Talking about language processing

[Auto-generated transcript. Edits may have been applied for clarity.]

All right. Good morning, everybody. Uh, welcome to the last cognitive lecture at the tavern.

Thank you for coming at 9:00 on. The last Friday of teaching.

Um. Before I start on the content today, two things.

One, uh, I've been asked to remind you about the end of module questionnaires.

Um. This year. We were particularly interested in what you think about the pace of the lectures and the level of the content.

When you're thinking about that last. Uh, issue.

Remember somebody said. And nobody said it was easy. Um, it's a degree course.

It's supposed to be quite hard. It's supposed to test you, but we don't want it to be too hard.

We don't. We talking so quickly that you can't process the information in the lectures or when you look at the, the recording.

So. The end of module questionnaires are in the usual place, wherever that might be.

Okay, so the other thing is the, uh, the exam tips I'm giving you two already.

One is the short answers and multiple choice.

Right? Something rather than nothing. Short answers.

Choose an answer rather than no answer for the, um, multiple choice.

My first tip today is a supplement to that. If you don't think you know the answer and you're just writing something or other.

If you can make an educated guess rather than just a guess.

Now that varies from question to question. It may simply be that there's something popping up in the back of your mind.

You don't really know if it's right or not, but that's the one that you think it is.

Quite often that's a good indication that something's lodged in your memory, that you're not too sure about it, but try to make an educated guess.

My main topic today is not specifically about the short answers.

It's about allocating time in the exam. So before you go into the exam, find out.

What proportion of the marks are allocated to mix?

What are allocated to show ounces of what's allocated to the essay?

And in the exam I calculate how much time therefore you should spend on each part.

Um, depending on what it's worth. If you calculate, let's say that you should spend 20 minutes on the M6.

Do not spend initially a lot more than that on the M6 session.

Obviously, if you, you know,

you've calculated 20 minutes and you've gone through the ABC queues and you think you've got the whole rights in two minutes,

well, now don't wait another 18 minutes before you start the next session.

Um, but try not if you spend too much time on the M6 on the short answers, you're never actually really sure that's a really.

Rubbish. Smart for it. So try and allocate your time according to the, uh, number of marks allocated to each part.

Okay, let's move on to today's topic, which is language processing talks about language and how complicated it is,

what sorts of structures you see in language, how the structures relate to meanings in a more or less arbitrary manner.

But now let's switch to focus more specifically on the psychology of language.

And what what we need to know and what we do know about the various aspects of language processing.

Bearing in mind that given that language is a very complicated, uh, our knowledge of language is going to be complicated,

and the mechanisms that use that knowledge are likely to be quite complicated as well.

So we want to know as psychologists how languages are used in everyday life.

And the study, the article about the psychology of language or psycholinguistics.

If you want to try and emphasise the, uh, the connection with linguistics, which people like me do and some people don't.

Okay, so it's the studying representations, mechanisms and processes that underlie our ability to use language and but also to acquire language.

Although I won't say anything very much about acquiring language, uh, today.

But it's a very interesting topic in its own right, obviously.

Uh, so, um. Press the button on the wrong computer.

That. So what are we? What are we trying to find out with you?

Psycholinguistics studies language use have said that there are two main aspects of language use.

There's the comprehension side where we understand things, whether that spoken language or written language.

And there's a production side where we speak or write stuff and, you know,

they come together in conversations is to and fro between comprehension and production.

There are three main levels at which processing takes place.

We need to, uh, process words.

We need to put words that I'm talking from a comprehension point of view here.

We need to put words together into complicated structures.

And then we need to think about what meaning is being conveyed by the words in the structures that they are, um, occurring.

And inevitably, the psychology of language is going to draw, um, uh,

descriptions of language that come out of linguistics, the kinds of things I was talking about on Tuesday morning.

But the psychology of language has its own quite different set of questions.

Linguistics itself is not really primarily concerned with details of of, uh, processing language on specific occasions.

Having said that, uh, Noam Chomsky, who's a.

A very influential figure in linguistics in from the mid to mid 20th century onwards.

And still to some extent today, still alive, although perhaps better known as a political activist,

and he made the distinction between linguistic competence and linguistic performance.

And linguistic competence is the knowledge that we have about the language or languages that we speak, whether they be and as they speak.

I mean, they would include, um, things like, um, sign language that I talked about last time.

Linguistic performance, on the other hand, is the use of language in specific situations,

including, uh, mistakes that we make way that we might use the wrong word.

