Finding words Lecture Flashcards
Briefly describe the concept of a semantic network
Meaning is represented in nodes within an interconnected network, activating one activates surrounding, connected nodes automatically.
Describe a study which investigates the automatic activation of this network
Lexical decision task (word/ non word)- participants trained to expect building related words (ie door) after body related primes (ie body). Therefore there is explicit knowledge that body parts are not to be expected after these. Despite this, after seeing the prime BODY, “arm” evokes a quicker response than “door.”
Evidence for automatic spreading activation
What possibilities are there for which nodes are connected?
Based on association- words occur together
Overlap in meaning- share sematic traits (Tea, coke; ket, coke)
How did Rhodes and Donaldson investigate this association vs meaning of the semantic network?
It was also known that the N400 amplitude indicates the contextual fit of a word within a sentence at the meaning level (peaks at a non-fitting word.)
They used association word pairs; such as traffic-jam,
compound words in which the second word is unrelated to the first outside of the word,
association plus semantic word pairs; brother-sister, which often co-occur
Semantic word pairs; cereal- bread, which are semantically related but rarely co-occur
and unrelated word pairs; beard-tower.
The biggest N400 was expected at the unrelated condition, the least in the association + semantic, and the other two conditions could be compared
What were the results of the investigation of the association vs meaning of the semantic network by Rhodes and Donaldson?
Semantic word pairs did not actually lead to smaller N400s, only the association and association + semantic word pairs. There was no real difference in effect between these, suggesting that it was mostly an effect of association.
Lexicon contains 45,000 to 250,000 words, describe a very bottom up model which attempts to explain how we can retrieve meaning from these words as they are spoken
Logogen model
input from sensory stimulus activates logogen. If the threshold is reached, based on frequency, recency, context etc, then the information in that logogen is integrated into the sentence.
What are the limitations of the logogen model?
Its a strictly bottom-up model and its winner takes all- if the threshold is reached then the information in that logogen is activated and integrated. This suggests that none of the other “logogens” will be activated which is not really in line with what we observe in that there are usually lots of words that are at least partially activated beside each other.
name an interactive lexical retrieval mode, what makes it an interactive model?
TRACE model
It is not sequential, also allows recurrent processing- therefore not only bottom up, but top down processing can be integrated (therefore an interactive model).
How does the TRACE model explain the word superiority effect? Does this explain it better or worse than the Logogen model?
Traditional models like the logogen model just can’t explain it
Refers to the finding that individual letters are easier to process (to remember, detect etc) if they are part of a word (DRAG rather than &&D& or nonsense word GADR). Typically the stimulus is shown quite fast and the participants are asked whether they saw a certain letter.
Describe the trace model using this superiority effect
If a letter is presented on its own, it will only activate that letter but doesn’t travel further into the representational network. In It causes cascaded activation. GADR would activate the different letters but does not travel further into the representational network. But if we look at the word DRAG which actually has representation in this network level, then by activating the individual letters, activation can travel through the network and actually travel back which reinforces the activation of the individual letters.
Describe an influential lexical retrieval model which is often contrasted with the TRACE model
The Cohort model is specific to spoken language
It has three levels
Access- Initial activation of word candidates based on phonetic match with the first part of the acoustic input
Selection- one of the word-candidates is selected based on the rest of the acoustic input and context (uniqueness point, b- br- bro)
Integration- Meaning of the selected words is activated and integrated within the preceding message
Is the cohort model an interactive or sequential model? Why?
It is a sequential model- it first deals with the bottom up input and only later adds high level knowledge to the lexical retrieval process.
Why is the cohort model also described as a modular model? Contrast this against the interactive model
Lexical retrieval is initially completed based on bottom-up information (Access), and only subsequently integrated into the broader context sequence (late selection)
Interactive (TRACE model) suggests that all information, including contextual information is immediately used in lexical retrieval
What are the three possible points at which context has an effect?
> Prediction: Prior to the onset of the critical word
> During lexical access: While processing acoustic or visual input (cohort model)
> During integration: after you have already recognised the word
Describe an early 1977 study which demonstrates the role of context in lexical retrieval
Used an effect known as phoneme restauration: One phoneme in a word is replaced by meaningless noise or silence. People only “hear” the missing phoneme if there is meaningless noise (context is sufficiently restrictive).
Warren and warren used this and did a study where they gave sentences as stimuli that were identical except for the last word: It was found that the *eel was on the orange/ axle/ shoe. People ‘hear’ the word that fits the end of the sentence i.e substitute the phoneme based on context.
