Speech production Lecture Flashcards
Give a very basic conceptual model of speech production
Message - Determine your message
\/
Syntax- determine the syntactical structure with slots for each word
\/
Morphemes- fill the syntactic structures with correct words based on morphemes
\/
Phonemes- choose the sounds that are pronounced
Describe the Sapir-Whorf hypothesis
The specifics of language change (or determine) the content of thought- the language and therefore words we have may determine the concepts we have/ understand
>Linguistic determinism
>Linguistic relativism
Where did Mr Whorf get this idea of language determinism
He was a fire prevention specialist and noticed that people were much less cautious about rules with empty oil drums than full ones despite them being more dangerous due to the vapours. Therefore he hypothesised that concepts in language determine cognition and behaviour.- “empty”
Contrast a weaker version of this hypothesis with a stronger one
Linguistic Relativism- Language can influence (but doesn’t determine) how we think
Linguistic determinism- Language determines thought; Some languages have no words for certain concepts therefore it will not be present in thought. It is the absolute difference.
What two language properties that differ between cultures were used for testing the Sapir-whorf hypothesis?
Words for colours- range anywhere from dark and light (then usually dark, light, red) to a wide range
Words for numbers- some languages very limited (one and many or 1,2,3) or infinite
Describe a set of studies which found evidence against the Sapir whorf hypothesis
The Dani (New Guinea) have two color terms. Dani subjects have no problem to distinguish colors that they cannot name. Just like Western subjects Dani subjects were better at remembering focal colours than non-focal colours
Berlin and Kay (1969): Despite the fact that the amount of colour terms varies with language, there is a constant underlying hierarchy. Indicates that perception shapes language, not the other way around.
Hieder (1972) / Rosch (1973): Dani subjects (2 colour terms) were able to perceive and remember colours for which they have no words. Perception is not determined or influenced by language
Describe a study which found evidence against for Sapir whorf hypothesis
Greeks have a word for dark blue (ble) and light blue (ghalazio). They were given an oddball task to detect the square among circles, some would be a deviant colour- a light or dark version. An ERP measurement was then taken to investigate the Mismatch negativity (MMN) ERP component- i.e are these perceptually different for greek/ English speaking people. this is an ERP component for automatic, pre-attentive change-detection.
The study found that there was, as expected, an MMN found for any deviance. However while MMN was around the same for light green and light blue violations for English speaking participants, the violations in the ‘colour blue’ was much stronger in greek speaking populations.
They concluded that having a separate word/ concept for a colour makes it more deviant in this manner therefore evidence for the weak Sapir Whorf hypothesis.
Retracting a bit, but name an influential early model of speech production
Levelt’s model of speech production
Look in copy
Which part of Levelt’s model does the Sapir whorf hypothesis concern?
There conceptual preparation in terms of lexical concept producing and lexical concept
What is the next process concerned with in Levelt’s model? describe it
Selecting the right word from memory is a 2-stage process in Levelt’s model:
Lexical selection = lemma (information on synactic and semantic properties) =>
Morphological encoding = morpheme/ lexeme (Information about phonetic properties) =>
Describe a study using images which demonstrates the need for these two separate layers
Word frequency was used- some words in English are more often encountered (he, she, can, will) than others (atone, regale, lechary). High frequency words are recalled more easily
However Wingfield found that in a task where participants had to match a word to a picture there was no effect of frequency (spoon/ spatula).
This suggests that at the level of matching a word to a picture (not having too generate it) lexeme extraction does not take place- this occurs at the level of pronunciation.
Describe a study using a literary technique which demonstrates the differences between lexemes and lemmas
A task was given where participants had to translate dutch words into English words. There were three stimuli:
Low frequency lemma, lexeme has a high frequency homophone (leer= leather, learn)
Low frequency lemma and lexeme (No homophone) (fracture)
High frequency lemma and lexeme (No homophone) (corner)
Homophones conceptually have two lemmas for one lexeme if at two different levels.
If linguistic information is combined in one package that included both sound form and meaning then frequency of actual target word is relevant (leather), frequency of homophone (learn) is not: therefore leather and fracture will be slow, corner will be fast
If linguistic information is also represented at the lexeme level, frequency of the homophone will also play a role: therefore leather would also be fast because of its fast homophone in learn, the two controls would remain the same.
Describe the results of the study using phonemes which demonstrates the differences between lexemes and lemmas
As expected, the low frequency control was slower than the high frequency control. However, low frequency words with a high frequency homophone were fast, indicating that connections between words at the phonetic (lexeme) level influence language production.
Describe the methods of a study which investigated which stage (lexeme/ lemma) came first
Used a picture-word interference task: people have to identify a picture (screw) vocally while hearing a distractor word in their ear which is related at the semantic (lemma) level (bolt), at the phonological (lexeme) level (shrew) or an unrelated distractor (tart). In addition to this, various stimulus onset asynchronies (OAS) of the distractor were manipulated:
Word before image (-150ms)
Word and image at the same time (0ms)
Word after image (150ms)
Response times in producing the correct word were recorded.
