Sounds and Words Flashcards
Define serial access
Searching through known words until you get the ‘matching’ word. Retrieve meaning of matching word
Define parellel/incremental access
Identifying intial letters/sounds. Look for partial match. When another letter comes in modify the short list.
Evidence for parallel access
Gating studies. Show the different stages of word recognition
Strong evidence for parallel access
Eye tracking (Dahan, 2001), pps played parts of word and shown photos, multiple words matching first letter are considered more than those that do not match. (Bell, bed, compared to Apple)
What are semantic competitors
Words that have a similar meaning are also considered in parallel access (Yee and Sedivy, 2006). Semantically related words are treated differently even before the listener fully commits to the target word.
What is semantic priming
Further evidence for parallel access. Priming studies use lexical decision tasks (words and non words). Words are chosen faster (especially if they are semantically related) compared to non words.
Outline other reasons why semantic priming may occur
Overlap at the conceptual level - similar to mediated semantic priming
Frequent co-occurrence - can help predict upcoming words
Outline cross-modal priming
Priming can also occur across modalities
Priming develops with incoming input (Different word forms AND their semantic relationship are considered from the beginning
“cap” primes both “money” (capital) and “ship” (captain)
What is form competition
Words with more neighbours are processed differently than words with few neighbours (“cat” has more neighbours than “scissors”)
In comprehension a larger phonological neighbourhood size often means slower target recognition. -> suggesting that other words related in form are considered too, and if there are more of those words, it influences word recognition.
Outline all evidence for parallel activation
Match with incoming signals
Semantically related words
Phonologically related words
Why are some words considered more during parallel activation
Because they are more frequent in everyday language. So that word is activated most strongly.
What are the three main ideas on how parallel activation works
Logogen
COHORT
TRACE
Outline the logogen model (Morton, 1969)
Each word has a ‘counter’ (logogen)
Each logogen has a baseline level
Activation is needed to pass a threshold
If the threshold is passed, the word is recognised.
Multiple logogens increase in activation -> this is in line with the parallel model
How does logogen interact with the interactive model
Both input and context can influence logogens and thus recognition
Describe how the baseline level of activation in the logogen model works
Bottom up input increases activation levels (this occurs in parallel)
If the input is ‘bench’ (the same as the target word) then the activation threshold will be reached
Words that “fire” more often get a lower threshold [this explains frequency effects]
What is the locus of frequency effects
Eye tracking data suggest gradual and very early effects of frequency suggesting frequency effects impact the baseline/ resting activation instead of a decrease in threshold.
More frequent words get a ‘head start’
Outline the COHORT model (Marslen-Wilson and Welsh, 1978)
Focus on spoken word recognition and some challenges specifically associated with spoken words
What are the three stages of the COHORT model
- Access stage - set up the word-initial cohort of possible words, based on initial portion of the word
- Selection stage - Eliminate cohort members as they mismatch the input (or don’t fit with context), at the uniqueness point of a word, only one member of the cohort remains
- Integration stage - semantic and syntactic information about the word are integrated with the sentence context
What are the features of the COHORT model
Simple
Optimal
Parallel activation
Issues with the COHORT model
COHORT model states that words are simply in or out of the cohort when in fact a more graded activation is possible
We can recognise words even if their onset is disrupted
Words that overlap in offset but not onset also compete for selection (e.g. rhyming words like “beaker” and “speaker”)
Outline the TRACE model
Deals much better with ambiguous or disrupted speech input
Interactive model, this makes it easier to implement effects of contexts
What are the features of the TRACE model
Connectionist -> different units of a word are connected
Localist -> words and sounds are represented as units
Lateral connections
Outline the context effects in TRACE
Hearing a contextually similar word can prime responses to the target word. A node is added to the model. However this could also occur in a different way, contextually similar word could prime a related contextual category (e.g. target word = cat, similar = dog, category = animals)
How does TRACE differ from COHORT
TRACE puts an emphasis on interaction between word and phoneme level in both directions.
COHORT has stronger focus on word onset.
