Lecture 6: Lexicalization Flashcards
(18 cards)
Stroop task
- important in understanding language production and other things
- > limiting because only colours
Distractor Effects
-picture word interferrance
-understand how production works
-timing
Glaser & Düngelhoff, 1984
• Pictures are displayed with visually or auditorily presented distractor words.
• Semantically related distractor words slow down picture naming relative to unrelated distractor
- semantic relationship (pig and hourse) impacts lexical selection, activation same type, deep but narrow, slower RT, competition effect
- compare horse in other instance ewith unrelated word, wide spread activation so faster RT
- eg looking for tall freiend in bar, easy to find but if other tall people harder because activation for them as well
• Phonologically related distractor words speed picture naming relative to unrelated distractor
- faster impact, phonology happens at later stage
- phonological relationship (horn), makes faster,
• Dissociation between effects supports separate stages
- meaning of word is separate from sound
- phonological only comes online after selected word (Lemarks model)
Discrete or Continuous?
• The output of lexical selection serves as the input to phonological encoding.
– Seriality: The selected lemma activates its
lexeme. (nothing should happen below line before above)
– Continuous transmission: Non-selected lemmas
also activate their lexemes.
• controversial point for many years.
• Temporal separation between semantic and phonological effects
Discrete->slower in processing but make less mistakes
Continuous->faster but maybe errors
Picture Word Interference to test Discreteness
-stimulus onset asynchrony (SOA): same time, picture then word or word then picture (change so interact at different times)
• By manipulating SOA, we can see the time course of
production.
• If word first and picture later, Semantic interference
large (early SOA).
• If picture first and word later, semantic interference
gone (late SOA)
• Phonological effects large at 0 and late SOA, gone at
early SOA
• Supports Discreteness
-semantic words distract early in semantic processing so if comes late no effect, if phonological similar word comes at same time no effect as well
-if manipulate SOA get temporal separation
-+SOA pic then word, -SOA word then pic
-if word first, semantic effect
-if word comes later, optimal point to distract from phonological info
-if semantic word comes too late then no effect, only effect if comes early (graph in notes!!)
Based on results of experiment Discrete or continuous?
■ The output of lexical selection serves as the input to phonological encoding.
❑ Seriality: The selected lemma activates its lexeme.
❑ Continuous transmission: Non-selected lemmas also activate their lexemes.
! controversial point for many years.
■ Evidence of phonological co-activation = the activation of wordforms of words you don’t want to say
Activation of non-target Lexemes?
-how far does activation of things you don’t want to say activated (Levelt says isn’t at phonological level) couldn’t find for years but evidences of phological co-activation (activation of words in phonological level)
Morsella & Miozzo, 2002
■ Picture-Picture interference: Two pictures are shown, one to be named (better task maybe b/c don’t have word effect b/c word processed faster, pictures superimposed on each other)
■ When two pictures which are phonological related are presented, phonological facilitation is observed, although the non-target pictures do not need to be processed. (faster because getting activation from two places)
■ Suggests continuous spread of activation from conceptual level through lexicon, down to the lexeme level.
-semantic competition get activation on two different representations
-phonological facilitation have two different representations that spreads activation to the same representation
LExical Selection Mechanism
■ The existence of word and sound exchanges also demonstrates that more words and sounds are active than are needed.
❑ Three aces ⇒ Three kings
■ How do we eventually SELECT the word we want?
❑ Competitive selection - tug of war
❑ Non-competitive selection - horse race
■ This debate continues
Phonological Encoding Definition
-after lexical selection
-The process of retrieving the sounds and preparing them for articulation
-Whether your model is discrete or cascading, the hard work of preparing to articulate only starts after a lemma has been selected.
-While non-target lexemes may become activated, they do not get fully processed.
-We’ve selected a lemma, we now want to
retrieve the associated phonological (sound) information.
What might that process look like? What
might we be retrieving?
What do our speech errors tell us?
What do our RT experiments tell us?
- spoonerisms
- swapping of sounds
- beginning of words swap with other beginnings of words and ends of words swap with other ends (position within words is important for where exists in mind)
- syntactic category constraints doesn’t effect swapping (information irrelevant at this point)
- phonological timing effect (from picture interference, can get phonological priming effect from words that partially over lap (lets us about constituent parts of words and share information in meaningful representation way)
Phonological Encoding Idea 1
-idea 1: lemmas point towards the articulatory blue print for saying ‘cat’ (single non analyzable part, just tells how to move to produce word)
-We know that words that share sounds must have some overlapping or related representation. (e.g., bottle – ball exp)
-We know that sounds can swap in errors.
-☞ some intermediate representation of sound segments needed.
