solely from observationof the contingencies for word use (word-to-worldpairing) favors concrete items.• Later in the learning sequence, other sources ofinformation become available to enable thelearning of other words: “the unobservables.• Multiple sources of information converge toexplain why learning is rapid and errorless6.863J/9.611J SP11Abstract verbs require different kinds ofinformation to acquire(Snedeker & Gleitman, 2003)Information SourcePercent of Verbs Identified6.863J/9.611J SP11Chase FleeFully overlapping contexts6.863J/9.611J SP116.863J/9.611J SP11The visual zoom lens• Directing subjects’ gaze to critical features of one interpretationsignificantly increases likelihood of corresponding interpretation(Georgiades & Harris, 1997)6.863J/9.611J SP11Linguistic duckrabbits(Gleitman January, Nappa & Trueswell, 2007; Nappa, Trueswell, & Gleitman, 2008)• “Oh look! He’s blorking him!”Chasing?)Fleeing?)Running?)Could)speaker)gaze)play)arole?6.863J/9.611J SP11Experimental DesignExp. 1. Pronoun Experiment 3yo’s (N=10) 4yo’s (N= 12) 5yo’s (N= 16) Adults (N=21)He’s)blicking)him!Neutral(SyntaxA (Dominant) Interpretation = CHASEB (Non-dominant) Interpretation = FLEE (RUN-AWAY)Character)A Character)BA?LookB?LookWill)the)speaker’s)direcBon)of)gaze)influence)meaning)of)blick??In)other)words:Will)B?Looks)increase)Non?Dominant)(B))interpretaBons)of)blick?6.863J/9.611J SP11Proportion of B (Non-Dominant) Responses(B / (A+B))0.00.10.20.30.40.50.60.70.80.91.0Char. AChar. B3 y.o. 4 y.o. 5 y.o. AdultsAndy looking at:Proportion non-dominant (B)interpretations of novel verb6.863J/9.611J SP11Summary of Experiment 1• Children battle against conceptual biases• Prefer Dominant (A) interpretations• Children use speaker’s eyegaze to infer verbmeaning• Does this happen in the absence of syntacticevidence? Or generally?6.863J/9.611J SP11meepsBlueBrownff6.863J/9.611J SP11meepsBlueBrown6.863J/9.611J SP11Experiment 2 DesignExp. 2. Full NP Experiment 3yo’s (N=12) 4yo’s (N= 25) 5yo’s (N= 16) Adults (N=24)The)dog’s)blicking)the)man!A.SyntaxThe)man’s)blicking)the)dog!B.SyntaxA)(Dominant))InterpretaBon)=))CHASEB)(Non?dominant))InterpretaBon)=)FLEE.(RUN3AWAY)Character)A Character)BA?LookB?LookCan)you)see)the)duck?Let’s)see)if)there’s)cheese)in)the)refrigeratorLet’s)see)where)the)cheese)is6.863J/9.611J SP11Proportion of B (Non-Canonical) Responses3 y.o. 4 y.o. 5 y.o. Adults0.00.10.20.30.40.50.60.70.80.91.0Type of SyntaxA B A B A B A BChar. AChar. BAndy looking at:Effect of SyntaxSmall effect of GazeConceptual (A) BiasEffects of Syntax & Gazen.s.n.s.*n.s.Proportion non-dominant (B)interpretation of novel verb6.863J/9.611J SP11Case 2: not salient, not observableTHINK6.863J/9.611J SP11Situations: When do we think about thinking?Idea: True Beliefs are notsalient candidates forlexicalization.Little Red Riding Hood’sgrandmotheris in bed. ?6.863J/9.611J SP11Little Red RidingHood thinks thather grandmotheris in bed.6.863J/9.611J SP11Scene type by Syntax typeNP compS-compTrue beliefFalse belief6.863J/9.611J SP11StimuliFalse Belief Scene6.863J/9.611J SP11Sentence and Query• Transitive Frame:“Josh’s brotherMEEPS his snack.What does thatmean?6.863J/9.611J SP11Sentence and Query• Transitive Frame: “Josh’sbrother MEEPS hissnack. What does thatmean?• Complement ClauseFrame: “Josh’s brotherMEEPS that it is abanana. What does thatmean?6.863J/9.611J SP11StimuliTrue Belief Scene6.863J/9.611J SP11Results: Adult data w/nonsense wordsDoes scene-syntax combination affect the type of V identified?% of mental V guesses6.863J/9.611J SP11Scene type by Syntax type (in %)NP comp S-compTrue belief 2.3False belief 11.2Information from the world…6.863J/9.611J SP112. And as for the environment…It’s not thetruth that setsyou free.6.863J/9.611J SP11Scene type by Syntax type (in %)NP comp S-compTrue belief 2.3 44False beliefThe information from syntax of syntax6.863J/9.611J SP11Scene type by Syntax type (in %)NP compS-compTrue belief 2.3 44False belief 11.2 82The power of multiple cues6.863J/9.611J SP11EXP 3: Children data w/GORP onlyDoes scene-syntax combination affect the type of V identified?% of mental V guesses6.863J/9.611J SP11Conclusions so far• Several cues contribute information for solvingthe mapping problem for vocabulary.• The weighting of these cues varies across wordtype (concrete > abstract)6.863J/9.611J SP11Why abstract verbs are learned late: Different verbsrequire different kinds of information to acquire(Snedeker & Gleitman, 2003)Information SourcePercent of Verbs Identified6.863J/9.611J SP11Informational inadequacies in young learnersmasquerade as ‘modules’To be used, a cue system• must be acquired. distribution: BAKE CHEF CAKE syntax: CHEF BAKE CAKE• must be reliable, so: lexical before syntactic generalizations conversational-pragmatic generalizations6.863J/9.611J SP11Feldman,)Goldin?Meadow,)&)Gleitman,)1978Intransitives always surface with their“actor” argument overtly signed6.863J/9.611J SP11This actor (“pro”) is selectedfor omission more often thanexpected by chance (33%).6.863J/9.611J SP11For 3-argument predicates, the rate ofactor production falls even further.6.863J/9.611J SP11So Let’s• Figure out how to get a program to learn…6.863J/9.611J SP11Constraints
View Full Document
Unlocking...