Lecture 3. Models of Word Recognition and Production Flashcards
What is our lexicon? Where does it exist?
It is our own personal mental dictionary, therefore it exists in our mind
Phonology?
What is spoken
Orthography
What is written
What is an example of visuo-spatual/gestural communication?
Sign langauges
a lemma is a head representation of a word and a specific concept of a word is a _____ (meaning)
semantic
Why do we require parallel activation of lexical competitors?
On average, spoken words take approximately 500ms to unfold
Word recognition would be simple if lexical selection process began only after the word had finished
However, this would not enable enough time to process the meaning of speech
To solve this problem, speech is evaluated and reevaluated continuously against numerous potential lexical competitors.
What did Spivey and Dale do? Twice?
Visual and mouse tracking. first it displayed that our cognition is more dynamic than we assumed in classical conditioning and secondly, and similarly that competition between co-activated lexical candidates is resolved over time.
When participants saw pictures of a piece of candy and a candle on the computer screen and were instructed to “click the candy” the trajectory of their mouse movements tended to exhibit spatial attraction to both objects as the wording of both objects start out the same “ca…” and we refine our selection as more part of the word unfolds
What is an even more interesting aspect of parallel activation in the Marslen-Wilson experiment?
Hint: its also cross modal
When we heard fragment “capt…” it first primes hearing the full word and then primes associated written words, like “ship” and “guard”
What is an example of priming for more ferquently used targets when we hear the start of the word is “d…”
We would be more like to be phonologically or orthographically primed to hear or see the word dog than dock, because dog is more common
What is McClelland and Elman’s model?
The TRACE model
The TRACE model. Is top down, bottom up or both top down and bottom up?
And what is special about interplay within the word layer?
Both top down and bottom up
In addition to excitory connections between layers, there are also inhibitory connections between nodes within the word layer.
So, the TRACE model is both top down and bottom up, what are the actions that flow as “bottom up” activation and what are the actions that flow as “top down” activation?
Bottom up is activation is the excitation activation through the layers of the system from the phonetic feature level to phonemes to word level.
Top down is the flow back down through the system from the word level to the phonetic feature level. This is referred to as “top down” activation
What are the layers of the TRACE model?
i.e. what is the bottom, middle and top layer?
Bottom layer is the phonetic features
Middle layer is the phonemes
Top layer is the words
So, the TRACE model is a three layer model consisting of phonetic features, phonemes and words, and there is bottom up activation in which the phonetic features activate to form phonemes and those phonemes form words, and there is also top down activation in which activation flows back down from the word level to the phonetic feature level.
SO, what is the deal with excitation and inhibition ?
Excitation is when an active node can both raise the level of activation of nodes that are consistent with it
Inhibition is when an active node will lower the level of activation of nodes that are inconsistent with it.
Inhibitory connections (or lateral inhibitory connections) between nodes in the word layer enable the suppression of activation of competing nodes
What does lateral inhibition allow for?
Selection between competitors.
Parallel activation causes multiple lexical ‘candiadates’ to compete during word recognition.
in the TRACE model, word recognition is a process of gradually building activation (or evidence) for a set of lexical candidates in parallel over time
As activation accumulates in multiple nodes at the lexical level, the nodes compete via lateral inhibitory mechanisms
lexical nodes for higher frequency words have a lower activation threshold and so exert inhibition more quickly and strongly on competing low frequency words.
The winner is the lexical node with the strongest activation once all excitatory and inhibitory inputs are accounted for
What do higher frequency words have? A higher activation threshold or a lower activation threshold?
A lower activation threshold
In the TRACE model, when there is phonological overlap, what is going on at the word level? Is there is inhibitory activation? What does this mean?
It enables the lexical entry most consistent with the input to inhibit (suppress the activity) of its competitors
Binlinguals! What type of language system is going on with bilinguals, are the two lexicons of languages “independent” or “integrated”?
Can you then select to only use one language or do elements of the other language still compete? Thus, is bilingual language processing language “non selective” or language “selective”
Bilingual processing is language integrated and non selective.
So this means that FMRI studies show that the brain areas are used to represent both languages in bilinguals and bilinguals represent language in the same areas of the brain that monolinguals do
Van Heuven sounds Dutch right? Therefore, what was the test that van Heuven did, orthographically?
van Heuven showed that English words with a high amount of Dutch words neighbours had a slower response times than than words that had fewer orthographic neighbours in the first language
oh, look who it is again, Spivey! Marian and Spivey (2003) showed what between german and english words in yet another eye trace task?
That we directed our attention, or our eyes tracked tentatively towards pictures of objects that had similar pronunciation in both German and English
i.e. click on the “d…” made eyes track towards both the desk (english) and deckel (german for lid)
Marian and Spivey, in their bilingual eye tracking experiment showed that eye tracking interference was particularly strong for what?
lexical competitors in another language that had a particularly higher lexical frequency i.e. lower inhibitory level than the target
so, van Heuven showed that the response time of Dutch/English bilinguals were longer when responding to English words that had more Dutch neighbours than straight English speakers, what model did van Heuven contribute to construction from this?