We might, uh, say something that isn't really a proper English sentence.

We might have, um, hesitations or ums and ahs and so on inserted into what we say.

All that comes under linguistic performance, according to Chomsky.

And so. If you look at how people actually use language,

you have to be quite careful in drawing conclusions about what they know, what their linguistic competencies.

And initially in the early 60s, Chomsky started working with some psychologists and a sort of thought that was quite an interesting thing to do.

But when they started say, hey Noam, lad, you might be wrong.

He didn't like that very much. So he decided that psychology was pretty much irrelevant to the study of linguistics itself.

Now. So as I've already said, when you're thinking about use of language in everyday life, you're talking about comprehension.

On the one hand, understanding language, whether you're listening to, uh, to other people,

to to TV to movies, to, uh, TikTok clips or whatever that might be, uh, or reading.

And then there's production, which is the confidence. It's the speaking, the spoken language in the writing we recommend.

And again, as I already mentioned in, in if you're having a conversation with somebody like you're saying that you're doing the production,

they're doing the comprehension and then it's the role switch and so on.

So there are some potentially additional questions to be asked about how that works.

So how are these two sides of language use related comprehension and production?

Well, a plausible view, and probably the correct one is that. They use a common store of knowledge.

So what we know about, uh, say, for English speakers, what we know about English is stored in a common form.

But you have a different set of processes for understanding that language and using the language in production.

There are other views. And you can have a view, for example, called Analysis by synthesis, which says that, uh,

when you're trying to understand something, what you do is you run your mechanism that would produce that.

And that's the that's the mechanism.

But they um, so that that's saying that actually the processes are somewhat the same, not different, but the most common view is that, you know,

when you're looking at the processes that use knowledge and understanding and then, uh, producing language that probably, uh, different sets.

Um. No.

We saw that on Tuesday.

Linguists talk about all sorts of different aspects of language when when you're looking at what's done in the psychology of language,

there's tends to be a focus on three main areas of processing, whether it's in comprehension or it's in production.

That's the processing of words. The processing infrastructure.

I know you've got a string of words, but what's that? What are the structural relations between those words?

Which ones group more closely and which ones group rather less closely?

And so those are to deal with, with, with formal issues.

But then once we've identified say we're thinking about comprehension.

Once we've identified words and the structure that they the full and what does that utterance mean.

How do we use the identities of the words and the structure that we've identified to determine the meaning that's being conveyed?

And of course, you know, it's, um, you know, I'm blathering on here, that's, uh.

Or if you're talking to people, uh, and roles and swapping whatever's happening,

processes at those levels, the levels of identifying words or producing words,

looking at structure, working out, meaning those are going on in a complicated way as the as the language that you're using unfolds.

Uh, and so then there are a number of questions about that, how that actually happens.

If you're considering comprehension. Basically,

the processing at the level of words has to happen first because you can't really decide what the structure is

until you know the words and you can't decide what the meaning is until you know the words and the structure.

So. Basically speaking, that's the order in which processing occurs.

But as I say, as you move, you know, as you go through a conversation, different bits have to be you have to identify the words as they come in,

and then you have to build a structure on top of that, and then you have to work out the meaning.

Um. There are obviously differences between spoken languages and written languages.

I mean, there are some other questions about sign languages and so on which I won't go into today.

Um. In spoken language, you're mainly constrained by how the speaker is speaking, how fast, how clearly, and so and you can't ask people to,

uh, if you're in a conversation with somebody you know quite well, you can ask them to repeat things.

But if you're listening to something on, uh, radio, TV, social media,

whatever you I mean, you can sometimes replay stuff, but your initial comprehension is,

is determined by how quickly and how clearly they're speaking,

whereas things are slightly different in reading because the text is there and you can, uh, skip back to bits if you need to, and, and so on.

So there are some different questions to be asked about understanding of spoken language and understanding of, uh, of written language.

Um. We know.

Quite a lot about, uh, uh, processing of written language.

And we know that if you're going through a text, obviously, um, if you say, say, take the A page or a regular book,

obviously you're tending to go through line by line in the order it's presented to the top to bottom,

as we usually print books in, uh, most of the world.

Uh, but as you go along an individual line of text, your eye is not moving smoothly.

It's helping you along like that. I actually, if you, um, if you get somebody next to you to, uh,

read the book and if you sort of peer down and look at their eyes, you can see their eyes going like this.

These jerky movements are called saccades. And where you stop is, uh, a fixation.

