Transcript Psycholinguistics & Cognitive Science:
Brain, Psycholinguistics, & Cognitive Science
Outline
• How does psycholinguistics fit within the umbrella of cognitive science?
• What do we know about language and the brain?
Inter-relationships
Cognitive Psychology Linguistics Psycholinguistics Cognitive Science Philosophy Computer Science/AI Natural Language Processing Cognitive Neuroscience Neurolinguistics
What do cognitive psychologists do?
Construct theories that describe how mental processing works in real time: – Within some domain, what are the mental representations and how are they manipulated?
– How does information flow through the mind? – How is processing impacted by memory constraints, stimulus quality, mode of input, tasks, etc?
Levels of processing Serial vs. parallel processing Top-down/bottom-up Automatic vs. strategic processing
What do
(formal theoretical)
linguists do?
• Construct formal theories of our linguistic knowledge – Sets of rules/principles/assumptions for generating utterances • Criteria for a good theory – The rules generate all and only grammatical outputs (intuitions) – In the simplest way
Derivational Theory of Complexity • • Miller’s clause-processing model of syntactic processing, motivated by Chomsky’s Transformational Grammar.
Example of taking a linguistic theory and trying to implement it directly as a psychological theory of processing.
Transformational Grammar
• Two levels of syntactic structure, related by transformational rules; accounts for similarity in meaning across different strings.
– Surface structure
((John) (picked __ (the box) up)).
– Transformations
Particle Movement
– Deep Structure
((John) (picked up (the box))).
• Phrase structure rules and lexical insertion rules used to construct DS
Diff SS, Same DS
Transformations can move, delete, or add words. Complex SS ’s require a sequence of transformations – Wh-movement & Subj/Aux Inversion SS:
Which book did Jim buy __?
SS
Jim did buy a book.
DS:
Jim did buy a/which book
– Passivization SS:
The beer was drunk by Jim
.
SS:
Jim drank the beer
.
DS:
Jim drank the beer
.
DTC
1.Determine the surface structure of the sentence
2.Reverse the transformations one by one
3.Recover the deep structure 4.Map DS to meaning
Processing Unit is entire Sentence • Wh-movement & Subj/Aux Inversion SS:
Which book did Jim buy __?
SS:
Jim did buy a book.
DS:
Jim did buy a/which book
• Passivization SS:
The beer was drunk by Jim
.
SS:
Jim drunk the beer
.
DS:
Jim drunk the beer
.
How would you test the DTC?
• Early evidence seemed to support it.
• Later evidence was problematic • No longer a viable theory of sentence comprehension.
Memory & Language
• Semantic Memory: LTM storehouse of conceptual knowledge –What is a cow? What is truth? • Lexicon: LTM storehouse of knowledge about words –Lexical Entry: For each word/morpheme, Spelling, pronunciation, syntactic category, pointer to semantic memory • Episodic Memory: LTM storehouse of our experiences
Spreading Activation Model
(Collins & Loftus, 1975) • Length codes typicality • Link codes type of relation • Some redundancy uses is is a
Commmon Current Assumptions about Semantic Memory & the Lexicon •
Semantic Memory
is a network of concepts, organized by semantic similarity •
Lexicon
is a network of words, organized by phonological similarity • Interconnections link meanings to words
Cognitive Neuroscience
• Cognitive psychologists tend to talk about the architecture of the mind in terms of functionality – E.g., what is the input to word recognition? What is the output?
• Ultimately, the mental operations described by cognitive psychology occur in the brain • In some cases, neuroscience can inform cognitive psychology – E.g., we may be able to learn about how words/concepts are represented by investigating activation patterns in response to different classes of words: action verbs activate motor cortex; perception verbs activate visual cortex
Brain Anatomy (& Language)
– Is language localized in the brain?
– Is language lateralized?
LH lobes
Frontal lobe Temporal lobe Parietal lobe Occipital lobe Cerebellum
Some Lg-relevant areas
Video Clip (15 min)
Brain Story: First among Equals – The first segment has been comparing human and chimp abilities to plan. They conclude that chimps can make and execute plans, but not as far into the future as humans.
– We’ll watch a segment on LG • Aphasia patient • Mapping the brain prior to surgery
What did you learn from the clip?
• What does it mean to have aphasia?
• Do aphasics recover language function? How?
• What is the current view on the role of Broca’s and Wernicke’s areas?
