Understanding the psychological sources of communicative behavior.

4a8d62be623e1a3e3cde003a6a810c0b?s=47 mllewis
May 03, 2016

Understanding the psychological sources of communicative behavior.

4a8d62be623e1a3e3cde003a6a810c0b?s=128

mllewis

May 03, 2016
Tweet

Transcript

  1. Understanding the psychological sources of communicative behavior Molly Lewis Michael

    C. Frank Stanford University 40th Annual Meeting of the Society for Philosophy and Psychology 20 June 2014
  2. point to the dax What cognitive processes underlie this behavior?

  3. DAT A THEORY Pragmatic reasoning Context- independent knowledge A challenge,

    when there are multiple consistent theories. …
  4. Task: infer cognitive process from data Two challenging features of

    this problem: – Proposed theories sometimes make use of information on different timescales – Theories are often not necessary mutually exclusive with each other How to address this challenge ① ②
  5. Task: infer cognitive process from data Two challenging features of

    this problem: – Proposed theories sometimes make use of information on different timescales – Theories are often not necessary mutually exclusive with each other How to address this challenge ① ②
  6. Disambiguation Bias – Children tend to map a new word

    to an object they don’t yet know a name for (e.g. Markman & Wachtel, 1988) – Potentially, supported by information at two timescales
  7. . . . . . . time Pragmatic Support Context-Independent

    Support Cognitive supports at different timescales (McMurray, Horst, & Samuelson, 2012)
  8. Multiple cognitive supports underlying communicative behavior Distinguishable by the timescales

    of the information sources they rely on (McMurray, Horst, & Samuelson, 2012) Pragmatic support: knowledge of the speaker’s intended referent – information available in the moment of reference (“pragmatic timescale”) Context-independent support: knowledge of the lexicon – information acquired over years of speaking language (“lexical timescale”)
  9. Disambiguation Bias: Pragmatic Support Principle of Conventionality: Speakers within the

    same speech community use the same words to refer to the same objects. Principle of Contrast: Different linguistic forms refer to different meanings. 1. You used a word I’ve never heard before. 2. Presumably, we both call a ball “ball”. 3. If you’d meant to refer to the ball, you would have said “ball.” 4. You didn’t, thus, the “dax” must refer to the new object. (e.g. Clark, 1987; Diesendruck & Markson, 2001)
  10. Disambiguation Bias: Context-independent Support Knowledge of the lexicon: One-to-one structure

    between words and objects. w1 w w2 w3 o1 o2 o3 (e.g. Markman & Wachtel, 1988)
  11. Pragmatic support Context-Independent support These supports, at different timescales, make

    similar empirical predictions -- How can we tell them apart?
  12. Task: infer cognitive process from data Two features of this

    problem: – Proposed theories sometimes make use of information on different timescales – Theories are often not necessary mutually exclusive with each other How to address this challenge ① ②
  13. Pragmatic support Disentangling the supports Context-Independent support

  14. Attempts to disentangle supports Preissler and Carey (2005) – Test

    children with autism, who have impairments in pragmatic reasoning – Typically developing children ≈ children with autism, on disambiguation task – Evidence for lexical support? Diesendruck and Markson (2001) – Compare performance on a novel facts about an object relative to a novel referential label – Label condition ≈ fact condition – Evidence for pragmatic support?
  15. Disentangling the supports - Neither support is necessary for the

    disambiguation bias - But: limited because experimental context differs from typical referential contexts
  16. Supports not necessarily mutually exclusive with each other - In

    a causal system, if A causes B, does not imply that C does not also cause B. – “explaining away bias” - Pragmatic and context-independent knowledge could BOTH be supporting behavior. - The relative contributions of the supports could vary across people, contexts, and development.
  17. A general empirical schema – Our data are always collected

    in a social context. – Thus, behavior could be supported by information on different timescales. • Pragmatic (~“task demands”) • Context-independent – Because these supports are not mutually exclusive, difficult to tease apart empirically.
  18. Task: infer cognitive process from data Two challenging features of

    this problem: – Proposed theories sometimes make use of information on different timescales – Theories are often not necessary mutually exclusive with each other How to address this challenge ① ②
  19. Addressing this challenge Qualitative Predictions Quantitative Predictions “Is this inference

    observed under condition X?” “How strong is this inference in condition X?” The Proposal: Make quantitative predictions by using computational models that instantiate multiple levels of constraints.
  20. Example: Hierarchical Bayesian modeling - Representations at multiple levels of

    abstraction - Can simultaneously acquire knowledge at multiple levels - Captures inferences at both the pragmatic and lexical timescales - Used this approach to begin to model disambiguation bias (Frank, Goodman, & Tenenbaum, 2009; Lewis & Frank, 2013)
  21. W I Situations S Referential Intentions L Lexicon Words O

    Objects Lexical Constraint C ball Hierarchical Bayesian Model Lewis & Frank, 2013
  22. Lexical Constraints 1-many w1 w w2 w3 o1 o2 o3

    many-1 w1 w w2 w3 o1 o2 o3 object object inconsistent w1 w w2 w3 o1 o2 o3 1-1 w1 w w2 w3 o1 o2 o3
  23. Model Predictions Theories may only be distinguishable by observing change

    in the system (i.e. development) 0 1 2 3 4 5 6 7 8 9 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Proportion Correct # of obs. of known word mapping 1−1 1−many many−1 inconsistent
  24. Conclusion An empirical challenge: - Theories at multiple timescales consistent

    with behavior - Theories not mutually exclusive Need quantitative predictions by using computational models (e.g. hierarchical Bayesian models) W I Situations S Referential Intentions L Lexicon Words O Objects Lexical Constraint C
  25. Thank you.