Language and Language Processing II

Question 1

What are psycholinguistics?

Answer 1

The study of the representations, mechanisms, and processes that underlie our ability to acquire and use language.

Question 2

Chomsky
- what do they make a distinction between?
- what did chomsky eventually conclude?

Answer 2

Made a distinction between competence and performance

• We know, in some sense (maybe only implicitly), about the structures of our language – structures of words, structures of sentences (Chomsky called that knowledge Linguisitic Competence)
• We use that knowledge in processing language, but we sometimes make mistakes or introduce extraneous elements into utterances (ums and ers for example). (Chomsky called that Linguistic Performance)
• So we need to be careful in using information from the use of language to make inferences about what knowledge people have stored.

Chomsky eventually concluded that Performance could tell us relatively little about Competence

Question 3

What are the two aspects of language processing?

Answer 3

Comprehension
–Listening
–Reading

Production
–Speaking
–Writing

They are intertwined in everyday dialogue

Question 4

How are Comprehension and Production related?

Answer 4

A plausible view is that:
–They use a common store of knowledge
–They each have dedicated processes for using that knowledge

Other views are possible
–E.g. analysis-by-synthesis – the use of production methods in comprehension

Use the mechanisms that produce language to project what you’re hearing

Question 5

What are the three stages of processing?

Answer 5

• Words
• (Sentence) structure
• Meaning (or words/ congrogated processes)

Question 6

Three stages of processing:
1- where does processing at stages take place?
2- for comprehension and for a particular part of the text, what order do processes have to occur?

Answer 6

1- As a text, discourse, or dialogue unfolds through time, processing at all three stages takes place for different parts of the text
2- words, structure, meaning

Question 7

Three stages of processing:
1- in spoken language, what is the listener largely constrained by?
2- in reading, what do we have more control over?

Answer 7

1- how the speaker is speaking – how fast, how clearly etc. (although, it might be possible to ask “could you speak more slowly/say that again?”).
2- the order in which information comes in, but we usually stick fairly closely to the order that it would come in if the same material were spoken

Question 8

Progressing through a text in reading – fixations and saccades

Answer 8

Eyes are not moving clearly, sometimes they’re skipping words or their eyes are moving backwards

Sentence includes: forward saccade, skip, return sweep, regression, refixation

Question 9

What happens with a harder text?

Answer 9

We don’t always progress forever onwards- more regressions
In the more complicated test, you’re more likely to go back and re read

Question 10

Three stages of processing - Production

Answer 10

• Production goes the other way round (message you want to display then construct/ populate sentence structures you want to display)
• You have a meaning that you want to convey (not yet expressed in natural language, but maybe in the “Language of Thought”)
• Then you need sentence structures that allow you to convey the complex meanings you have in mind
• And then you need to find the words to express your ideas.

Question 11

Psycholinguistic research: A pragmatic point
- What has there been a tendency to focus on?
- What has there been a tendency to study?

Answer 11

• Historically, there has been a tendency to focus on comprehension in psycholinguistics, because it is easier to set up well controlled experiments on comprehension than it is for production or for studying dialogue.
• There has also been a tendency to study written/printed language rather than spoken language, also for pragmatic reasons.

–This tendency is not ideal, given the primacy of spoken language.

Question 12

Words

Answer 12

• We hear noises of a particular kind, or see (usually) black marks on a white ground, and have to divide them into (probable) words and then identify each pattern that is a probable word with one of the words we know and have stored in memory (in our “mental lexicon”)

– Then we can find out what it means

• In written/printed language the division into words is usually clearly signalled by spaces and punctuation
• In spoken language there are not usually breaks in the sound stream, even though we think we hear them

– To some extent we have to break the sound stream into words according to what makes sense (the segmentation problem), on the assumption that all (or most) bits of the sound stream have to be assigned to one and only one word.

Question 13

Noises:
- what do you listen to?

Answer 13

• Try to listen for breaks between words. There are some clear breaks but mostly there aren’t
• Article 1 of the Universal Declaration of Human Rights
• All human beings are born free and equal in dignity and rights. They are endowed with reason and conscience and should act towards one another in a spirit of brotherhood.
• Read in Basque

Question 14

What is our knowledge of words stored in?

Answer 14

The mental lexicon

Question 15

The mental lexicon

Answer 15

–If we know two or more languages, there are questions about how the two or more dictionaries are related (or are they separate?)

–There are also questions about how both speech and writing access the same dictionary

• If they do, which seems likely, do auditory properties of words influence spoken word recognition, and vice versa?

– And there are questions about whether comprehension and production use the same lexicon

• And given that words can be morphologically complex, is that complexity represented in the mental lexicon or are related forms (“eat” and “eats” for example) just stored separately? Work by Marcus Taft suggests not.

