origins of modularity

49
Origins of Modularity

Upload: hewitt

Post on 06-Jan-2016

38 views

Category:

Documents


1 download

DESCRIPTION

Origins of Modularity. Preliminary Binding Study. Results 13 participants tested using Linger Data can be analyzed using Lingalyzer Lite Region encoding needs to be added ( perl script, Excel ?) Short-term objective: means & standard errors, accuracy Longer-term objective - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Origins of Modularity

Origins of Modularity

Page 2: Origins of Modularity

Preliminary Binding Study

• Results

– 13 participants tested using Linger

– Data can be analyzed using Lingalyzer Lite

– Region encoding needs to be added (perl script, Excel?)

– Short-term objective: means & standard errors, accuracy

• Longer-term objective

– Fine-tune experimental items

– Test with naïve participants

– ANOVA analyses

Page 3: Origins of Modularity

Regions

Page 4: Origins of Modularity

Statistical Analyses

• Raw vs. residual reading times

• ANOVA– MS = s2 = SS/df

– F-ratio = MSbetween/MSwithin

– Interaction in 2-way ANOVA: between group variance that is not accounted for by main effects

• Repeated Measures ANOVA– Intuitively: are all subjects/items affected similarly by the experimental

factor?

– Statistically: F = MSrepeated-measure/MSinteraction

– Power of subjects analysis depends on number of items, and vice versa

Page 5: Origins of Modularity
Page 6: Origins of Modularity

• Effects of informational encapsulation due to

– Architectural constraints

– Computational constraints

Page 7: Origins of Modularity

(Trueswell et al. 1999)

Page 8: Origins of Modularity

Summary from Last Week

• ‘Autonomy’ in Generation vs. Selection

– Most evidence does not specifically identify effect of semantics/pragmatics on structure generation

– What would needed for semantics to impact generation• In order to block generation of one analysis, semantic information

must have its effect prior to the ambiguous input

• E.g., anomaly of ‘the frog’ in multiple frog context triggers search for restrictive modifier

• Requires a link from failure to resolve definite description to generation of specific structure

• Predicts limited effects on generation, at best– Multiple referent contexts

– NOT plausible combinations, e.g. evidence examined, warehouse fires

Page 9: Origins of Modularity

(Tanenhaus et al., 1995)

Page 10: Origins of Modularity

• Implication

– “Our results demonstrate that in natural contexts, people seek to establish reference with respect to their behavioral goals during the earliest moments of linguistic processing. Moreover, referentially relevant non-linguistic information immediately affects the manner in which the linguistic input is initially structured. Given these results, approaches to language comprehension that assign a central role to encapsulated linguistic subsystems are unlikely to prove fruitful.” (p. 1634)

– Do these results imply strong interactivity?

(Tanenhaus et al., 1995)

Page 11: Origins of Modularity

How Representative are VW Studies?

• Strong context effects occur in narrowly constrained situations, where

– Prior material generates expectation for one variant of an ambiguity, with sufficient specificity to predict individual words

– Context is constrained enough such that

• Specific expectation can be elicited (i.e. notice ‘Oops, there are two apples in this scene, which one can he be referring to?’)

• There is a good understanding of what are the relevant objects in the visual context

– “… visual stimuli can combine with linguistic stimuli in a manner that effectively constrains the referential domain in ways that may be unrepresentative of the challenges that readers and listeners face when dealing language about displaced objects and events” (Pickering, McElree, & Garrod, 2004)

Page 12: Origins of Modularity

• “Second, the effects of the objects on processing an utterance may be exacerbated by the small number of objects presented (typically about four). Presumably, small sets are necessary because larger sets would increase memory demands (both for the retention of the objects and their locations) and would dilute effects. However, small sets enable particular strategies that may be optimal for the experimental task but not necessarily representative of general operations involved in processing language. For instance, participants may circumvent standard processing operations by developing strategies based on a limited number of representations held in working memory. Immediate effects that depend on properties of the paradigm cannot be used to provide strong support for interactive theories of comprehension.” (Pickering, McElree, & Garrod, 2004)

Page 13: Origins of Modularity

(Trueswell et al. 1999)

Page 14: Origins of Modularity

(Trueswell et al. 1999)

Page 15: Origins of Modularity

(Trueswell et al. 1999)

Page 16: Origins of Modularity

QuickTime™ and aVideo decompressor

are needed to see this picture.

QuickTime™ and aVideo decompressor

are needed to see this picture.

“Put the hippo on the towel in the basket”

“Put the bear on the plate in the box”

Adult

(Trueswell et al. 1999)

Page 17: Origins of Modularity

QuickTime™ and aVideo decompressor

are needed to see this picture.

QuickTime™ and aVideo decompressor

are needed to see this picture.

“Put the frog on the napkin in the pot”

“Put the hippo on the towel in the basket”

5-year Old

(Trueswell et al. 1999)

Page 18: Origins of Modularity

Kindergarten Path Effect

• Why are the children failing to act as adults do on the 2-referent context?

– Overall VP-attachment bias, stronger than adults (due to minimal attachment, or to statistical bias of put for a location argument)

– Unable to take advantage of referential context

– Insensitive to definiteness

– Unable to revise initial commitments/plans of action

Page 19: Origins of Modularity

One Answer

• Compare put and wiggle (2-referent contexts only, n=16)

– Put the frog (that’s) on the napkin in the box.

• Actions Ambig 55% Unambig 16%

• Looks (n=8) Ambig 71% Unambig 44%

– Wiggle the frog (that’s) on the napkin in the box.