Um. You can. If your neighbour will allow you to do that, you can try that while I'm generating an attendance code today.

Um. They pin.

So, um, the pin for today is 35833583.

Okay, so some of you might have seen that.

That's the pattern of movement. We said particularly with simple text we tend to move forward all the time.

But sometimes we move backwards. Uh that's called a regressive eye movement or regression.

And in more complicated text, that happens quite a lot even when we go.

I mean, sometimes you sort of step back from say, oh, I can't remember what what I read two seconds ago, I'm gonna have to go back and read it.

We're not talking about that. We're just talking about as you go through the first time,

sometimes you instead of going forward, you go backwards to reread something you saw before.

Mostly we have one fixation in the word, but sometimes you can see that sometimes you skip forward within a word, particularly long words.

Lots and lots of nothing about that. It's quite, quite an interesting topic in itself.

So what about, uh, production? What about speaking and writing?

We still got to do processing at the level of meaning, at the level of structure, the level of words.

I've deliberately put those in the reverse order there. Because in production, in producing language, things tend to go the other way round.

You've got an idea about what you want to say, then you've got to come up with a structure, uh, uh, to convey that meaning.

And then you've got to get the individual words to fit into that structure.

Um, so things tend to go the other way around.

There are lots of complicated questions about exactly how these processes,

the levels of words, structures and meanings are intertwined with one another.

They may have slightly different answers in comprehension than in production.

Um, and as I say, there's always this issue, that language, you're always moving it in a sense forward it moving forward in time.

And more language is coming in to do the bits and pieces at each level as you move ahead, whether you're understanding or whether you're producing a.

Um, I think I sort of hinted at this the other day, but that when you're doing.

Uh, research in the psychology of language. There is a tendency to work on comprehension rather than production.

And the reason for that is that it's much easier to find a dependent measure in, uh.

In experiments on comprehension. It might be that the time you spend reading a particular bit of a text or something like that,

whereas in production, particularly if you allow free production,

then what you get out of your participants is some complicated bit of language,

and you've got to decide how to process that to get a dependent variable, feel something.

So that's just for pragmatic reasons.

If you look at research in psycholinguistics, it tends to be on comprehension.

Uh, and. Again, slight, perhaps slightly more.

Unfortunately, it tends to be on written language rather than spoken language again, because it's easier to set up experiments.

I say unfortunately, because of what I was saying last time, which is that, you know, written language is clearly derivative of the spoken language.

Pretty much everybody learns. Spoken language learns.

I mean, they, you know, they acquire it rather they don't, you know, they don't struggle to learn it in any sense.

Literacy comes much later, both in, as they say in biology, in ontogeny, in phylogeny, I ontogeny,

the development of the individual and phylogeny, the development of species in biology.

But it would be a society in which case okay.

So let's let's talk, uh, briefly about these various stages of processing.

Let's start with words. So what's happening is that we're hearing some noises.

Initially that just noises. It's very hard for us to hear speech in our own language is just a noise, but it is just a noise in the sense.

Or we see marks on, let's say, black marks on white paper,

or it might be black characters on a sort of white background on a computer screen or whatever it might be.

Um, and we've got to divide these up into probable words.

Okay. Now, in written language, in, uh, certainly in the way English is usually written, then the divisions between the words are marked by spaces.

So dividing the marks up into words is, in a sense, easy.

Um, whereas um, in spoken language.

Although when we hear spoken language in our native language or a language that we know well,

we seem to hear a stream of words as though they were separated.

If you actually look at the acoustic signal, you know the kind of thing that it looks at.

Um, then you say most of the time, unless you've got a specific pause in what you say, there ain't no boundaries there.

It's a continuous stream of noise. Um.

So we think we hear the brakes, but they're not that and.

Dividing the speech stream into words is called the segmentation problem.

We have to have a. A theory about how that happens.

And the basic theory is that. You have to assume that almost all of every bit of the same stream is assigned to one word.

And there are not very many bits that are assigned to no words.

And you have to use your knowledge of what the words are in your language to do that.

Segmentation. So, um.

Just to illustrate this idea that in written language, we just got the black box on the white background.

Here's a language that almost certainly hardly any of you know my name, which is a language of South India.

Um. As with English, you have it with standard English text.

You have spaces between the words. Um, but beyond that, we don't we don't really know if this is written left to right or right to left.

You know, some languages, like Arabic and Hebrew are written, um, right to left rather than left to right.