Question 16

Processing words
1- what does the process of identifying words rely on?
2- who has the original version of this type of network model
3- what is the corresponding model for spoken words?
4- how does this system work?

Answer 16

1- Since about 1990 most people have agreed that the process of identifying words relies on a set of interconnected detectors, one for each word you know, together with detectors for letters (or phonemes) and for subcomponents of letters (or phonological features).

2- See McClelland & Rumelhart, 1981, for the original version of this type of network model (the Interactive Activation Model for written word recognition).

3- The corresponding model for spoken words is the TRACE model of McClelland & Elman, (1986)

4- The system works very quickly – too quickly for us to notice it working. But experiments can detect effects of factors such as how long a word is, how common it is, whether it has a regular or irregular spelling, and whether there are many or few other words that are spelled the same or sound the same

Question 17

McClelland and Rumelhart’s Interactive Activation model

Answer 17

Note the cascade of activation (from bottom to top) from visual features, to letters, to words

Arrows ending in a▲indicate facilitation: features to letters that have them, letters (in positions) to words that have those letters in those positions.

Arrows ending in a ● indicate inhibition. Note that as well as between-layer inhibition, there is within-layer inhibition: If this word is ABLE, it is not TRAP or TRIP or….

feature detectors at the bottom then they feed into letter detectors

A series of connectors- some are excitatory (facilitates the recognition of other letters) and others are Inhibitory (If you think you recognise one word this would be evidence against recognising another word)

Question 18

Processing Structure:
- why do we have to use our knowledge of what structures are allowed in our language?
- how many structures for sentences in english?

Answer 18

• to help us to structure the words which we are just identifying and in the order in which they occur
• Unlike words, of which there are a finite number, there are indefinitely many structures for sentences in English (or any other language), but the possibilities can be represented finitely in terms of rules of combination

SENTENCE = NOUN PHRASE + VERB PHRASE ((the chair)(fell over))

NOUN PHRASE = ARTICLE + NOUN ((the)(chair))

VERB PHRASE – VERB + PARTICLE ((fell)(over))

Question 19

What is syntactic processing or parsing

Answer 19

The process of working out structure in comprehension, using stored rules

Question 20

Structure
What is an issue?

Answer 20

• A major issue is that different structures might be possible, and which is the correct one may be discovered later in the sentence

–I told the man that I saw……..
- ….last night to meet me at the station tomorrow (THAT introduces a relative clause)
- ….four foxes eating from my neighbours’ wheely bin (THAT introduces a complement clause)

• Or not at all
  –Visiting relatives can be boring
  -Relatives who are visiting or the act of visiting relatives
  –British left waffles on the Falkands
  –The cop saw the man with the telescope
  
  • Who had the telescope?

Question 21

Processing structure:
Intuitively we feel we are understanding a sentence as in unfolds, not waiting till the end to figure out what it means.
What are the two main ideas about how this happens?

Answer 21

1. At each point (each word) we make a choice (e.g. chose the simplest structure that the sentence up to that point could have)

If it’s the wrong choice, we have to revise later
The garden-path theory (Lyn Frazier)

2. We develop all possibilities in parallel, and discard them if they become incompatible with later parts of the sentence

Constraint-based theories (MacDonald, Seidenberg, McClelland)

Question 22

Structure – do we compute it?

Answer 22

• Another idea is that we don’t always fully analyse the structure of a sentence, but just do processing that is GOOD ENOUGH to produce the right meaning
  –At least most of the time
• Back in 1976, Caramazza and Zurif showed that agrammatic (Broca’s) aphasics seemed to process sentences by putting the main content words together in a plausible way, so they interpreted:

–The lion that the baby is scaring is yellow
–To mean that The (yellow) lion is scaring the baby

• More recently it has been suggested that this kind of process is common is unimpaired ordinary language users too (e.g. Ferreira et al., 2002)

Question 23

Can Referential Context determine the initial analysis? Garden paths vs contextual constraint

Answer 23

Altmann et al. (1992): use an ambiguity in the meaning of “that” to test the idea that people always chose the simplest structure first (the “garden path” theory)

complement clause (simpler structure):
He told the woman that he was worried about many other people

relative clause (more complex structure, but can be used to distinguish between different women):

He told the woman that he was worried about to wait outside

control:
He asked the woman that he was worried about to wait outside
- Context: no context or a context with two women that you might need to distinguish between
- When you get to the bit in grey (“many people” or “to wait”) you know if you have made a mistake in your analysis.

Question 24

Eye-Tracking Data (Reading Times)

Answer 24

When theres no context (null context) of they’re given the sentence with the relative clause, when you come to the bit that disambiguates the sentence and you find that it should’ve been a relative clause and not a comp clause, then you slow down. This indicates you’re having difficulty.

