Psychology 247 Cognitive Psychology
Language Comprehension
Erwin Segal
Return to syllabus
What is available for comprehension?

1. Context of utterance.
        Situation Semantics  (Situations and Attitudes, Barwise & Perry, 1983)

"[o]ne of the simplest facts about human language [is the fact that] an utterance must be made by someone, someplace, and sometime. That is, an utterance always takes place in a discourse situation, and so the facts about the discourse situation can always be exploited to get from the meaning of the expression used to whatever information is to be conveyed." (pp. 32-33).
        This includes the relation of the interlocutors, what has been previously said, the body language and expressions of the speaker, the social status of the speaker, what the interlocutors know or assume about one another, etc. If I (E. Segal) had to choose the most important component of the situation for understanding and interpreting a sentence, I would have to say that it is where in the discourse the sentence occurs. The interpretations of new sentences are hugely dependent upon the previous discourse.

2. The facts of the utterance
    This includes the choice of words, the grammatical form of the utterance, and the intonation pattern of the speaker. Each of these are the same, or may be simulated to a greater or lesser degree if the expression is in writing. Writing also includes white spaces, paragraph structures, headings, etc. Anything available may be exploited for communication.

Levels of meaning:

Anderson begins with a discussion of parsing. This is a critical issue, because comprehension is dependent upon semantic combinatorics. That is, the meaning conveyed is dependent upon combining many different components of meaning from many different sources, particularly the words and grammar of the utterances. If there were no process of cognitively isolating some of the parts there would be nothing to combine. Sentence constituent is the term that is most often used to identify the parts of a sentence that are the results of accurate parsing.

Parsing
1. Logically, the first task for comprehension is identifying the lowest level units in the utterance. What are the letters or phonemes? This is a tough computational task, and many researchers, such as Jim Sawusch of our Cognitive area of psychology have worked many years on. How do you know that the sound represents a /b/ rather than a /p/, an /s/ and not an /sh/, an /uh/ not an /ah/? Especially, considering the fact that in a dynamic context, some of the components get transformed.

Although one can fairly clearly identify a logical order in which the units in an utterance get combined into higher order units, the different levels do not all proceed sequentially, there are times that top-down processes are necessary. That is when information from some beginnings of higher level processing gets sent "down" to help decide on the units at a lower level.
2. Given the elements, logically, the next task is to parse the utterance into basic syntactic and semantic units. This is also a complex task; there are times when the same elementary units, with identical physical properties can be combined into different words. This weekend the expression "New dancers" got parsed as "nude dancers" in my presence. Not only does one have to identify words, but also affixes of words, such as tense, and number in English. English generally uses function words such as prepositions to mark case relations, but these and other categorizations are often marked by affixes in other languages. These grammatical markers are extremely important in understanding language.

3. The next level is the combining into units greater than the word. There are several possible levels above the word within a sentence. This level also is not always obvious, and mis-parsing easily leads to misunderstanding. Although words are always temporally contiguous, i.e., words cannot be separated into two components with other linguistic units between the components, higher order units at times have the potential to become discontinuous.

There is much data that shows that readers and hearers parse utterances into the different constituents. For example, when recalling a sentence people are more likely to recall a whole constituent when the get the first word, and they often forget the first word of a constituent. The higher in the parsing tree a constituent, the more likely a subject will not get the transition right. If a text is presented to be read and it is presented in parts, if the parts are not broken at constituent boundaries, the text is much harder to read. When trying to remember a sentence, a person is more likely to pause at a constituent boundary than at other locations.

Extracting meaning

Relevant information tends to be accessed as soon as possible. After a meaning bearing constituent is identified, the next step is to access its meaning and to combine it with other meanings. The data show that within milliseconds of receiving a word Ss start accessing its meaning. There seems to be a "lexicon in the head" that has meanings of the words associated with it which are activated.
    Data show that if you test immediately (within 200 msec. or so) multiple meanings of ambiguous words are seen to be activated, but within a very short time only the relevant ones are still active. The relevant meaning is integrated with the rest of the currently active meanings. The combinations of meanings from different sources narrow down the possible interpretations.

