1 / 38

Rich lexical representations and conflicting features

Lotte Hogeweg. Rich lexical representations and conflicting features. Overview . My goal in this talk: propose a different view on lexical interpretation Rich lexical representations Conflicts arise when words are combined Conflicts are solved through optimization

bart
Download Presentation

Rich lexical representations and conflicting features

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. LotteHogeweg Rich lexical representations and conflicting features

  2. Overview My goal in this talk: • propose a different view on lexical interpretation • Rich lexical representations • Conflicts arise when words are combined • Conflicts are solved through optimization • Provide (preliminary) experimental evidence

  3. Lexical semantics . • My mouse is eating cheese • My mouse is broken • She is poor but nice • Life is but a dream • She knocked on the door • He walked through the door

  4. Lexical semantics • Sense enumeration • Word • Sense 1 • Sense 2 • Sense 3 • New meanings • Separate senses? • I love fast food • I love fast cars

  5. Lexical semantics • Underspecification (e.g. Reyle 1993, Pustejovsky 1995, Blutner 1998, 2004) • Underspecified representations containing for example place holders

  6. Lexicalsemantics • However, most authors agree that the interpretation of an utterance includes what is mostly considered “conceptual” knowledge: Bierwisch and Schreuder (1992): “… the concept structure CS, in terms of which the actual interpretation of linguistic expressions is specified…” (p. 32, boldface mine]. Blutner (2009): “Of course it is not sufficient to postulate underspecified lexical representations and to indicate what the sets of semantically possible specifications of the variables are. In order to grasp natural language interpretation it is also required to provide a restrictive account explaining how the free variables are instantiated in the appropriate way” (p. 5).

  7. Lexicalsemantics Anderson and Ortony (1975) • Recall test • Sentence: He pounded the stake • Trigger: Hammer • “The present research suggests that sentence comprehension and memory involve constructing particularized representations whose sense cannot be reliably predicted from the dictionary readings of the constituent words”

  8. Lexical semantics • Hardly ever specified how the detailed representation is derived • An exception is Blutner (2009): Lacking information is filled in by means of abduction rules. The price is determined based on a Horn Clause Knowledge base containing clauses of the form p1 , ... pn -> q, where the literals pj in the antecedent are annotated with weights.

  9. Lexicalsemantics • Most common view in formal semantics: word meanings are underspecified and detailed interpretation is ascribed to world knowledge • How does a detailed representation of meaning come about? • Insightful to look at research on metaphors

  10. Most extreme examples of the flexibility of word meanings: My cousin is such a mouse Lots of psycholinguistic research on metaphors Metaphors

  11. Metaphors • Most theories assume some role for the mechanism of suppression • Next: study by Rubio Fernández (2007) to illustrate the mechanism of suppression in the interpetation of metaphors

  12. Subjectspresentedwith spoken sentences: a context sentence and a target sentence: Nobody wanted to run against John at school. John was a cheetah. 60 participants Set of 22 commonnouns (e.g. cheetah) withpredictablesuperordinates (e.g. cat) and distinctive property (fast) Rubio Fernández (2007)

  13. The subjects carried out a lexical decision task at 0, 400 or 1000 ms after the word recognition point of the critical word The target words were presented visually If the critical word was cheetah, the target words were for example cat and fast Assumption: facilitation relative to an unrelated control is indicative of property activation Rubio Fernández (2007)

  14. Rubio Fernández (2007)

  15. Beyond metaphors • Rubio Fernández: the interpretation of metaphors involves the suppression of irrelevant features • Are metaphors unique? • Next: experiment testing the interpretation of coerced nouns as in stone lion • In between literal and figurative language • More in focus of formal semanticists • More directly compositional

  16. Privative adjectives entail the negation of the noun property they co-occur with: I prefer to wear fake fur A similar effect arises in: In front of the building is a stone lion This change of meaning of the noun due to privative adjectives or adjectives like stone, is called coercion (e.g. Partee 2001). The denotation of the noun is coerced to include referents not normally denoted by the noun. Beyond metaphors

  17. Beyond metaphors • Hypothesis based on metaphor research: • Stone lion: features like ‘roars’, ‘hunts’ etc. are suppressed • If hypothesis is confirmed… • Underspecification view (with enrichment process as described by Blutner 2004) cannot be correct • In contrast, words would be stored with a very reach meaning that can be weakened in a context