Schiff et al. (2005)
• Do minimally conscious patients process speech?
Yellow= forward; Blue = backward Red = both
Figure 1. Functional maps obtained during listening to narratives
Averaged healthy data
Schiff, N. D. et al. Neurology 2005;64:514-523
Figure 2. Volumes of activation during the passive listening tasks. The 2 patients are in blue and red; Averaged healthy activation in black.
Schiff, N. D. et al. Neurology 2005;64:514-523
Schiff et al.
– paper was published shortly after the Terry Schiavo media/political frenzy.
– How do we decide whether a minimally conscious patient is experiencing a life worth living?
• Is language comprehension relevant?
• Why don’t we care as much about how their brain responds to smells?
Speech & Spoken Word Recognition
Outline
• Why is speech perception difficult for computers to do?
– Problem of Invariance • How do humans do it so easily?
– Bottom-up information (acoustic signal) – Top-down information (higher level context)
Automatic Speech Recognition Follies (David Pogue, NYTimes, 8/15/02) phonemes form different words What I said bookmark it Motorolla modem port a procedure Different phonemes, at underlying and then stick it in the mail movie clips I might add Inscrutable the right or left What was transcribed book market motor roll a mode import upper seizure and dense thicket in the mail move eclipse I my dad in screw double the writer left
The Problem of Invariance
• Individual phonemes do not have invariant acoustic cues.
• There’s a lot of variability in the acoustic signal!
• Variance in the acoustic signal has many sources: – Coarticulation – Differences among speakers – Differences within speakers: yelling/whispering, phone/in-person, etc.
If the bottom-up acoustic signal doesn’t provide consistent cues, how do we recognize phonemes/words?
Reading a spectrogram formants Can you see invariant cues associated with /i/ ?
If we can’t/don’t rely solely on bottom-up input, how do we recognize speech?
• Perceive speech as (intended?) articulatory gestures, not as acoustic signal?
– McGurk Effect – Sine Wave Speech: There are no essential acoustic properties that enable speech perception. Rather second-order changes in frequency and amplitude over time are responsible. sine wave speech demos • Use top-down information (word and sentence context) to complement bottom-up information – If so, when and how?
Is Speech Special?
Modularity Thesis (Fodor, 1983) – The mind is not a unified whole. In addition to central processes, there are specialized input output modules • Central: decision making • Input: color perception, voice recognition • Output: throwing, touch-typing, articulation – Modules are fast, informationally-encapsulated, mandatory, exhibit characteristic breakdowns, and have shallow outputs.
– Speech perception may be handled by a specialized input module
The Essence of Motor Theory • Speech perception is grounded in our knowledge of speech production. We recognize phonemes by covertly re-creating the articulatory gestures. (Lieberman et al., 1967) • Consistent with philosophy that performance & perception are inextricably linked.
• Assume innate, encapsulated phonetic module
Spoken Word Recognition
• Overcoming the problem of invariance in speech perception – Motor Theory – Top-Down Feedback: Word to Phoneme • TRACE • Cohort theory of Spoken Word Recognition
Evidence for Top-Down influence on speech perception • Phoneme Restoration Effect (Warren, 1970) • Lexical bias in categorical perception task, e.g.
dype
vs.
type
(Clifton & Connine, 1987)
• • • • •
TRACE
(McClelland & Elman, 1986) Interactive connectionist model Nodes in network represent phonetic features, segments, & words Feature nodes activated by consistent input Activation spreads up through network & back down again Predicts top-down effects
Example: initial phoneme in “pick” is ambiguous betwn /b/ & /p/.
(Lexical Bias)
Is word recognition Automatic & Modular?
Automatic Processes – Fast – Do not require attention – Feed-forward (can’t be guided, controlled, or stopped midstream) – Not subject to top-down feedback (informational encapsulation)
Stroop Effect
Name font color
RED GREEN BLUE YELLOW GREEN
What happens if you have to name word?
Differences between spoken and written word recognition • For relatively short words, letters in a written word are processed in parallel – Eye movement data – Word superiority effect – Letter-Search Task • Spoken word unfolds across time – Can recognize some words before they are completely pronounced.
Research on the Lexicon (Outline) • How are lexical entries accessed? (Word Recognition) What is the input?
– Speaking (Ashcraft) – Reading – Listening • How is lexical ambiguity resolved?