Date Bidirectional OT and language acquisition Petra Hendriks ESSLLI 2008 course “Bidirectional OT in natural language” Hamburg, August 15, 2008
Date >The elephant is hitting himself. Children: NO >The elephant is hitting him. Children: YES >Here is an elephant and an alligator. >The elephant is hitting himself. Children: YES >The elephant is hitting him. Children: YES Comprehension
Date Production/comprehension asymmetry: >Pronoun Interpretation Problem (e.g., Jakubowicz, 1984; Chien & Wexler, 1990; Grimshaw & Rosen, 1990, for English; Deutsch, Koster & Koster, 1986; Koster, 1993; Philip & Coopmans, 1986, for Dutch) The elephant i is hitting him i/j Until 6-7 years old >However, children’s production is adult-like from age 4 on! (de Villiers, Cahillane & Altreuter, 2006, for English; Spenader, Smits & Hendriks, in press, for Dutch) Puzzle
Date >Children: The elephant is hitting him/the alligator. >Cf. adults >Children: The elephant is hitting himself. >Cf. adults (Spenader, Smits & Hendriks, in press) Production
Date Binding Theory (Chomsky, 1981): >Principle A: A reflexive must be bound in its local domain. >Principle B: A pronoun must be free in its local domain. How can comprehension of pronouns be delayed, while production of pronouns is adult- like? Binding Theory
Date Explanations of PIP: Children possess the linguistic knowledge, but make errors due to: >Lack of relevant pragmatic knowledge (e.g., Chien & Wexler, 1990; Thornton & Wexler, 1999) >Interference of task factors (e.g., Bloom, Barss, Nicol & Conway, 1994; Grimshaw & Rosen, 1990) >Lack of sufficient processing resources (e.g., Avrutin, 1999; Reinhart, 2006) Explanations
Date If the Pronoun Interpretation Problem lies outside the grammar: >Why is production unaffected? >Why does the PIP not arise in all languages? >Why does the PIP not arise in all constructions in a language? Aim of this talk: Investigate the hypothesis that the PIP (and other asymmetries) can be explained from the grammar itself. Aim
Date Outline: The grammar: Optimality Theory Constraint reranking OT is a direction-sensitive grammar Production/comprehension asymmetries Bidirectional OT results in a symmetric system Predicting further asymmetries in acquisition (e.g., PIP, subject anaphora) Outline
Date Markedness constraints, e.g.: NoCoda: No syllables with a coda. *Dors: No dorsal segments. Faithfulness constraints, e.g.: Parse: No unparsed underlying material. Fill: No insertion of new material. Optimality Theory
Date Input: /kæt/ FAITH (Parse, Fill) MARK (NoCoda, *Dors) [kæt]* [ta]*! Tableau 1 Adults’ grammar: FAITH >> MARK Input: /kæt/ MARK (NoCoda, *Dors) FAITH (Parse, Fill) [kæt]*! [ta]* Tableau 2 Children’s grammar: MARK >> FAITH Language acquisition involves constraint reranking: Constraint reranking
Date Input: /kæt/ MARK (NoCoda, *Dors) FAITH (Parse, Fill) [kæt]*! [ta]* Tableau 2 Children’s grammar: Production Input: [kæt] MARK (NoCoda, *Dors) FAITH (Parse, Fill) /kæt/ /hæt/*! Tableau 3 Children’s grammar: Comprehension Production and comprehension yield different results: Smolensky (1996)
Date >Optimality Theory is output-oriented: Markedness constraints penalize outputs Faithfulness constraints penalize input-output mappings >If the direction of optimization is reversed, this affects the application of markedness constraints (but not faithfulness constraints). Output-oriented
Date >Production: Meaning form Faithfulness constraints Markedness constraints on form >Comprehension: Form meaning Faithfulness constraints Markedness constraints on meaning >Because different constraints apply in the two directions of optimization, OT is direction- sensitive. Direction-sensitive
Date So there is evidence for early delays in production. Do we find similar delays in comprehension? Yes, if Chapman & Miller (1975) are right in that production precedes comprehension w.r.t. early word order. >The car is pulling the cow. Comprehension delay?
