1. Optimality Theory Lexical Semantics Tandem workshop on Optimality Theory in language and geometric approaches to language

2. Overview  Overview OT & lexical semantics  Consequences of OT view on lexical semantics  Content words and OT  Uni- versus bidirectional OT  Experimental testing  Conclusions

3. OT lexical semantics Previous work: Fong (2003): use of already in Colloquial Singaporean English and Standard English Zwarts (2004): interpretation of preposition (a)round Zeevat (2002): various discourse markers Zwarts (2008): production of prepositions Hogeweg (2009): discourse marker wel

4. OT lexical semantics Word 1: {f1, f2, f3, f4} Word 2: {f1, f5, f6, f8, f9} Word 3: {f3, f9, f10, f11} Word 4: {f11, f12, f13} Word 5: {f15} Lexicon Input: meaning/intention (set of features). For example {f1, f2, f9} M ARK F AITH D EP word 1: {f1, f2, f3, f4}**  word 2: {f1, f5, f6, f8, f9} **** word 3: {f3, f9, f10, f11}***** word 4: {f11, f12, f13}*** word 5: {f15}**** Production Input word from lexicon (set of features) + context. For example: word 4 {f11, f12, f13}. F ITFAITH * INVENT {f1, f2, f3, f4, f5}******** {f1, f11, f12, f13}** {f11, f12, f13}*  {f12, f13}* {}*** Interpretation

5. Consequences of OT view on LS  There is no strict relation between word and meaning but a meaning is the output of the process that takes a word as an input or the input to the process has a word as an output  Whether a concept is labeled by a particular word does not only depend on the stored information for that word but also on the stored information of competing words, that is, competition is important  The meaning of words is overspecified in the lexicon

6. Consequences: overspecification  Most current theories of lexical semantics argue that lexical representations are underspecified and can be strengthened by contextual information (e.g. Reyle 1993, Pustejovsky 1995, Blutner 1998, 2004)  The previous studies (except for the original analysis in Zwarts 2008) assume an overspecified lexical representation, which can be weakened by the context

7. OT and the semantics of content words  An overspecified lexical representation includes what is usually considered conceptual, encyclopedic or commonsense knowledge  For example, lexical knowledge specifies that an apple has a stem, pulp, a peel and that apples can be red or green: {pulp, stem, peel, red ∨ green}

8. OT and the semantics of content words Lolly pop body stick green round apple brown long BODY COLOR SHAPE TASTE COLOR SHAPE STICK Lolly pop-frame (based on Petersen 2007) λ x[lolly pop(x) ∧ body of (body, x) ∧ color of (green, body) ∧ shape of(round, body) ∧ taste of(apple, body) ∧ etc.] λ x ∃ y[lolly pop(x) ∧ body of (y, x) ∧ color of (green, body) ∧ shape of (round, body) ∧ taste of(apple, body) ∧ etc.]

9. OT and the semantics of content words  Hogeweg (submitted):  Ranking of faithfulness constraints pertaining to attributes and values  Interacting with well-known semantic principles:  Selective binding (Pustejovsky 1995)  If α is of type, β is of type b, and the qualia structure of β, QS β, has quale, q of type a, then αβ is of type b, where { αβ } = β ∩ α (q β )  Non-vacuity principle (Kamp & Partee 1995)  In any given context, try to interpret any predicate so that both its positive and negative extension are non-empty’  Fit (e.g. Zwarts 2005)

10. OT and the semantics of content words  Well known constraint F IT which penalizes candidates with inherent contradictions  When there are two or more values for the same attribute,  When a value or attribute is of the wrong type as determined by a type hierarchy (for example the value warm for the attribute color or the attribute color for the type smell)  When a value or an attribute is in incongruence with an inherited value