Those are just I mean, you could you could identify the same mark or the same symbol occurring in multiple places there,

but you can't really make anything much of that unless you know the language and its script.

Right? Uh, let's hope this works.

Here's the similar, um, demonstration for, uh, spoken language.

Um, and here this is not in English.

So listen to it and see if you can hear the breaks between the words.

This one. Yeah, yeah, yeah. They were if they start with right then it's good to encourage me.

Okay. There are a couple of pauses in there that is basically just a stream of what you do not hear unless you speak best.

You will not hear the breaks between the words in that.

That's what it is. Article one of the Universal Declaration of Human Rights.

That's the English version of it. And it was spoken in Basque. So there is a genuine problem with spoken.

Spoken language is the basic form of language.

There's a problem in spoken language, which you don't find in the written forms of most languages.

Of finding those boundaries between words and the way you have to do is use your knowledge of what words are in your language,

and try and map those onto bits of the speech stream so that.

You get a segmentation of the speech stream with with hardly any bits that are not words.

Uh, but no overlap either between bits. So this bit of sand goes with that word.

And that word, uh, the h bit has to go with one word only.

Okay. So in order to, uh.

Carry out tasks of identifying words and and hence fun.

Get what they mean? We need a store of knowledge, which is a bit like a dictionary.

In the literature it's usually referred to as the mental lexicon.

The mean lexicon is just a, uh, fancy word for dictionary.

And there are lots of interesting questions.

I'm not going to answer these questions. There are lots of interesting questions that you can ask if you know two languages.

Are dictionaries related or separate? If there's a, you know, if they're closely related languages, or at least the example of Italian and Spanish,

which are two of the most closely related of the Latinate languages.

Do you have completely separate mental lexicons for those of, you know, Italian and Spanish?

What if you know English in Malayalam, uh, which are completely from completely different language families?

Two dictionaries or what? It would have, you know, there'd be a meaning, uh, say the word apple in English, and then the corresponding form,

written form in spoken form for both English and Malayalam would be associated with it.

Is that how it works? Uh. Well, it may be different depending what the languages are.

Just I already said there's the question about, uh, how are the processes with comprehension and production related?

Do they use the same dictionary?

Well, for one language, they probably do, because they've already said that the most plausible accounts and probably, uh,

reasonably well supported accounts is that comprehension and production use the same store of knowledge but different processes.

And then you can ask more detailed questions like.

Do the auditory properties of words what a word sounds like?

Does that affect processing when you're processing the written form of the word?

And there's quite a lot of evidence that it does. Okay.

Another issue that's been looked at. I talked to very briefly on Tuesday about morphology.

That is how complex words can be built up by different bits.

Yeah, that's from a linguistic point of view, from a psychological point of view.

Do we actually still build and build as two separate words, or do we, uh, store them as, uh, derivatives of the same basic morpheme build?

Uh, one of which is just the, uh, the basic morpheme itself, and the other is the derived builder,

with an additional um derivational morpheme added to it to turn the verb of building into the noun of the person.

Does that. Um. So there are quite a lot of questions that, uh, that you can answer that.

And, uh, probably, uh, the words are represented in a way that reflects this morphological complexity.

Okay, so if we start asking about more specific processes, say, in the comprehension of,

uh, of language, uh, what what sort of models for identifying words do we have?

Well, since about 1980, there's been a general agreement that identifying words relies on an interconnected set of detectors,

one for each word in the language that, you know, I mean, that's setting aside this issue of morphology.

Um, uh, and they that will also as well as the detectors for the words themselves, they'll be detectors for letters or whatever, you know,

if it's if it's written language letters or if it's spoken language detectors for phonemes and then subcomponents of letters,

uh, or uh, subcomponents of phonemes that I didn't mention or the other day.

But if you're doing phonetics and you're trying to analyse the, um,

meaningfully different sounds of speech, then phonemes themselves have various kinds of properties,

like whether you breathe that when you say that, or whether they whether your tongue's at the back,

middle or front of your mouth when you say them, and so on.

And they. Around 1980.

Uh,

Game of Thrones and Dave Rumble Hart have proposed a model called the Interactive Activation Model for the identification of printed words in English.

And that's the original version of this network type of model. But there's general agreement that that is that that's the right model for, uh.

Identify words, and there's a corresponding model for spoken words uh, again developed by uh McClelland with Jeff Elman.