The young women looked beautiful as they danced at the colorful ball.
The young boys kicked the colorful ball.
Syntactic cues: People use function words and grammatical information to structure the sentential content. Len Talmy wrote an important paper "The relation of grammar to cognition" (B. Rudzka-Ostyn,
Ed., Topics in Cognitive Linguistics) in which he argued that grammar as represented by word order, affixes, and function words structure the cognitive representation of the utterance. The grammar represents a set of notions which are important in the understanding of most utterances. Some of the kinds of things which can be presented by grammar include number, (singular-plural) spatial relations (in, on, near) temporal relations (tense--present, past, past perfect) relation to the speaker (I, you, here, there, this, that) abstract paths (through, into, onto) and many others, some of which are not true about English. When we hear these features, we know how to relate the content terms that are in the standard vocabulary. By using the grammatical structure when we parse a sentence we can relate the lexical items to one another in a meaningful way. Interestingly this structure aids our recall even in 'content free' situations. This "sentence" The brammers ricked the lerses before the grom maset corted. is easier to remember than the list of nonsense words [bram, rick, lerse, grom, maset, cort]. Nonsense sort of makes sense.
    LouAnn Gerken showed that young children can understand sentences better when they are presented with the appropriate grammatical context than when it is omitted or substituted for. This is true although the children are in the one or two word stage and there is no evidence that the children ever speak those forms. The grammar seems to be backgrounded for children, it structures their understanding and probably their speech, but they have not learned to incorporate it in their utterances. Thus children at early stages of speech tend to utter only the content words (R. Brown).
    The words have meanings. These can be accessed via their form. The order of the words and the grammatical markers also have physical manifestations which can be accessed by the listener/reader. These help tie the meanings of the different words together to constrain which of different meanings are intended, and to add other aspects of the sentential meaning. These different parts of the meaning analysis seem to play different neurophysiological roles in sentence processing. There are different regions of the brain which are tied to lexical and grammatical processing and ERPs seem to react differently to anomalies in syntax and semantics. (p. 398f).

Sense making
We make use of all kinds of information in order to understand language. It seems that we do not automatically apply algorithms to process the information. Rather, we seem to be governed by an attempt to make sense out of the received discourse. This starts with the beginnings of language acquisition. Dan Slobin gave young children reversible 'The lion chased the tiger.' and semantically nonreversible sentences 'The girl picked the flower.' He found that the nonreversible were interpreted faster, and if he made a sentence anomalous, e.g. 'The flower picked the girl.' they would understand it as if it were said the other way around. This kind of hearing or reading it to make sense occurs with adults as well. The processing of sentences with errors may take a little longer, but they usually are interpreted to make sense.

Kinds of meaning
All meaning is not the same. There are no satisfactory characterizations of meaning.
Following the lead of Henning Anderson, (citation?) I identify at least 3 kinds of meaning.
Conceptual meaning, usually associated with words--This is what most people think of when they think of meaning. 'dog' is a canine, a carnivorous mammal that has been domesticated, etc.
Referential meaning, usually associated with phrases--this is a mapping of a concept with some actual, imaginary, or potential entity in the world, in some mental model, or both. If the speaker utters the phrase 'The dog' she would expect that the hearer would be able to associate that phrase with a particular dog. It might be fictional e.g. Lassie, or real, the speaker's pet. The hearer/reader must be able to identify the entity to understand the utterance.
Propositional meaning, usually associated with sentences--There are several variants of this meaning. It is a claim about a relationship between the entity referred to and something else. 'The dog chased a rabbit.' This utterance makes a claim about an event concerning a particular dog. Different kinds of claims are sometimes marked in the grammar, 'Did the dog chase a rabbit?' and are sometimes marked in the situation.

Speech Acts (John Searle, John Austin)
Language is used for a purpose. A speaker may tell an addressee something, ask for information, or request an action. She may make a promise or threaten her addressee. She may try to be funny. All of these are illocutionary acts. Each of these actions have particular conditions to be met in order for them to be satisfactory speech acts. There are also perlocutionary acts. These are expected actions or processes by the addressee. For example, accepting as true an assertion, answering a question, acting on the request, be entertained. These acts may be specified in the sentences, or implied in the situation.

Discourse processing.
Readers/hearers start tying a text to the domain of the discourse as soon as they have enough information to do so. They identify the referent of the first noun phrase (the topic) when they hear it, if it is already in the discourse (Olson & Filby, 1972; Greenspan & Segal, 1984, Reference and Comprehension, Cognitive Psychology; Segal, 1995, A Cognitive-Phenomenological theory of Fictional Narrative, in Deixis in Narrative). Next, they tie the predicate to the topic in the relevant way based on the illocutionary act and their perlocutionary response to it. If the speaker/writer misleads the hearer/reader by not presenting the appropriate discourse cues, comprehension is slower. The hearer has to make adjustments to comprehend. This is demonstrated in a number of experiments.
    If the discourse concerns objects and events in the environment, the hearer is likely to attend to the referent and evaluate the predicate about it as it is being uttered. A study by Olson & Filby (1972) illustrates this. click here for details
    Early research on the use of negative sentences, such as "The bus is not on time today."  by Wason, (1965) The contexts of plausible denial, Journal of Verbal Learning and Verbal Behavior, 7-11. clearly shows that the grammatical form selected normally depends on focus and expectation. Negative sentences tend to deny expectations of several different sorts.
    If the discourse is expository text, then the primary goal of the reader is to build a model which includes the set of meaning relations among the propositions as best she can. Background knowledge and structure aid this process because the reader can tie the new information into a previously constructed model. If the information seems to be basically disjointed so that it is difficult to build a coherent model, learning it will be difficult and many errors are likely to ensue.
    If the discourse is fictional narrative, i.e. short stories, novels, plays, movies, etc. then mental representation becomes more prominent and other cognitive activities such as what Segal calls a 'Deictic Shift' becomes more obviously operative. The reader/hearer/viewer imagines herself in the world of the story and uses that position to interpret the incoming text. A short paper describing this approach can be found if you click here.