The Bilingual Interactive Activation (BIA) Model
The BIA (bilingual interactive activation) model is top down and bottom up, yes?
Yes, it is an algorithmic model of bilingual word recognition that implements non-selective bottom-up processing (letters activate words from both languages in an integrated lexicon) and language-specific top-down processing (language nodes selectively inhibit activity in words of the other language).
By using this language-specific top-down control mechanism, the BIA model can handle both selective and non-selective results.
Green’s inhibitory Control Model of Bilingual Language Production is a what kind of system?
Hint: an SAS
Green’s system is a Supervisory Attentional System
The SAS (supervisory attentional system) activates language task schemas
What is the wole of language task schemas in Green’s Inhibitory Control of Bilingual Language Production?
Schemas enable selection of the intended language by inhibiting the activation of task-irrelevant language tags
Greater activation of what part of the brain indicated a bilingual advantage?
in the dlPFC and the ACC
Processing within the bilingual lexical system is believed to be:
A. Selective
B. Non - Selective
C. Integrated
D. Separate
Answer: B!
Language non selective is the term used to describe processing within the bilingual lexical system as it refers to the fact that there is evidence that words from both languages are activated automatically during language processing. i.e. non selectively, even when only one language is being used.
Language independent (separate) is not the correct answer because the question referred to processing within the system, rather than the structural organisation of the system. The language of bilinguals are integrated, rather than there being separate lexical systems for each language, but this is not the answer to the processing question.
In the domain of recognising spoken words, bilinguals have been shown to experience what?
I.e. even when one language is being spoken to them, what languages are activated as they listen?
Activation of a cohort words from both their known languages as the sound of a target word unfolds over time, even when the linguistic input is purely in one of their languages.
Which of the following research findings provides support for the Bilingual Interactive Activation (BIA) Model?
a. Bilinguals show increased effects of incongruent stimuli when performing non-linguistic conflict tasks such as the Erikson Flanker Task.
b. Bilinguals show increased effects of incongruent stimuli when performing non-linguistic conflict tasks such as the Erikson Flanker Task.
c. Bilinguals respond more quickly to written words presented in language A if those words are spelled similarly to many words in language B.
d. Bilinguals respond more slowly to written words presented in language A if those words are spelled similarly to many words in language B.
Option D is the correct answer. The structural assumptions of the BIA model include an integrated lexicon for words from both languages, and the assumption that the written input activates words from both languages that share letters with the input non-selectively. Competition between languages is resolved when input from activated lexical nodes to the language nodes results in inhibition of words from the non-target language.
The co-articulation of phonemes during speech production often results in ambiguous input for the listener. According to the TRACE model of spoken word recognition, the ambiguity caused by the co-articulation of phonemes can be resolved by the mechanism of:
a. Parallel activation
b. Lateral inhibition
c. Lexical feedback
d. Segmentation of speech system
The correct answer is C. The coarticulation of phonemes during speech results in ambiguous input to the phoneme recognition system. Feedback from the lexical level to the phoneme level (lexical feedback) provides a means to clarify or ‘clean up’ the ‘noisy’ (imperfect) information at the phoneme level based on knowledge stored at the lexical level about the phonetic structure of whole words. Option A (parallel activation) is wrong because this feature of the model relates to the flow of information through the system rather than to mechanisms for resolving ambiguity. Option B (lateral inhibition) is not the best answer as this mechanism, as discussed in the lecture, operates at the lexical level to resolve competition between co-activated word nodes, rather than acting directly to resolve ambiguity at the phoneme level (although this is not to say that lateral inhibitory processes are not indirectly involved in this process). Option D is wrong as segmentation of the speech stream is the problem to be solved by the mechanisms proposed in the model, rather than a mechanism per se.
In the TRACE model, what level does lateral inhibition operate on?
lateral inhibition operates on the lexical level, to resolve competition between co-activated word nodes
van Heuven came up with the Bilingual Interactive Activation (BIA) model for bilingual word recognition.
- Is this an algorithmic model?
This model of bilingual word recognition implements non selective ___2.___ processing (letters activate words in an integrated lexicon) and language specific ___3.___ processing
- Yep
- bottom up processing
- top down processing
When a string of letters is presented to the BIA model, this visual input affects particular features at each letter position, which subsequently excite letters that contain these features and at the same time inhibit letters for which the features are absent. The activated letters next excite words in both languages in which the activated letter occurs at the position in question, while all other words are inhibited. At the word level, all words inhibit each other, irrespective of the language to which they belong. Activated word nodes from the same language send activation on to the corresponding language node, while activated language nodes send inhibitory feedback to all word nodes in the other language. The language nodes collect activation from words in the language they represent and inhibit active words of the other language. The activation of the language nodes reflects the amount of activity in each lexicon.
What are the layers of the bilingual Interactive Activation model?
Language
Words
Letter
Feature
In the TRACE model, when we hear similar words, what is the ordering of lexical selection, parallel activation and lateral inhibition
first parallel activation i.e. hearing “beetle” at first accrues both beetle and beaker nodes
then lateral inhibition, as we hear more of the word, the other word is suppressed
then lexical selection
