Previously, I wrote that language processing could be broadly divided into language production and language comprehension. In this post, I will talk about comprehension.
Firstly, I will discuss the ways we perceive words and their components and then will come to processing the meaning; especially I will stop on words and sentences which have several meanings (that is, ambiguous ones): how do we choose the right one?
Firstly, I will discuss the ways we perceive words and their components and then will come to processing the meaning; especially I will stop on words and sentences which have several meanings (that is, ambiguous ones): how do we choose the right one?
Perceiving Phonemes and Words
Phonemes
Phonemes are the smallest components of words; sounds (don't confuse them with letters, which can sound differently in various cases). We perceive them automatically, relying on words which they are part of. In 1970, Richard Warren made his participants to listen to sentences, in which one phoneme, 's', was replaced with a sound of a cough. None of participants noticed the missing 's'. Warren then came up with a term phonemic restoration effect: 'filling in' of the missing phoneme based on a context. It is an example of top-down processing, in which the processing of overall meaning precedes and aids processing of separate components.
Words
Number of experiments showed that top-down processing is again at work here; words comprehension is aided by context provided by the rest of the words and sentences in the conversation. In isolation, it is quite hard to perceive separate words, as there are many different ways of saying them - depending on pronunciation, accent and intonation.
In a speech words are often pronounced together, without a perceivable break between them - however humans are very good in identifying and perceiving individual words in a continuous speech flow. This process is called speech segmentation.
Segmentation is helped very much by a context and semantic meaning of a sentence. Consider: 'Be a big girl and eat!' and 'Big Earl loved his car'.
There are other clues for the listeners as well. Learning a language, we naturally, subconsciously learn that certain sounds follow some sounds more frequently within a word, while others are more likely to be separated by a space.
Phonemes are the smallest components of words; sounds (don't confuse them with letters, which can sound differently in various cases). We perceive them automatically, relying on words which they are part of. In 1970, Richard Warren made his participants to listen to sentences, in which one phoneme, 's', was replaced with a sound of a cough. None of participants noticed the missing 's'. Warren then came up with a term phonemic restoration effect: 'filling in' of the missing phoneme based on a context. It is an example of top-down processing, in which the processing of overall meaning precedes and aids processing of separate components.
Words
Number of experiments showed that top-down processing is again at work here; words comprehension is aided by context provided by the rest of the words and sentences in the conversation. In isolation, it is quite hard to perceive separate words, as there are many different ways of saying them - depending on pronunciation, accent and intonation.
In a speech words are often pronounced together, without a perceivable break between them - however humans are very good in identifying and perceiving individual words in a continuous speech flow. This process is called speech segmentation.
Segmentation is helped very much by a context and semantic meaning of a sentence. Consider: 'Be a big girl and eat!' and 'Big Earl loved his car'.
There are other clues for the listeners as well. Learning a language, we naturally, subconsciously learn that certain sounds follow some sounds more frequently within a word, while others are more likely to be separated by a space.
Comprehending Words
Word Frequency Effect
Now, I would like to discuss our understanding, rather than perceiving, of the words. One of the effects that is involved in comprehension of words is word frequency: the relative frequency of words usage.
This effect is apparent in the Lexical Decision Task, in which participants are presented with two sets of words (one of frequently used and another one - of rare words), and are then told to decide which of them are actually real words and which are not.
On average, it takes people 750ms (with 1% errors) to determine whether a highly frequent word is a real word, and 800ms (5% errors) - to do the same with the words of low frequency. Thus, our past experience with the words influences our ability to access their meaning.
While reading, we fixate words for certain time (that is, look at them), and then jump to others (in something which is called a saccade). However, it is important to understand that sometimes we come back to the same word several times during reading; in other words, we might fixate it numerous times. In such cases we can talk of gaze duration: total time of all the fixations on a word before leaving it.
With the use of Eye Tracking devices, it was found that people tended to fixate low frequency words for longer than high frequency words (on average, 288ms and 258ms respectively). More interestingly, gaze duration is significantly higher too: 328ms compared to 289ms. It means that we are more likely to come back to a rare word after the initial fixation.
