Chapter 8 Linear notes

Page 1

Speech recognition

pre lexical and lexical models of speech segmentation

hard to find word boundaries bc silences within words and coarticulation of phonemes

universalities of speech segmentation - non-word language sentences - infants followed - universals e.g. rhythm and stress

lexical advantage: uses stored knowledge of phonological representation

parallel activation: cohort models allow access to all candidates and process of elimination - time efficent

cross modal priming: semantic relations between spoken and written target and prime = aids access to meanings - better word recognition

limitation of parallel processing - too many meanings

TRACE: phonetic feature, phoneme and word - inhibition and activation - helps speech segmentation bc winner takes all - word that wins dictates boundary (confess - boundary after fess, fester - boundary before fess)

McQueen - duhmess vs nuhmess - find inner word easier if non word

Visual Word recognition:

IAC model: visual features, letters, words - bottom up and top down (weak sensory input can be helped by stored knowledge)

lexical effects e.g. WSE - non word harder to find syllable within bc no stored info of non words

but is top down needed? contentious

orthography and phonology links - assembled and addressed phonology - assembled good for simple words but addressed helps with unusual spellings

dual route DRC: rule based and lexical route - regular words named faster than irregular but only if low freq - regular can be named by both routes.. low freq - 2 routes confers advantage for reg words, high freq - doesn't matter

Neighbouring effects challenge DRC because similar spellings but irreg pronunciations - DRC can't explain as it applies same rules to reg and irreg - should there be single connectionist network?

Eye movement article: parafoveal region, occulomotor and lexical features, serial or parallel

eye tracking techniques useful for visual but not spoken recognition

Mental Lexicon

contains info re word sounds, spellings, meanings

morpheme - smallest word unit

morphology: inflectional and derivational

inflection more simple e.g. adding an s, derivational more complex e.g. changing grammatical class by adding suffix 'ness'

full listing and decompositional approach - extreme positions

spreading activation vs featural (properties)

semantic priming - assoc and non assoc

semantic ambiguity - autonomous and interactive view (bug context didn't matter so autonomous but there is implicit support for interactive re ERP rather than RT)

ERP: N400 peak re strength of fit - INTERACTIVE SUPPORT

Sentence Comprehension

hierarchical structure - explained by parsing - individual grammatical elements (syntax) - phrase structure rules - sentence creation and understanding

incremental or syntactic boundaries parsing? incremental - refine plausible syntactic structures when each word is recognised - incremental, maximises available info

appropriate vs non appropriate continuations named quicker - context bias for parsing ambiguous phrases (landing planes is, landing planes are - plus context)

Two models: garden path and constraint

Garden path - backtracking when wrong path chosen - horse race past barn fell - alternative version not accessed -- serial parsing

Constraint - parallel and interactive like the cohort model - potential parses evaluated simultaneously - eliminates autonomous stage - more than just syntactical features but sem plausibility and freq

is parsing autonomous? defendant examined - if reduced relative clause then garden path effect

word meaning before ambiguous verb affects gp effect --- only applicable for some contexts e.g. fewer plausible meanings

lexical freq influences parsing:

constraints: lexical freq - horse raced - rarely used as past participle - people can keep track of how words are usually used in sentences to combat against ambiguity

visual info: towel apple box napkin - external environmental features inform situation and reduce gp effect

COMPETITION AND INDEPENDENCE OF SYSTEMS