The middle part of each bar- the difference between comp and rel goes away so no longer having additional difficulty with the rel clause therefore the context has forced you.

Question 25

Altmann et al. (1992) Conclusion

Answer 25

• When there is no context and a complex sentence structure (relative clause) you slow down in the disambiguating region suggesting you have made the wrong analysis so far (complement clause) and need to rethink.
• When there is an appropriate referential context (two women you need to distinguish between, e.g. by a relative clause!!) you don’t slow down at the disambiguating region suggesting you have made the right analysis all along.
• The appropriate (referential) context has led you to choose the right structure (the relative clause that allows you to distinguish between the two women), even though that is the more complex structure.

Question 26

Meaning:
1- who study word meaning
2- what do they tend not to study
3- what do they study
4- what does making those links depend on?

Answer 26

1- Psycholinguists
2- They tend not to study compositional meaning within clauses
3- They study aspects of processing related to some types of pragmatic meaning
–Particularly inferences, which are related to, among other things, presupposition and implicature (from Pragmatics)
They also study ways of making links between clauses
–Pronouns (“he”, ”she”, “it”, “they”, etc.)
–Connectives (“because”, ”and so”, “before”, “after”, etc.)

4- Making those links correctly may depend on inferences (e.g., to work out which of two woman is being referred to by “she” or “her”)

Question 27

Meaning and Inference – just one example Implicit Causality and Consequentiality

Answer 27

1. John charmed Bill because he had an engaging manner
2. John charmed Bill and so he was keen to continue the friendship
   - In 1 the cause of the charming is probably (something about) John
   - In 2 the consequence of the charming is probably (something to do with) Bill
   - Do we anticipate (infer) these causes and consequences when they remain implicit and before they are confirmed at the end of the sentence (and in any case, they might not be confirmed: John charmed Bill because he was easily taken in.)
   - And do we at least wait till we see/hear “because” or “and so”?
   - Or do causes take priority? Or consequences? (if so, we would have to ask: why?)

Question 28

Garnham, Child and Hutton (2020)

Answer 28

Visual world experiment – look at four pictures while you listen to a sentence.

• Looking at a particular picture indicates that the thing depicted is in your mind

The bird charmed the horse during the delicious meal because he had repeatedly sung so many beautiful songs.

The bird charmed the horse during the delicious meal and so he obviously wanted a repeat of the evening

What are people looking at (Bird? Horse? Meal? Something else?) when they hear?

• “the delicious meal” (no info about cause vs consequence, but is there a general tendency to favour one over the other?)
• “he had repeatedly ”/“ he obviously ” (you’ve heard “because” or “and so”, though you don’t know what the exact cause or consequence is going to be, but do you assume it will be the the implicit cause or the implicit consequence?)
• “beautiful songs”/”repeat of the evening” (now you know, so no excuse!!)

Question 29

Garnham, Child and Hutton (2020)
Results

Answer 29

Green line is looks to Bird
Light blue line is looks to Horse

Top two panels are for verbs like “charm”, where the cause is the subject (Bird) and the consequence is the object (Horse). Bottom two panels for verbs with reverse pattern (e.g., “fear”).

No difference at “meal” – first (blue) vertical line

Difference BEFORE last black line - after “because”/”and so” but before end of sentence

So, for “charm”, after “because” or “and so” people look more at NP1 or NP2 respectively

Question 30

Dialogue
Comprehension and Production Intertwined

Answer 30

• What else is involved in engaging in dialogue other than the basic processes of com- prehension and production?
• Alignment - Pickering and Garrod (2004) - participants in a dialogue align their linguistic output at a variety of levels, from low-level factors such as how they pronounce the sounds of their language and how fast they talk, to high-level factors such as the expressions they use to refer to the things they are talking about.
• Much previous work, e.g. by Herb Clark and colleagues, shows how, in dialogue, pairs of speakers home in on agreed expressions for objects they have not seen before, but which they mention several times.

Wilkes-Gibbs and Clark (1992) - “tangram experiment”
“resembles someone that looks like they’re trying to climb stairs. There’s two feet, one is way above the other”
“the person climbing the stairs”
“Stair-climber”

Question 31

Alignment in Dialogue

Answer 31

• Alignment is not total: speakers don’t say exactly the same thing in exactly the same way!
• But there is greater similarity than would be expected if alignment processes were not taking place.

Question 32

Audience Design in Dialogue

Answer 32

• Is ‘audience design’ an important factor in language production?
  –Tailoring what you say to what you know about who you are speaking to?
• In a clear sense, it ought to be, if your goal is to be understood.
• And it’s reflected, for example, in the Wilkes-Gibbs and Clark results on tangram naming (“stair-climber”)
• But it’s hard to do quickly (e.g. Keysar, Barr, & Horton, 1998). Communication is often more egocentric than it should be.