It was also found that fixation times overall are much shorter than lexical decision times. Therefore, we make semantic decision after we have read a word/sentence - but not at the same time.
This effect is apparent in the Lexical Decision Task, in which participants are presented with two sets of words (one of frequently used and another one - of rare words), and are then told to decide which of them are actually real words and which are not.
On average, it takes people 750ms (with 1% errors) to determine whether a highly frequent word is a real word, and 800ms (5% errors) - to do the same with the words of low frequency. Thus, our past experience with the words influences our ability to access their meaning.
While reading, we fixate words for certain time (that is, look at them), and then jump to others (in something which is called a saccade). However, it is important to understand that sometimes we come back to the same word several times during reading; in other words, we might fixate it numerous times. In such cases we can talk of gaze duration: total time of all the fixations on a word before leaving it.
With the use of Eye Tracking devices, it was found that people tended to fixate low frequency words for longer than high frequency words (on average, 288ms and 258ms respectively). More interestingly, gaze duration is significantly higher too: 328ms compared to 289ms. It means that we are more likely to come back to a rare word after the initial fixation.
It was also found that fixation times overall are much shorter than lexical decision times. Therefore, we make semantic decision after we have read a word/sentence - but not at the same time.
Semantic Priming
In the same Lexical Decision Task, two types of lexical pairs are presented to participants: those semantically related and those which are not. In the case with semantically related pairs, the first word is a prime, while the second one - a target. On average, people tend to determine whether the target word is a real word 30ms quicker when it is preceded by a prime than when preceded by unrelated word (750ms and 780 ms respectively).
For example, on average we would expect people to respond 30ms quicker to 'nurse' if it is preceded by 'hospital' (semantically related word), than if it is preceded by 'table' (semantically unrelated).
It shows that people have a 'mental lexicon' which is organised by meaning; when one meaning is activated, the activation spreads to other, semantically related concepts, too.
For example, on average we would expect people to respond 30ms quicker to 'nurse' if it is preceded by 'hospital' (semantically related word), than if it is preceded by 'table' (semantically unrelated).
It shows that people have a 'mental lexicon' which is organised by meaning; when one meaning is activated, the activation spreads to other, semantically related concepts, too.
Lexical Ambiguity
One word can have separate meanings; how do we decide which one is right in a specific case?
If we encounter a word in isolation, we are likely to choose its most frequent meaning. But what if the word is a part of a sentence? Surely, context plays a crucial role in determining the word meaning - but do we only think of the right one straight away, or do all the possible meanings become activated before we select the one to fit the context?
In 1979, David Swinney conducted a famous research on how we access the meaning of ambiguous words. You can read more about Swinney's experiment here. What he found was that the moment we encounter a bias word (in his study, 'a bug', which could be either an insect or spying device), we initially access all its meanings - no matter what the context (or, semantic meaning of an overall phrase) is. However, the context kicks in couple of milliseconds later, making it possible for us to choose the meaning appropriate for the context.
However, comprehension of ambiguous words also depends on frequency of their meanings. If all of them are equally frequent (as in 'a bug'), we access all of them initially - and then decide which one to choose, depending on a context. In some cases one of the meanings may be much more frequent than the other(s); in this case, we only access the most frequent one when we encounter the word, and only if it does not fit the context we reconsider its meaning.
It can sound a bit confusing; let's have a look at the example. Read the following sentence: 'Last night, the port had a strange taste'. People typically think of 'port' as of a harbour, and it is the only meaning that pops into their head when they are reading the sentence. However, after they finish the sentence, it becomes apparent that this meaning does not work. We then come back to the word 'port' and find its less frequent but appropriate for the case meaning, which is wine.
This was found with the use of Eye Tracking devices. When reading sentences similar to the one about 'port', people fixate the rare ambiguous word for just as long as other words (despite its ambiguity), however after finishing the sentence they tend to come back to it again - presumably, accessing its second meaning.