  18. Experiment: material • 64 Dutch adjective-noun combinations (e.g. stone lion) • Adjectives all express materials or implied a material (e.g. knitted) • Three target words per combination (matched for frequency) • A word expressing a property that is compatible with the noun but not the adjective-noun combination (e.g. roar) • A word expressing a property that is compatible with the noun and the adjective-noun combination (e.g. mane) • A completely unrelated word (e.g. book)

  19. Experiment: material • Pretest: Online questionnaire How “typical” is the property for the noun or adjective-noun combination on a scale from 1-7? lion - roar lion - mane stone lion - roar stone lion - mane

  20. Experiment: material • Pretest lion – roar at least 5 lion – mane at least 5 stone lion – roar at most 3 stone lion - mane at least 4.5 • 36 sets left (after testing 35), e.g.: Wooden moon (round, light) Paper sheep (white, soft) Felt plum (purple, pit) Plastic turd (brown, smell)

  21. Experiment: material • For each adjective-noun combination a context was constructed containing two sentences e.g.: De dierentuin in Antwerpen kan je niet mislopen. Links voor de ingang staat een stenen leeuw You can’t miss the zoo in Antwerp. In front of the entrance, there is a stone lion.

  22. Experiment: design • Contexts presented auditory • Target word presented visually 0/400/700ms after prime offset (SOA) • Each participant saw only one type of target per prime-sentence • Each participant saw 12 targets per type • Each participant was assigned a SOA

  23. Experiment: design • Factor 1: relatedness • 3 levels: AN-compatible/AN-incompatible/unrelated • Within subjects • Factor 2: stimulus offset asynchrony • 3 levels: 0 ms/400ms/700ms • Between subjects

  24. Experiment: subjects • 108 (12 per condition) subjects • Native speakers of Dutch • Mostly students (mean age 21.6) • 69 females, 39 males • They received 5 euro

  25. Experiments: results

  26. Experiment: results

  27. Experiment: results • Hearing a noun activates semantic features • Features compatible with the adjective are still primed at 700 ms • Features incompatible with the adjective are no longer significantly primed at 700 ms • Scale of “literalness”: metaphors (1000ms) → noun coercion (700ms) → classic ambiguity (400ms)

  28. Optimality Theory • When words are combined, their semantic features may be in conflict • The features that cannot be integrated in the context are suppressed • Conflict between being faithful to lexical meaning and avoiding contradictory interpretations

  29. Optimality Theory • Suppression is not random • There are rules/constraints that operate at a level of representation at which features like ‘shape’, ‘taste’ and ‘material’ are represented

  30. Optimality Theory • Rich lexical representations (for example as in Generative Lexicon (see also McNally 2005) • More and possibly defeasible information

  31. Optimality Theory Input: underlying representation • (Partial) representation lion: λx [Lion (x) ∧ animate(x) ∧ material of (organic, x) ∧ mamal(x) ∧ suckles young(x) ∧ color of (yellowish, x) ∧ image of (lion image, x) etc.] • Representation stone: λx[stone(x)]

  32. Optimality Theory Animal MATERIAL: organic Mamal FEATURE: suckles young Fish Herring Lion IMAGE: lionshape COLOR: yellowish MATERIAL: stone

  33. Optimality Theory (Partial) representation of lion including information about inheritance: λx [Lion0 (x) ∧ animate2(x) ∧ material of2(organic, x) ∧ mamal1(x) ∧ suckles young1(x) ∧ color of0(yellowish, x) ∧ image of0 (lion image, x) etc.]

  34. Optimality Theory Constraints • Faith: features in the input must be present in the output • Avoid contradiction (Hendriks and de Hoop 2001) • Non-vacuity Principle (NVP): In any given context, try to interpret any predicate so that both its positive and negative extension are non-empty (Kamp and Partee 1995)

  35. Optimality Theory

  36. Conclusions • Interpretations are fine grained • Hearing words activates semantic features • Combining words may lead to conflicting features • Conflict is best analyzed in OT

  37. To do • More fine-grained statistical analysis of experimental results • More precise representations such that together with the constraints they predict the right interpretation when words are put in several contexts • Experiment testing “simple” noun-adjective combinations like rotten banana

  38. Thanks to • Netherlands Organization for Scientific Research (NWO) • My student assistents: Judith Schellenberger & ThijsTrompenaars • You all for your attention

More Related