Date >Q: Does the adult constraint ranking always result in the same pairing of form and meaning in production and comprehension? >A: This depends on the constraints. Particular combinations of constraints give rise to a different pairing in production and comprehension. Example: Object pronouns (A)symmetry
Date >Principle A (FAITH): No reflexives with a locally disjoint meaning. >Referential Economy (MARK): No full NPs >> No pronouns >> No reflexives (Principle B need not be assumed, but rather is a derived effect) Pronouns
Date Input: coref. FAITH Principle A MARK Ref Econ reflexive pronoun*! Tableau 4 Production of coreferential meaning Input: disjoint FAITH Principle A MARK Ref Econ reflexive*! pronoun* Tableau 5 Production of disjoint meaning Production yields the adult forms: Production
Date Input: reflexive FAITH Principle A MARK Ref Econ coref. disjoint*! Tableau 6 Comprehension of reflexive Input: pronoun FAITH Principle A MARK Ref Econ coref. disjoint Tableau 7 Comprehension of pronoun But comprehension results in a non-adult pattern: Comprehension
Date This is exactly children’s pattern w.r.t. the Pronoun Interpretation Problem. >Q: But why aren’t pronouns ambiguous for adults? >A: Because adults optimize bidirectionally, whereas children are not yet able to do so. (de Hoop & Krämer, 2005/6; Hendriks & Spenader, 2005/6; Hendriks et al., Conflicts in interpretation) Ambiguity
Date Bidirectional optimization (Blutner, 2000): A form-meaning pair is bidirectionally optimal iff: a.there is no other bidirectionally optimal pair such that is more harmonic than. b.there is no other bidirectionally optimal pair such that is more harmonic than. Blutner (2000)
Date FAITH Principle A MARK Ref Econ * * * Tableau 8 Bidirectional optimization of anaphoric objects Principle B A symmetric system arises through bidirectional optimization: Bidirectional OT
Date Language acquisition in bidirectional OT: >Initial constraint ranking (presumably MARK >> FAITH) >Error-driven constraint reranking (e.g., Tesar & Smolensky, 1998; Boersma & Hayes, 2001) >Adult constraint ranking >From unidirectional to bidirectional optimization Language acquisition
Date How can we decide between biOT explanation and alternative accounts? >Alternative accounts predict that production in general is relatively easy. Example: Subject pronouns Bidirectional OT
Date Him Ladies and gentlemen, we got him! Paul Bremer at press conference in Baghdad, 14 Dec. 2003
Date Pronouns refer to very salient referents, usually mentioned in the linguistic discourse. >ProTop (FAITH): No pronouns that refer to a non-topic. Topic
Date MARK Ref Econ FAITH Pro Top * * * Tableau 9 Bidirectional optimization of anaphoric subjects The adult pattern can be modeled by bidirectional optimization: Recoverability
Date Input: +topic MARK Ref Econ FAITH Pro Top pronoun full NP*! Tableau 10 Production of topical referent Input: -topic MARK Ref Econ FAITH Pro Top pronoun* full NP*! Tableau 11 Production of non- topical referent Predictions with respect to production: Predictions
Date Input: pronoun MARK Ref Econ FAITH Pro Top +topic -topic*! Tableau 12 Comprehension of pronoun Input: full NP MARK Ref Econ FAITH Pro Top +topic -topic Tableau 13 Comprehension of full NP Predictions with respect to comprehension: Predictions
Date >If children are unable to optimize bidirectionally, it is predicted that: They overuse pronouns to refer to non- topics. They fail to interpret full NPs as marking a topic shift. >This was tested in a production/ comprehension experiment with 4- to 6-year- old Dutch children. (Wubs, Hendriks, Hoeks & Koster, to be presented at GALANA 3) Experiment
Date
A pirate is walking with a ball. He kicks away the ball. But then the ball falls into the water and he starts to cry. A knight arrives with a fishing net. He scoops the ball out of the water. And then the pirate has his ball back again.
Date % produced forms Children (4-6 y.o.)Adults Production of referring expression to refer to old topic after topic shift Results
Date Adults: >And then the pirate has his ball back again. Many children: >And then he has his ball back again. By using a non-recoverable pronoun, children as speakers do not take into account the hearer. This suggests lack of bidirectional optimization. Egocentric
Date Input: pronoun FAITH Principle A MARK Ref Econ FAITH Pro Top coref. & -topic *! disjoint & +topic Tableau 14 Comprehension of pronoun Prediction: Pronoun Interpretation Problem disappears if there is a clearly established topic. Another prediction PIP dissolves entirely in single topic context: “Here is an alligator. The elephant is hitting him” Spenader, Smits & Hendriks, in press
Date Young children (<4 years old) Older children (>4 years old) Delay in production First wordsAnaphoric subjects Delay in comprehension Early word order? Pronoun Interpretation Problem Asymmetries Bidirectional OT predicts four types of asymmetries:
Date How does bidirectional optimization develop? >Blutner & Zeevat (2004): Pragmatic reasoning about form-meaning pairs that can become conventionalized >Hendriks, van Rijn & Valkenier (2007): Online mechanism, dependent on processing resources: Form meaning form Meaning form meaning Development
Date Do processing resources matter? YES >Also overuse of subject pronouns by elderly adults (>60 years old). (Hendriks, Englert, Wubs & Hoeks, 2008) >Overuse of subject pronouns appears to be related to working memory capacity. (Wubs, Hendriks, Hoeks & Koster, to be presented at GALANA 3) >Children’s comprehension of object pronouns improves when speech is slowed down. (Van Rij-Tange, Hendriks, Spenader & Van Rijn, to be presented at GALANA 3 & BUCLD 33) Processing
Date Can the data also be explained by extra- grammatical factors? >Pragmatic knowledge: Separate explanation required for each phenomenon >Task factors: Methodological pessimism >Processing limitations: May account for late asymmetries, but weaker explanation Other explanations
Date Testing theories
Date Because OT is direction-sensitive, it allows for a straightforward explanation of production/comprehension asymmetries in language acquisition: >Early asymmetries can be explained as the result of a non-adult constraint ranking. >Late asymmetries can be explained as the result of the inability to optimize bidirectionally. Conclusions