And that's called trace. Uh, but it's very, very similar to, say, there there are, there are detectors for the words,

but those words are made up of letters in the printed row, phonemes in the spoken form.

And then the letters themselves are made up of, you know, geometrical bits and vertical on the,

on the left hand side or, uh, a diagonal from top left to bottom right or whatever it might be.

And then the phonemes are made up of phonetic features.

Uh. And these are they're interconnected in certain ways, so that the system works in a way that mirrors what you see in spoken language.

So, um. You can make fairly specific predictions.

What you see now, obviously, as I've already hinted, when we're reading or when we're listening to people speak,

this identification of words because very, very quickly you don't notice what's going on.

No idea. That's just as an ordinary language user.

There are complicated mechanisms underlying what was happening and that there are all sorts of known effects of.

On the process of identifying birds. Some of them are fairly obvious, like it's easier to identify a short word.

That alone would, um, it's easier to identify a common word than a, uh, a rare word.

Not not because you might not know the right word.

Even if you know the rare bird, you know the word, and you know what it means, it's still harder to identify.

And then you have things like, does it have it particularly for English?

Does it have regular or irregular spelling? Regular words are easier to, uh, to identify.

And then are there? Many other words that have the same kind of sound pattern.

Does that does that affect how you identify the spoken word?

But as I've hinted above, maybe the having the same kind of sound pattern,

there's lots of other words affects how easy the word is to identify printed form.

There's all sorts of detailed experiments on those, um, kinds of questions.

Um, and we know that these these effects do occur.

Okay, let's turn from words to structure.

So again, just as we we've got this store of knowledge about the words in a language called the mental lexicon.

And we use that to identify words because, you know, the, the each detector is, it corresponds to an entry in the mental lexicon, each word detector.

Again, we have to use our knowledge of how words can be structured into phrases and sentences.

So once we've identified words, how did they go together? If we had no.

The red brick fell onto the ground. Then, you know, we were chunking the red brick into a meaningful unit.

But we're not chunking. Fell onto the.

Fell onto that. Another set of three words. That's not a unit that we want to say is a primary unit in the analysis of those words.

So we got to use knowledge of how words can be structured into phrases to process the words that are coming in,

in comprehension or to produce meaningful sentences or utterances when we're speaking or write.

Um. Again, there's been a lot more work on comprehension here,

and the process of working out structure during comprehension is called syntactic processing.

Parsing syntax is the name for the study of how words are structured into phrases and sentences.

If you look at the, uh, cycling literature on this topic,

one of the major issues that arises is that as you're going was focusing here on on comprehension,

as you're going through a printed text or you're hearing words coming in,

obviously, that are coming in one at a time, literally in spoken language, in written language,

because you're moving across the text, you're processing them in sequence.

Um, but, uh, the problem for the comprehend there is that there may be, you know, you're trying to find the structure,

but given what you've heard so far, there may be a number of different structures that the words could have.

So, for example, if you if you hear the beginning of a sentence, uh, I told the man that that I saw, um, that could continue in two kinds of way.

It could continue. I told the man that I saw last night to meet me at the station tomorrow.

That's all I saw last night is so-called relative clause.

And it's telling you which men are we talking about? It's the one I saw last night.

That's the man I'm talking about. What did I do? I told him to meet me at the station.

Or it could continue just in a different kind of way.

I told the man that I saw four foxes eating from my neighbour's wheelie bin, so that he would not say anything about which man we're talking about.

We'll just say, there's a man. What did I tell him?

I told him about these foxes. Now that's different.

It's easier to see the different types of meaning, but those meanings are associated with different structures.

And at this point, I told the man that I saw, you don't know which of those structures is going to be the one that the sentence folds out into.

So what do we do? Just to say, sometimes those issues are not resolved at all within the sentence itself.

So. A version of an example that's used very widely in the linguistics literature.

Introduced or made famous by Chomsky is something like visiting relatives could be boring, that that sentence is ambiguous.

That is it. Relatives who are visiting me, they're boring.

Or is it my activity of going to visit relatives?

That is boring. And those two meanings, again, is associated with different structures.

In this case we don't know within the sentence itself which is intended.

You have to look at uh, contextual issues that may or may not result.

Another example is the one I gave last time about the British, which is select waffles on the Falklands versus British left waffles on the Falklands.

Now, in spoken languages, you see that the intended structure can be indicated by the way I say it,

but in printed language it's not so, particularly for printed language, there's a.

There's an issue. And another kind of example that's been used widely in psycholinguistics research is the corpse of the man with the telescope.