If we encounter a word in isolation, we are likely to choose its most frequent meaning. But what if the word is a part of a sentence? Surely, context plays a crucial role in determining the word meaning - but do we only think of the right one straight away, or do all the possible meanings become activated before we select the one to fit the context?
In 1979, David Swinney conducted a famous research on how we access the meaning of ambiguous words. You can read more about Swinney's experiment here. What he found was that the moment we encounter a bias word (in his study, 'a bug', which could be either an insect or spying device), we initially access all its meanings - no matter what the context (or, semantic meaning of an overall phrase) is. However, the context kicks in couple of milliseconds later, making it possible for us to choose the meaning appropriate for the context.
However, comprehension of ambiguous words also depends on frequency of their meanings. If all of them are equally frequent (as in 'a bug'), we access all of them initially - and then decide which one to choose, depending on a context. In some cases one of the meanings may be much more frequent than the other(s); in this case, we only access the most frequent one when we encounter the word, and only if it does not fit the context we reconsider its meaning.
It can sound a bit confusing; let's have a look at the example. Read the following sentence: 'Last night, the port had a strange taste'. People typically think of 'port' as of a harbour, and it is the only meaning that pops into their head when they are reading the sentence. However, after they finish the sentence, it becomes apparent that this meaning does not work. We then come back to the word 'port' and find its less frequent but appropriate for the case meaning, which is wine.
This was found with the use of Eye Tracking devices. When reading sentences similar to the one about 'port', people fixate the rare ambiguous word for just as long as other words (despite its ambiguity), however after finishing the sentence they tend to come back to it again - presumably, accessing its second meaning.
Comprehending Sentences
Semantics and Syntax
All kinds of stimuli activates neurones in brain: light, sound - and different types of language too. This activation (Event Related Potential) can be measured by disc electrodes which are placed on scalp. When encountering a sentence brain sends two waves: N400 - a negative wave around 400ms after the stimulus, and P600 - a positive wave around 600ms after the stimulus.
Using the electrodes, it was found that brain responded differently to grammatically incorrect and semantically implausible sentences. It sends a much larger N400 wave with semantically anomalous stimuli, and a much larger P600 with syntactically anomalous stimuli. Such finding suggests that brain distinguishes between syntax and semantics.
With this in mind, a question arises: how do we deal with syntactic ambiguities? Sometimes we encounter a sentence with ambiguous meaning such as: 'I ran to a man with a briefcase'. The problem is, do we group 'I' with 'briefcase' or 'man' with briefcase'? This sentence is syntactically ambiguous and studying how people deal with such ambiguities can help us to understand how we comprehend language.
Using the electrodes, it was found that brain responded differently to grammatically incorrect and semantically implausible sentences. It sends a much larger N400 wave with semantically anomalous stimuli, and a much larger P600 with syntactically anomalous stimuli. Such finding suggests that brain distinguishes between syntax and semantics.
With this in mind, a question arises: how do we deal with syntactic ambiguities? Sometimes we encounter a sentence with ambiguous meaning such as: 'I ran to a man with a briefcase'. The problem is, do we group 'I' with 'briefcase' or 'man' with briefcase'? This sentence is syntactically ambiguous and studying how people deal with such ambiguities can help us to understand how we comprehend language.
Parsing
Parsing is the process of grouping words into phrases. To study it, psychologists present the sentences which create temporary ambiguity, that is, sentences which could unfold in two different ways. The following sentence could be an example: 'Cast iron sinks quickly rust'. You probably were assigning different meanings to the sentence as you were reading it? First you could think it was about iron which sinks, but only when later it becomes clear that it is about iron sinks which rust. In this case, we can say that the perceiver was 'garden-pathed': that is, led to a wrong meaning.
There are two approaches to how we deal with such ambiguities: modular and interactionist approaches.
a) Modular theory (syntax-first approach)
This approach suggests that there are independent areas of the brain which analyse syntax and semantics independently from each other; parsing is based on grammatical structure of a sentence and the principle of late closure. Late closure principle, in its turn, states that every new word we encounter we perceive as a part of a current phrase - and we do so until it stops making sense.Think about the iron sinks sentence once again - we tend to add every word to the current phrase, but when it stops making sense we go back and rearrange the parsing basing on semantics.