What conclusions can be drawn from this study on lexical access and context?
Context had an effect during the integration stage; wasn’t during prediction or when processing the input
Summarise the position of Cohort and Logogen models on context in lexical access
The first activation of lexical candidates is completely driven by the visual or auditory signal. No predictions based on the context (top-down)
Initially, only bottom-up perception, followed by top-down integration. So only then can context play a role
Describe a study which investigated when context takes place in lexical retrieval
They used cross-modal fragment priming
The participants would be given a sentence which gives context so that the last word is subject to prediction (with heavy hearts the mean surrounded the grave. They mourned the loss of their … (must be a previously living entity.)
They then heard a word fragment (kap/ kapit- could be capital (unlikely) or captain (likely) in dutch) or they didn’t hear one. Then the sentence would be interrupted and the participants were given visual targets to respond to either related to a word in line with the context or not but both in line with a word with started with the fragment(ship/ money).
RT was then measured to see which words people were more primed towards.
What were the results of the study which investigated cross-modal fragment priming? What conclusions can be drawn?
When there was no word fragment there were no differences in RT, when there was only nap fragment there was no difference.
However there was a 60ms difference when kaput was presented.
Therefore only when quite a lot of audial information is presented, then there is an effect of context. This suggests that context effects occur during the processing of the words, definitely not before, and a bit later in the processing as the effect is not seen with the smaller fragment.
Describe a study which gives evidence for the effect of context at a different stage of processing
Used homophones: same sound, different word (wood, would); and sentences supporting closed class word (words central to a language and unlikely to change e.g pronouns, would): “John said he didn’t do the job but his brother…”, and sentences supporting open class words (generally nouns and vocal that can be added or removed from a language without it changing in a meaningful way, wood): John said he didn’t want to do the job with his brother’s…
If you present a little bit of the word, There is priming for a target word such as timber, only in the open class condition. It is observed very early, seemingly in the access stage.
Describe a study which used eye tracking to determine context effects
A dutch study which questioned whether a clearly predictive context gave rise to lexical predictions. They used a constraining condition (Never climbed a goat so high) and a neutral condition (never has a goat so high climbed).
A display is presented where there is a fixation point and four visual stimulus surrounding the point. One is a contextually appropriate target (goat), a cohort competitor-phonetic similarity (goal), semantic competitor (spider, can climb), and an unrelated distractor (island). Eye movements and fixations are recorded.
What were the expectations and results of the study which used eye tracking to determine context effects?
In the neutral condition it was hypothesised that when hearing “never has a goat so high climbed” that eye movements are random before beginning of goat word, then eye gaze will start to shift at the the /g/ to either the target word or cohort competitor. However when the /t/ sound was made, the cohort competitor (goal) would be out of the question and gaze would resume on the target word. This was found to be the case.
In the constraining condition:
Prediction context effects predict that in the sentence “never climbed a goat so high”, eye movements would shift towards the target word (goat) or the semantic competitor (spider) after the word ‘climbed’ but before ‘goat’, however later context effects predict nothing happens here. Access context effects would predict a shift in attention at the beginning of the /g/ sound, when there is acoustic target information to start the retrieval process. No context effects would predict both Neutral and Constraining context have the same effect during initial stages of semantic retrieval.
There was no evidence found for prediction effects- people were not looking more at goat or spider before onset of the target word. However gaze was directed at goat from the onset of /g/- suggesting that very early on during the very first access stage people immediately integrate the context
These have been three studies providing evidence to context effects during the access stage. Describe a study which found evidence for predictive context effects in certain instances
In languages like dutch and German (and French) there are gendered (I can’t even say it I’m going to get sick)- nouns. This effects the word for ‘the’, ‘a’ and adjectives before it for example. Word gender is stored in our lexicon with the words and this can be used to make overt violations to people who speak the language e.g le Maison.
A dutch study used the sentence “The burglar had no problem locating the family safe. Of course it was located a … “, since painting is a neutral gendered word, it would usually be substituted. But if big was added with a common gender inflection, painting would be a violation. If bookcase is added instead of painting, then this sentence is grammatically correct so there is no overt violation however there is a covert violation at the possible prediction at the assumed incorrectly gendered adjective. These two sentences were then used; big bookcase and painting to see if there was an overt violation in the bookcase condition.
EEG was analysed when hearing these sentences and a differential ERP was found, when an adjective was presented that was incongruent witch the expected noun: suggesting predictive context effects in certain constrained instances.
What type of model does this research fit best
Interactive model- early effects of context; TRACE model.