• Actions Ambig 23% Unambig 8%

• Looks (n=8) Ambig 52% Unambig 38%

(Trueswell et al. 1999)

Page 20: Origins of Modularity

Another Answer

Modifier-biasChoose the cow with the stick.

Equi-biasFeel the frog with the feather.

Instrument-biasTickle the pig with the fan.

(Snedeker & Trueswell, in press)

Page 21: Origins of Modularity

Another Answer

Modifier-biasChoose the cow with the stick.

Equi-biasFeel the frog with the feather.

Instrument-biasTickle the pig with the fan.

(Snedeker & Trueswell, in press)

Page 22: Origins of Modularity

Memory

Page 23: Origins of Modularity

The defendant examined by the lawyer…The evidence examined by the lawyer…

(Just & Carpenter 1992)

Page 24: Origins of Modularity

The defendant examined by the lawyer…The evidence examined by the lawyer…

(Just & Carpenter 1992)

Page 25: Origins of Modularity

The defendant examined by the lawyer…The evidence examined by the lawyer…

(Just & Carpenter 1992)

Page 26: Origins of Modularity

The soup cooked in the pot but was not ready to eat.The soup bubbled in the pot but was not ready to eat.

(Pearlmutter & MacDonald, 1995)

Page 27: Origins of Modularity

The soup cooked in the pot but was not ready to eat.The soup bubbled in the pot but was not ready to eat.

(Pearlmutter & MacDonald, 1995)

Page 28: Origins of Modularity

The soup cooked in the pot but was not ready to eat.The soup bubbled in the pot but was not ready to eat.

(Pearlmutter & MacDonald, 1995)

Page 29: Origins of Modularity

Processing Overload

• RC/RC center embedding

– The school board which the teachers who were neglecting the students had angered troubled the superintendent.

Page 30: Origins of Modularity

A Contrast (Gibson 1998)

• Relative Clause within a Sentential Complement (RC SC):

The fact [CP that the employee [RC who the manager hired] stole office supplies] worried the executive.

• Sentential Complement within a Relative Clause (SC RC):

#The executive [RC who the fact [CP that the employee stole office supplies] worried] hired the manager.

RC SC is easier to process than SC RC

Page 31: Origins of Modularity

A Contrast (Gibson 1998)

• Relative Clause within a Sentential Complement (RC SC):

[SC that the employee [RC who the manager hired] stole

• Sentential Complement within a Relative Clause (SC RC):

[RC who the fact [SC that the employee stole office supplies] worried]

RC SC is easier to process than SC RC

Page 32: Origins of Modularity

A Contrast (Gibson 1998)

• Relative Clause within a Sentential Complement (RC SC):

[SC that the employee [RC who the manager hired] stole

• Sentential Complement within a Relative Clause (SC RC):

[RC who the fact [SC that the employee stole office supplies] worried]

RC SC is easier to process than SC RC

Page 33: Origins of Modularity

Complexity Measures

• Memory cost associated with

– Incomplete syntactic dependencies (prediction, memory cost)

• Number

• Length

– Completion of syntactic dependencies (integration cost)

• Number

• Length

(Gibson 1998)

Page 34: Origins of Modularity

• Role in ambiguity resolution

– Minimization of memory costs contribute to choice among alternative resolutions of an ambiguous structure

– Prediction: when difference in memory costs is large, this factor overrides other factors

Page 35: Origins of Modularity

Plausibility

• Grodner et al. 2002

The evidence examined by the lawyer turned out to be unreliable.The witness who the evidence examined by the lawyer implicated seemed to be very nervous.The witness thought that the evidence examined by the lawyer implicated his next-door neighbor.

Page 36: Origins of Modularity

Plausibility

• Grodner et al. 2002

The evidence examined by the lawyer turned out to be unreliable.The witness who the evidence examined by the lawyer implicated seemed to be very nervous.The witness thought that the evidence examined by the lawyer implicated his next-door neighbor.

Large syntactic difference overrides plausibility bias

Page 37: Origins of Modularity

(Grodner, Gibson & Tunstall, 2002)

Page 38: Origins of Modularity

• Experiment 2: N-N compound vs. RC

– PlausibleThe alley (which) mice run rampant in is damp and dimly lit but relatively clean.

– ImplausibleThe tool (which) plumbers need to have is a good monkey wrench for loosening rusty pipes.

(Grodner, Gibson & Tunstall, 2002)

Page 39: Origins of Modularity

(Grodner, Gibson & Tunstall, 2002)

Page 40: Origins of Modularity

Plausibility

• Eastwick & Phillips, 2000

The judge remembered that the document…

stating that the [ defendant/evidence (that was) examined by the lawyer was unreliable ]

…had been stolen from the filing cabinet.

Page 41: Origins of Modularity
Page 42: Origins of Modularity
Page 43: Origins of Modularity
Page 44: Origins of Modularity
Page 45: Origins of Modularity

Plausibility

• Eastwick & Phillips, 2000

The judge remembered that the document…

had stated that the [ defendant/evidence (that was) examined by the lawyer was unreliable]

…and should be withdrawn from the testimony

Page 46: Origins of Modularity
Page 47: Origins of Modularity
Page 48: Origins of Modularity

Architecture Again

Page 49: Origins of Modularity

Three Architectural Issues

• Abstraction (fall)– Relation between language and conceptual/perceptual systems

– Abstraction (in phonology, lexicon, syntax, etc.) allows for powerful & efficient combinatorial systems

– …but exacts a cost for recognition and learning

• Incrementality (winter)– Relation between systems for understanding, speaking, and

acceptability judgments

– Computation is fast, but how is this achieved?

• Modularity (spring)– Relation between computations at different levels of representation

– Various possible sources for constraints on interaction