Because modular theory suggests that we analyse syntax separately from semantics, according to it, initially we adopt the syntactically simplest analysis: that is, main clause analysis. Then, if the sentence is incompatible with such analysis, we reanalyse.
b) Interactive theory
This approach states that semantics determines parsing at the same time with syntax - not after, as syntax-first approach assumes. It is quite easy to see the evidence for this approach. Consider the same sentence: 'I ran to man with a briefcase'. The ambiguity arouses because both interpretations make perfect sense. Now let's change one word without altering syntax: 'A cat ran to man with a briefcase'. You must have immediately understood that briefcase belongs to a man, not a cat. Thus we can see, that parsing is guided by semantic meaning as well as by syntax. This idea was further developed by Michael Tanenhaus (1995), who used Eye Tracking in his 'Apple and Towel' experiment; you can read on it here.
This idea was also supported by a Trueswell's study (1994). He used Eye Tracking to see how long people would fixate on syntactically ambiguous and unambiguous sentences, in order to test which theory would prove to be right. He found that people used semantics in order to adopt an appropriate syntactic analysis, thus proving Interactive theory to be right. You can find his paper HERE if you want to read it online; or, download it from this page (no viruses or pop ads, guaranteed)).
There are two approaches to how we deal with such ambiguities: modular and interactionist approaches.
a) Modular theory (syntax-first approach)
This approach suggests that there are independent areas of the brain which analyse syntax and semantics independently from each other; parsing is based on grammatical structure of a sentence and the principle of late closure. Late closure principle, in its turn, states that every new word we encounter we perceive as a part of a current phrase - and we do so until it stops making sense.Think about the iron sinks sentence once again - we tend to add every word to the current phrase, but when it stops making sense we go back and rearrange the parsing basing on semantics.
Because modular theory suggests that we analyse syntax separately from semantics, according to it, initially we adopt the syntactically simplest analysis: that is, main clause analysis. Then, if the sentence is incompatible with such analysis, we reanalyse.
b) Interactive theory
This approach states that semantics determines parsing at the same time with syntax - not after, as syntax-first approach assumes. It is quite easy to see the evidence for this approach. Consider the same sentence: 'I ran to man with a briefcase'. The ambiguity arouses because both interpretations make perfect sense. Now let's change one word without altering syntax: 'A cat ran to man with a briefcase'. You must have immediately understood that briefcase belongs to a man, not a cat. Thus we can see, that parsing is guided by semantic meaning as well as by syntax. This idea was further developed by Michael Tanenhaus (1995), who used Eye Tracking in his 'Apple and Towel' experiment; you can read on it here.
This idea was also supported by a Trueswell's study (1994). He used Eye Tracking to see how long people would fixate on syntactically ambiguous and unambiguous sentences, in order to test which theory would prove to be right. He found that people used semantics in order to adopt an appropriate syntactic analysis, thus proving Interactive theory to be right. You can find his paper HERE if you want to read it online; or, download it from this page (no viruses or pop ads, guaranteed)).
trueswell_tanenhaus.pdf |
Comprehending Text and Stories
Inferences
Comprehending a story, we don't solely rely on information given directly in the story. It is a creative process in which we use our prior knowledge about the world and make inferences. These are needed to create a coherence: key feature of any story or text, which states that of a narrative relates to others. There are three types of inferences that we make to reach coherence.
1. Anaphoric inference
This inference connects an object/person in one sentence to objects/persons in other sentences. We don't even realise we are making any effort in cases when these inferences are obvious: 'I have been to this cafe before. Coffee there is good and quite cheap!' It is obvious to us that 'there' relates to a 'cafe' from the first sentence. However sometimes making anaphoric inferences is slightly more challenging. Think about this one: 'Thelma and Louisa attend statistics lectures with me. They are so boring!!' Is it stats lectures which are boring (no way!) or is it Thelma and Louisa who annoyed me to the point of exclamation marks (perhaps with their pretentious names)?