Is it the man who had the telescope, or did the Cop use the telescope to see it?

And again, that's something that isn't resolved within the census.

You can sort of disambiguate that in spoken language.

The corpse or the man with the telescope, or the corpse or the man with the telescope.

But you don't always do that. I mean, you have to, you know, if you're going to do that disambiguation in spoken language, you have to try quite hard.

So there's an issue. There's an issue of what happens when what you've heard so far might develop into two different structures.

What are you what are you doing? And, uh.

You might say, okay, let's wait till the end of the sentence. Maybe things will be sorted out by then.

But if you say, what's the phenomenology of understanding sentences?

Well, that doesn't seem to be what? What's happened? We have the feeling that we understand things as we go along.

So there are two. There are two main kinds of theories.

One is that we. So we commit ourselves to one of the versions.

And if that turns out to be wrong, we have to go back and reanalyse the sentence.

That's called the garden path theory. Um.

And because there are different structures, it's possible to make a judgements about which structure is more simple and which is more complicated.

Uh, and the garden path theory says, initially you always try for the simplest structure, and if that's not the right one, then bad luck.

And then you kind of expect, you know, you'd expect to sort of feel that somewhere later in the sentence,

you get to a point where you realise it's wrong. In the average to reanalyse it's doesn't happen very often.

Um. A different type of theory.

Um. Uh, again, associated with Jay McClelland and other people is that, you know, we develop all the different possibilities in, in parallel.

And, uh, once a later part of the sentence shows that one of those is wrong, we drop that one, and then we could focus on the other one,

and then we don't have to go back because we've we've had both possibilities in mind all along.

Um, and so you wouldn't expect any, uh, major disruption of processing.

Now, before I move on to an example experiment, I just want to mention another idea.

These sentence structures sound a bit complicated, don't they? Do we really have to worry about them in detail?

Well, one idea is that quite a lot of the time we actually don't.

Uh, have to go into all the detail of working out the structure.

We just do processing. That is good enough. To get or to probably get the right meaning.

Um. This idea derives from?

Very well. I've always thought was a very interesting study done back in the 1970s by Alfonso Mazza.

Um, on aphasic patients. And these are brokers that face people with very hesitant, stuttering speech.

But you're looking at that comprehension, uh, and what, uh, catamounts and zeros found was that.

It looks very much as though what these people were doing is you take the main

words in the sentence and you construct a plausible made meaning out of them.

Um. So, um.

It's rather unlikely that a baby is alive, and it's quite likely that a lion would scare a baby.

So if you read that, the line that the baby's scaring is yellow, you might say, okay,

it's a lion and it's a, it's a lion and a baby, and there's some scaring going on.

What's the plausible message that's being conveyed?

The plausible message is the lion scaring the baby. That's not what the sentence says.

The sentence says that the baby is scaring the lion. But with the brokers they phases who have very hesitant, unstructured speech.

It seems that in comprehension they don't develop this full structure of the sentence.

They just take these words and they put them together and make a plausible meaning.

More recently, although my fellow PhD student Dennis Norris actually pointed this out.

But at the same time, it. Korematsu did his stuff a bit later, but since about 2000 it's been become quite common to say,

well, maybe this kind of process is really quite common in all comprehension,

because most of the time you get the right meaning if you go in the most plausible way, because that's what we mean by most plausible.

That's the one that's likely to be that language is complicated so that we can express implausible things like babies scaring lions,

but that doesn't happen very often. So if we just go with the plausible thing, most of the time we're going to be right most of the time.

And perhaps that's good enough. We know people don't understand what's being said to them exactly, but a lot of the time.

So maybe actually we don't um, maybe we actually don't worry about the fine details of the structure too much.

Anyway, setting that aside, let's, um. Just give you an example of the kind of experiment.

This is an experiment that I did with Jerry Altman back in the uh, 1990s, in the late 1980s.

I can't quite remember that. Uh, it's based on the ambiguity that I was talking about before between relative clauses that tell you,

the ones that tell you which man or woman you're talking about,

and the complement clauses that just tell you what you said or asked or whatever it was.

Um. So the simplest structure in these cases is a sentence like he told the woman that he was worried about many other people.

So. We don't distinguish. There's nothing in there that tells you which woman we're talking about.

It just gives you the content of what was told to the particular woman that's mentioned.

So what was told was that I'm worried about many other people.

Whereas if you say, tell the woman that she was worried about to wait outside, then it's the woman.