11. OT and the semantics of content words Input: λ x[apple(x) ∧ peel of (peel, x) ∧ color of (red ∨ green, peel) ∧ stem of (stem, x) ∧ color of (brown, stem) ∧ etc.] + λ x[lilac (x)] F IT NVP*NB F AITH ∃ x[color of(x, peel)] F AITH color of (red, peel) F AITH color of (green, peel) F AITH ∃ x[color of (x, stem)] F AITH color of (brown, stem) y = apple (peel = red, stem is brown) ****  y = peel ** y = stem (peel = red)*** y = peel and peel = red** peel = red** Interpretation lilac apple

12. OT and the semantics of content words Word 1: {f1, f2, f3, f4} Word 2: {f1, f5, f6, f8, f9} Word 3: {f3, f9, f10, f11} Word 4: {f11, f12, f13} Word 5: {f15} Lexicon Input: meaning/intention (set of features). For example {f1, f2, f9} M ARK F AITH D EP word 1: {f1, f2, f3, f4}**  word 2: {f1, f5, f6, f8, f9} **** word 3: {f3, f9, f10, f11}***** word 4: {f11, f12, f13}*** word 5: {f15}**** Production Input word from lexicon (set of features) + context. For example: word 4 {f11, f12, f13}. F IT NVP*NB FAITH {f1, f2, f3, f4, f5}******** {f1, f11, f12, f13}** {f11, f12, f13}*  {f12, f13}* {}*** Interpretation

13. Bidirectional optimization  Aforementioned studies are all cases of unidirectional, interpretive optimization  Blutner (2004) argues that some lexical phenomena can be explained by bidirectional optimization based on a underspecified lexical meaning

14. Bidirectional optimization  f = form/underspecified representation  m = possible enrichment  f1> f2  More features > less features QI f1 { f a}, m1{ f a, f b} f1 { f a}, m2{ f a, f b, f c}* f2 { f a}, m2{ f a, f b, f c}** f2 { f a}, m1{ f a, f b}*

15. Bidirectional optimization ‘lilac’ + ‘apple’ I y = apple (peel = red, stem is brown)*  y = peel y = stem (peel = red)** y = peel and peel = red*** peel = red***** Not completely comparable: for example interpretation lilac apple Blutner 2009: bidirectional optimization is an offline process that forms language in an evolutionary sense

16. Metaphor  Metaphor: most extreme examples of the flexibility of word meanings My cousin is a grey mouse

17. Psycholinguistic testing  Experiment Fernández (2007):  Subjects presented with 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 common nouns (e.g. cheetah) with predictable superordinates (e.g. cat) and distinctive property (fast).

18. Psycholinguistic testing  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.

19. Psycholinguistic testing

20.  The hypothesis of overspecification leads to a testable prediction for a similar lexical decision experiment about other semantic phenomena, e.g. coerced nouns and “simple" noun adjective combinations.  Stone lion  Halve appel ‘half an apple’, rotten banana  An overspecification view predicts that upon hearing a noun with an adjective that is in conflict with a highly ranked property of the noun, this property is initially activated and repressed later.

21. Psycholinguistic testing  An underspecification view would predict that the feature ‘yellow’ is not activated because the lexical representation of banana does not include information about its color.  In contrast, this aspect is filled in in the enrichment process based on the context, the immediate context being the adjective rotten in this case.

22. Psycholinguistic testing  A similar hypothesis holds for the interpretation of stone lion. An overspecification view predicts that interpreting this phrase involves the suppression of features such as ‘carnivore’, ‘runs fast’ and ‘hairy’ because those are not properties of a lion made of stone while they are properties of a (real) lion.

23. Psychological testing  The experiments could show that “regular interpretation”, (metonymic) type coercion and metaphors work similarly, they all involve the suppression of irrelevant features

24. Conclusions  Previous studies in lexical semantics focused on grammaticalized, functional items  The studies assume an overspecified lexical representation  Analyses of the interpretation of content words require a more complex representation of meaning and additional constraints  Unidirectional and bidirectional optimization are perhaps not mutually exclusive  The hypothesis of overspecification leads to a testable prediction for a priming experiment