2. Instrument inference
These inferences are all about tools and methods. So, if I say 'I finished writing my essay yesterday!' you would probably have understood I used a computer to type it, not a pen or a duck feather.
3. Causal Inference
These are our inferences that the actions/events in one sentence/clause were caused by the actions/events in a previous sentence/clause. For example: 'I never read JungMinded statistics posts. My exam went really bad!' So, this causal inference is really obvious, but they not always are like this. 'I thought of him for couple of minutes. Soon he called.' Have her (or his?) thoughts cause him to call? Unlikely. However, what if it is a sic-fi book about a girl (boy?) with an ability to communicate with others mentally?..
Situation Models
Situation Model is a mental representation of a text's meaning, its message. It does not involve information about sentences, phonemes and morphemes; only objects, characters and their actions, locations...
There is a theory that one of the ways people form these models is as follows: they mentally simulate motor and perceptual characteristics of the objects and actions of a story. The study was made by Rolf A. Zwaan; you can find the report of his experiment here.
Another great experiment was conducted by Horton and Rapp in 2003, in which they showed that the participants experienced a story just as if they were really experiencing the described events themselves.
Physiology of Simulations
But how do these simulations really work? In 2004 Hauk conducted an interesting experiment, in which he measured participants's brain activity twice: in one case - when they were moving their foot or finger, and in another - when they read 'action words', such as 'pick' or 'kick'. In both cases approximately same brain areas were active: a bit closer to the brain centre line in case of an actual action, and slightly further from the centre - when reading about the same action.
Comprehending a story, we don't solely rely on information given directly in the story. It is a creative process in which we use our prior knowledge about the world and make inferences. These are needed to create a coherence: key feature of any story or text, which states that of a narrative relates to others. There are three types of inferences that we make to reach coherence.
1. Anaphoric inference
This inference connects an object/person in one sentence to objects/persons in other sentences. We don't even realise we are making any effort in cases when these inferences are obvious: 'I have been to this cafe before. Coffee there is good and quite cheap!' It is obvious to us that 'there' relates to a 'cafe' from the first sentence. However sometimes making anaphoric inferences is slightly more challenging. Think about this one: 'Thelma and Louisa attend statistics lectures with me. They are so boring!!' Is it stats lectures which are boring (no way!) or is it Thelma and Louisa who annoyed me to the point of exclamation marks (perhaps with their pretentious names)?
2. Instrument inference
These inferences are all about tools and methods. So, if I say 'I finished writing my essay yesterday!' you would probably have understood I used a computer to type it, not a pen or a duck feather.
3. Causal Inference
These are our inferences that the actions/events in one sentence/clause were caused by the actions/events in a previous sentence/clause. For example: 'I never read JungMinded statistics posts. My exam went really bad!' So, this causal inference is really obvious, but they not always are like this. 'I thought of him for couple of minutes. Soon he called.' Have her (or his?) thoughts cause him to call? Unlikely. However, what if it is a sic-fi book about a girl (boy?) with an ability to communicate with others mentally?..
Situation Models
Situation Model is a mental representation of a text's meaning, its message. It does not involve information about sentences, phonemes and morphemes; only objects, characters and their actions, locations...
There is a theory that one of the ways people form these models is as follows: they mentally simulate motor and perceptual characteristics of the objects and actions of a story. The study was made by Rolf A. Zwaan; you can find the report of his experiment here.
Another great experiment was conducted by Horton and Rapp in 2003, in which they showed that the participants experienced a story just as if they were really experiencing the described events themselves.
Physiology of Simulations
But how do these simulations really work? In 2004 Hauk conducted an interesting experiment, in which he measured participants's brain activity twice: in one case - when they were moving their foot or finger, and in another - when they read 'action words', such as 'pick' or 'kick'. In both cases approximately same brain areas were active: a bit closer to the brain centre line in case of an actual action, and slightly further from the centre - when reading about the same action.