Describe how the model would predict the results of the study which investigated which stage (lexeme/ lemma) came first
The model shows the lemma (image- concept) coming before the lexeme (the noise)
This therefore predicts that hearing the unrelated lexeme (tart) shouldn’t be a too problem when hearing it while seeing the lemma- at least no worse than if presented when reporting the lexeme (no effect of order.)
It would also predict that hearing a semantically related word (bolt) while being presented the image would interfere with the lexical access and therefore be slower. However if it was presented at the lexeme stage, it shouldn’t make much of a difference compared to an unrelated word.
It would predict that hearing a phonologically related word (shrew) when presenting the image shouldn’t fuck up lexical access too much but would actually help when presented while trying to receive the lexeme (known from previous research.)
Semantic distractors should have an early effect while phonological distractors should have a late effect.
Describe the results of the study which investigated which stage (lexeme/ lemma) came first
At -150ms before stimulus:
Phonological distractor has no difference in effect to unrelated stimulus
A semantic distractor has a much slower response than the unrelated stimulus
At the same time as stimulus:
A phonological distractor has a fatser response than the unrelated stimulus
A semantic distractor has a more comparable response to the unrelated stimulus
150ms after the stimulus:
A phonological distractor has a much faster response than the unrelated stimulus
A semantic distractor has almost the same response to the unrelated stimulus
Therefore semantic has an early effect, phonological has a late effect as predicted: Lexeme => Lemma
Outside of experimental evidence, where else can evidence for/ against for Levelt’s model be found? Describe how this can then be manipulated in a study
Speech errors (jeremy cunt), which can be be induced experimentally e.g b- d, b- d, b- d, d- b* can make a b- d error at the phonological level. This results in 30% speech errors if the switch would result in a real word (ban- man), 10% speech errors if the switch leads to non-words (Dart- board). This effect is called lexical bias, which is problematic for Levelt’s model.
describe why lexical bias is problematic for Levelt’s model
Phoneme selection occurs at the phonetic encoding stage, this stage which supposedly occurs after the lemma is selected (also after morphological encoding and syllabification) therefore the lemma decision stage should be finished with before the phonemes are selected. Yet there are more phoneme switches if the error results in words than in non-words, showing an influence of lexical selection on on this later stage of phonetic encoding. This violates the predictions of the model
Describe a model which better describes this effect, and how it explains this lexical bias
Dell’s interactive model shows interactive activation in speech production. There are interactive connections between the phonemes and the lemmas and interactive connections between the meanings and the lemmas. This means that lower levels can affect higher levels and higher levels can affect lower levels (recurrent processing.)
Similarly to how the TRACE model facilitates the word superiority effect, Dell’s interactive model better explains lexical bias. The phonemes activate the possible lemmas which reinforce the letters and also activate the possible semantics. Therefore the lemmas that are both activated by the possible phonemes and have semantic connections will be the most activated during selection, which can lead to a mistake and the wrong phoneme/ lemma being used.
Describe a study which tests Dell’s model
Used words pairs with potential speech errors to test the interactive activation in speech production.
The prediction of Dell’s model is that phoneme exchanges that lead to non-words happen most often when there is little time to spread (recurrently) between the phonetic and lemma level. Therefore:
Fast speaking rate: more speech errors leading to non-words
Slow speaking rate: less speach errors leading to non-words
Describe the results of the study testing Dell’s model
Fast speech: slightly more non-word than word outcome speech errors
Slow speech: lexical bias (more errors that lead to word outcome)
Giving evidence for Dell’s model.
Describe a study that possibly demonstrates the self monitoring aspect of Levelt’s model
Moller et al. found distinctive EEG activation around the pre-SMA just before you make a mistake in the word-pair task. Due to it being before the mistake was realised and due to its location, it is thought that this could represent competition between two motor implementations of speech.
Describe the monitoring stage of Levelt’s model using the correct terminology in two different aspects
There is a prediction of what you intend to say, known as the auditory target map, and then there is what you actually say called the auditory state map. Monitoring inspects whether there is a difference between these two, known as the auditory error map
It is also supposed that not only is there a prediction of how it will sound, but how you will make that sound- the somatosensory target map vs the somato sensory map = the somatosensory error map.
What brain area is associated with the auditory monitoring function?
The Planum Temporal (Heschl’s Gyrus)
Remember: How stream of speech perception, listening to stuff and relating it to motor activation
Describe a study which investigated which brain areas are associated with this auditory speech monitoring and what they found
The study had participants wear headphones and pronounce phonemes. What the subject hears is slightly distorted from what is actually said; when the target map differs from the state map. fMRI measured brain activity while this occured and found brain activation in the posterior STG and Spt.
What brain areas are associated with the somato-sensory monitoring function (2)? Describe a study which demonstrated this
Ventral somatosensory cortex, anterior supramarginal gyrus
The study had participants pronounce pseudowords (Abi, agi) and for some words , a little balloon was inflated between the upper and lower jaw, therefore disrupting the motor plan.
When this occurred, activation was observed in these areas.