– Words broken down into parts
*seems inconsistent with facilitation and error effects
*can’t be right
Phonological Encoding Idea 2
-Idea 2: We need to assemble the sounds for the words we want to say. (eg Ikea bed)
– We retrieve the sounds but they aren’t properly structured
-phonological priming evidence pieces
! Why is assembly needed?? What are the
parts that we retrieve?
! Think again about the pattern of speech
errors we discussed
Lexeme
How assembly conducted?
How much is stored and how much is assembled?
How assembly conducted?
How much is stored and how much is assembled?
-To account for speech error patterns, many researchers have proposed that lexemes include segmental, syllabic, and metrical information.
-These are then assembled.
-One assembly process is ‘segment to frame association’
->segmental, syllabic, metric (what syllables stressed), stored then assemble together (segmental-to-frame association)
-retrieve info, map information from the word
-need to know syllable? but also need onset?
segment to frame association
- proposed by Dell (diagram in notes!!) eg
- retrieve unordered, unstructured segmental information (C, T, A )
- also retrieve a syllabic frame (consonant front, vowel middle, consonant at end)
- retrieve them then have to map them
- map sounds within position in word
- need to know onset, and coda ect
- if store information, retrieve, map
- gives us error pattern
According to this view Lexemes:
-Pre-syllabified lexemes explain why sound errors respect syllabic positions.
– Errors arise during segment-to-frame association.
– Positions are respected because they are stored with segments.
What other evidence do we have for
frame-based approaches?
Priming phonological syllable
- Several studies tried to facilitate lexeme retrieval by presenting whole syllables or syllable parts.
- If syllable frame stored, you should be able to speed retrieval using priming
- > segments are same (bacon and banker with baby), 2 segments
- > diff in syllables (bacon syllable match, banker syllable mismatch)
- > hypothesis get priming effect from bacon because syllable
- Several studies tried to facilitate lexeme retrieval by presenting whole syllables or syllable parts
- > also balsam (3 seg) and bald with ball
Results:
-Phonologically related distractor words facilitated
picture naming.
-no added bonus of sharing syllable has priming effect, no added effect (no evidence syllable being stored)
-more segments had effect
-must build syllable not stored
->syllable boundaries sensitive to context in connective speech , eg s, ing, changes syllable (constantly re-syllabify so doesn’t make sense to store ( Levelt argued against pre-syllabified lexemes due to re-syllabification.)
->Syllable boundaries are sensitive to
context.
– HORSE (1 syllable) vs. HOR.SES (2 syllables)
– OWNS vs. OWN.SIT
(look over notes unsure about this)
Idea 3 for phonological encoding
Given
– No evidence for syllable priming
– Resyllabification
Idea 3: Levelt argued that lexemes only store ordered set of segments, which have to be assembled into syllables.
-minimal storage hypothesis, segments and construct something meaningful
Prosodification
-After (all) the lexemes are retrieved, the segments are assembled into units appropriate for articulation, syllables.
– Unstructured segments not valid input to
articulators
-Prosodification is the process by which segments are assigned to syllables and syllable positions.
->have to do work for articulators so have appropriate info to work with
-Important: process unfolds in one direction (beginning to end
->middle of word can get distracted, always know how starts, try to put together from start, never end
Incremental encoding
-We store ordered set of segments and build syllables.
-The wordform is encoded, or prosodified,
incrementally from start to finish.
-The beginning of the word is ready before the
end of the word.
Preparation Effect (experiment) -Prosodification
-Preparation effect: word onset
-Preparation task: implicit priming task
-learn associations between words, cue and from cue have to produce a word
-word producing will either be related or not related
-if word you have to say is phonologically related, sharing beginning of word (loner, local, lotus)
-train initially on associations then cues
loner beacon major
local beadle maker
lotus beaker maple
-you know no matter what going to say going to start with ‘lo’ can prepare
-homogenous and hetergenous blocks
-saying same words just in different order
-how long does it take you to say loner in contect of other words that begin with ‘lo’ and how long it takes you to say loner in context of words that have other word beginnings
Preparation effect: word offset
-can also do same thing with the end of word
salto feable lava
veto stable nova
photo rubble diva
-if prosodification process can happen from both ends then you’ll find the preparation effect in end syllables as well
Results:
- when word beginning is repeated get priming/ preparation effect
- when word ending, don’t get any preparation effect, no speeding up of processing
- evidence that prosodification process only occurs incrementally from the start of the word
- when start speaking don’t have to have all word planned
Lexeme summary based on different ideas
-Lexeme is ordered set of segments, not syllabified.
-Segments are assembled, incrementally,
into phonological syllables.
-Start at the beginning
- No RT evidence for stored syllables. Only
assembled syllables
