Introductiontwo major types of information available when we access the lexical entry for a wordinformation about the word’s meaninginformation about the syntactic and thematic rolesgoal of sentence interpretation is to assign thematic roles to words in the sentence being processedthematic rolesagentthemerecipientlocationsourcegoaltimeinstrumentverb’s argument structure (subcategorization frame)verbs and argument structures are important in parsingParsing: a process of computing the syntactic structure of the sentence1st step: determine the syntactic category to which each word in the sentence belongs2nd step: combine those categories to form phrases3rd step: construct a representation of the meaning of the sentenceDealing with structural ambiguityambiguity causes processing difficulty iserial autonomous modelconstruct parse tress on purely syntactic ground, then decide using semantic information whether it makes sense or not.if it does, we accept that representation; if it does not, go back and try againparallel autonomous modelconstruct all possible syntactic representations in parallel, again using solely syntactic information, and then use semantic or other information to choose the most appropriate oneuse semantic information from the earliest stages, only constructing semantically plausible syntactic representations<> interactive modelother sources of information (Semantic) influence the syntactic processoruse semantic information from the earliest stages to guide parsing, constructing semantically plausible syntactic representationsactivate representation of all possible analysesambiguitypermanent (global) ambiguitystill syntactically ambiguouslocally (temporarily) ambiguityambiguity resolved by subsequent material (the disambiguation region)e.g.) “the horse raced past the barn fell”the verb “raced” is ambiguous. could be a main verb or a past participlethis sentence contains a reduced relative clauseEarly work on parsingSize of the syntactic unit in parsingclause is a major unit of perceptual and syntactic processingclause is a part of a sentence that has both a subject and predicateclick displacement techniqueparsing is an incremental processrapidly construct a syntactical analysis for a sentence fragment, assigns it semantic interpretation, relates this interpretation to world knowledgepeople anticipate properties of upcoming words in the sentenceEarly work on parsingParsing strategies based on surface structure cuessurface structure of the sentence provides a number of obvious cues to the underlying syntactic representationtwo approachesuse cues and strategies that enable us to compute the syntactic structurecanonical sentence strategyFodor, Bever and Garrett argued that goal of parsing was to recover the underlying, deep structure of a sentenceTwo early accounts of parsingKimball proposed seven principles of parsing1st principle: parsing is top-downstart from sentence node and predict constituents2nd principle: right associationnew words are preferentially attached to the lowest possible node in the structure constructed so far3rd principle: new nodes4th principle: processor can only cope with nodes associated with two sentence nodes at any time5th principle: processor prefers to close a phrase asap6th principle: fixed structureexplains our difficulty with garden path sentences7th principle: principle of processingwhen a phrase is closed, it exits from short term memory and is passed on to a 2nd stage of deeper, semantic processingcriticism of Kimball’s ideaprinciples of processing underlie many other modelsthe role of function words in parsingFrazier and Fodor simplified Kimball’s account“Sausage machine model”:two stage model of parsingPPP (preliminary phrase packager)SSS (sentence structure supervisor)this model evolved into the garden path modelcriticism of sausage machine modelcannot account for the preference for right association in some six-word sentencesProcessing structural ambiguityGarden path model1st stage: build a tree based on syntactic categoryIf information is ambiguous, we resolve ambiguity… byuse minimal attachmentincoming materials should be attached to the phrase marker being constructed using the fewest nodes possiblelate closureincoming information attaches to the tree structure you already have2nd stage: check the semantics & see if it makes senseif it doesn't, revise the treeConstraint-based models of parsingthe processor uses constraintsconstraints: multiple sources of information, including syntactic, semantic, discourse, and frequency-baseddecide between the choices based on which has the most activation at the end of parsingsemantic information can also cause garden pathsAutonomy (Garden path model) vs. interaction (constraint-based)Evidence for autonomy in syntactic processing1) Ferriera & Clifton (1986)(26) the defendant examined by the lawyer turned out to be unreliable(27) the evidence examined by the lawyer turned out to be unreliableaccording to garden path theory, because of minimal attachment, when we come across the word “examined” we should take it to be the main verb in (26) and (27)“examined” requires an agentin (26), “defendant”can be an agent (animate), also can be what is examined.structure is ambiguous between a reduce relative clause and a main verb analysisin (27) “evidence”can’t be an agent (inanimate) it must be what is examinedthis structure can only be a reduced relative clausehowever, (27) still made participants garden-pathed.Therefore, semantic information does not prevent or cause garden-pathingsemantics do NOT matter!2) Van Gompel and Pickering (2001)(28) After the child had visited the doctor prescribed a course of injections(29) after the child had sneezed the doctor prescribe a course of injectionsaccording to garden path theory, assign “the doctor” as direct object of “visited”if semantic and thematic information about verbs is available from an early stage… “sneezed” in (29) cannot take a direct object (a process called lexical guidance)however, (29) still garden-pathed.tells us that initial parse ignores verb informationTherefore, first stage of parsing does not use semantic or thematic information. Verb information does not affect.3) O’Seaghdha(30) the message that was shut(31) the message of that shutsyntactic analysis precedes semantic analysis and is independent of it4) evidence from neurosciencesemantic and syntactic processing are independenta