Yeah, it's the woman that she was worried about.

So there's an implication that there's more than one woman they might be talking about, and he has to specify which one it is.

And, uh, once what he told her was just to wait outside.

The other bit tells you which woman. And then we've had a lot of control condition.

And that was asked because asks doesn't take a compliment clause.

You can't say. He asked the woman. Uh.

Uh, that he was worried about many other people. So we had those sentences,

and then we put those in a little context that either did or didn't introduce to women that you might need to distinguish between.

So if there are two women you need to distinguish between then this sentence with the relative clause might seem appropriate in that context.

Whereas if there aren't two women, then there's no need to have the relative clause.

So what are you doing? Including it? Um.

These have got the bits in blue there.

That relates to what I was sort of hinting at before that if you start developing a structure and it turns out to be wrong,

but something later in the sentence that tells you whether it's right or wrong.

And that's the point. Then the other also what? Those bits of the sentence away.

Find out whether you were wrong of. Um.

So the results were. If there's no context at all, you just get the sentence.

So with no context, there's no reason to think there are a wrong number of different women.

Um, in that case, when you get to the disambiguation region, if this if you lose in the complicated structure with the relative clause,

you slow down at that point and that suggests, well, you would think it was going to be a relative clause.

You would just think he was going to be this simple structure.

He told her that he's worried about many other people, but if there is a context, if there's a context that's already said,

there are these two women, there's one I'm worried about and there's another one that I'm not worried about.

Uh, then in that context, uh, you don't have the same difficulty with that complicated, more complicated structure.

So the context suggests you need to say which woman you're talking about, and you use a complicated structure that allows you to do that.

And that's understood quite simply. So the appropriate context leads you to choose the right structure, even if it's a complicated.

So it doesn't. It seems that it's not true.

If you read a sentence in context that you just go for the simple structure, first you go to the appropriate structure.

So as this suggests that the garden path is wrong. There's a bit more detail on that in the supplementary slide.

Okay, let's now turn to the meeting. Said words, structure, meaning of the three things that people look at in the psychology of language.

In the. In the first lecture, I talks about concepts. That's the stuff we do on the word meaning.

There's not very much on. Straightforward compositional meaning.

Uh. Putting together the, um, literal meaning of sentences from.

But, uh, the bits and pieces in the sentence structure there,

in that there is quite a lot of work that relates to the kind of pragmatic meanings that I talked about briefly at the end of the lecture on,

on Tuesday. Um. Presupposition and implicit of that kind of thing.

There's also quite a lot of work on two types of links between sentences.

One is use of pronouns. Um.

The man he who sits the or connectives like because and some before and after and.

It may not be completely obvious, but just take my word on this for now.

When you're studying how those types of things create links between sentences,

you often have to make inferences as well to use information that's not implicit and so not explicit in what's written or set.

Up. So I'm going to give you another example of one of my own experiments.

It's a more recent one that I did with Sam Hutton and Scarlet Child, who is a PhD student of mine.

Um. It's based on a. Phenomenon called implicit causality or consequence reality.

Um. And this is true if you have a, um.

A sentence like 1 or 2. Here you go. Because indicating a cold or and so indicating consequence.

But if you just look at the first part of the sentence, says John charmed Bell, it turns out that if you say John charmed Bell,

people tend to think that the reason that happened is because of something about John.

He was a charming person in some sense. Um, whereas there, uh, so in that case, the, um, if you just read the first bits of the sentence,

you think the cause is going to be John, the first person mentioned.

Whereas, uh, you think the consequence is going to be something focusing on Bell.

So John and Bill, what went on to happen after that was something to do with Bill, probably.

And there are other there are other verbs in which this, uh, works the other way round,

that the implicit cause is the second person mentioned, and the implicit consequence is the first person.

Uh. So the question, the kind of question that we asked in the experiment I'm going to mention is.

Do we use that? Information about who the cause is likely to be.

Do we use that right up front in the sentence?

So we already, uh, we already considering that when we're working out what the sentence is going to mean,

or maybe more plausibly, because it could go to a cause and focus on John, or it could go to a consequence and focus on Bill.

Maybe it would be a bad idea up front to be focusing on one of them or the other, because we don't really know how things are going to unfold.

But of course, as soon as we see the because of the answer, then we do know we're going to get a cause or a consequence.

So maybe at that point we do the focusing or maybe we don't.

Maybe we just wait right to the end of the sentence and see how things unfold, because, um.

I don't think I've seen it on this side, but. It's more common for a sentence to continue.

If it's a causal sentence with because it's more common to actually explicitly mention, uh,

something about the implicit clause, but it's not absolutely definite that you do that.

You can go against that.

Uh, you know, you could have something like John Child build because he was easily taken in and that he would be Bill, not John.

So this is a it's a kind of probabilistic phenomenon.

If you go to the implicit clause and you go on to give an explicit clause, those two are likely to match, but they don't have to match.

So again, because things could develop, by the way,

maybe it would be silly just because you saw because to assume you were going to be talking about John rather than Bill.

So the question is, can we find out where you do start thinking more about John than Bill in these sentences?

Is it in the first part before you even know it's going to be a clause or a consequence?

Is it after. Because when you know it's going to be a cause or a consequence, but you're not sure it's going to go with the implicit one,

or do you wait till the end of the sentence and just say, okay, let's wait and see what this person's going to say.

All right. So. So in this this experiment, use the so-called visual world paradigm,

where you you're listening to the sentence and you're looking at a computer screen.

There are four pictures on it. Two of the pictures correspond to the, uh, to, to to John.

And below that we actually used animal names because it's quite hard to have pictures that are clearly of children and Bill.

And then there's one that mentioned something else in the sentence, like the male.

And then there's one picture that represents something that's not even mentioned in the sentence at all.

And so what we're saying is, when is it that you start preferring to look at John over Bill,

indicating that that's the person you're considering the rest of the sentence will be about?

Is it? In the first clause, but I don't even know whether it's going to be a clause or a consequence.

Is it after the big codes are.

And so where, you know, it's a cause or a consequence, but you don't know what exactly it's going to be, or do you wait till the end of the sentence?

So the answer in this study was that before you get to the because of the answer, um.

You do not look preferentially at the whatever corresponds to John or Bill.

Neither is your. You're not favouring either the implicit calls or the implicit consequence.

But as soon as you see the because of the. And so then you start to see the favouring John for the cause of Bill for the consequence.

And you don't obviously that continues to the end of the sentence, but you do as soon as you see it's going to be a cause or a consequence.

You seem to be thinking, well, it probably be the implicit one. It'll probably develop to, to match.

That probably won't go the other way. So.

That's a bit surprising in a way, because, um, yeah, it might it might go the other way.

And if you if you're thinking it's going to be John and it turns out to be Bill,

that's probably going to cause a little bit of a disruption to your understanding of the sentence.

Again as a supplementary slide that shows some of the rather complicated data that comes after that sort of expense.

Um, okay. Let me just say, before I finish a little bit of that dialogue in which, you know, dialogue comprehension,

in which, um, is dialogue or conversation, in which comprehension and production are intertwined.

Um. There's a relatively small amount of work on this,

but one phenomenon that you see all over the place is something called alignment between the speakers.

So they tend to speak more in the same way, all that in the same way than they would if they were talking on their own.

And that's the words they use, how fast they speak, whether they speak in a formal way or an informal way.

There's much more alignment than you would expect by, uh, how they speak on their own.

Um, than, uh, when they're in a conversation than when they're they're not.

Um, um, the. That's for example, in this experiment here I will skip games and have clock they where people had to name,

uh, figures of this kind where they weren't it wasn't clear what they were pictures of.

They tended to in the course of a conversation. They tended to focus in on particular terms.

Obviously it's not total. And the final phenomenon that you might expect to see in dialogue is audience design.

You might expect not just to, to, to, to sort of speak in the same kind of way, but to, uh, coordinate with what you know, that they know.

And what you find is that, uh. This happens to some extent, but it's really quite difficult.

It's quite difficult to tailor what you say to what you know about what the other person knows, because that can be very complicated.

So it doesn't happen quite as much as you might like it to.

And that's a, uh, a finding associated particularly with Boaz, uh, Chicago.

Okay, so to summarise what I've been saying today.

So psycholinguistics that is stored information way.

It's used to produce and understand language. There are different issues in written language in spoken language.

There are questions about structure and how it's used to compute meaning dialogue, which is relatively difficult to study.

The processes are intertwined, but there are some interesting findings about dialogue, about alignment, and about audience design.

Okay. And again, as in the other lectures, there are some supplementary slides.

So some more details of some of this stuff if you want some of it.

Okay. Thank you very much. You are through with cognition apart from your exam in general.

So good luck with your exams. Now I don't want to throw.