Psycholinguistics James Myers March 26, 2004 Discourse comprehension OVERVIEW: 1. Discourse and discourse coherence 2. Discourse and memory 3. Inferences 4. Schemata ============================================================= 1. Discourse and discourse coherence 1.1 DISCOURSE: any group of sentences combined in a meaningful manner (e.g. stories, lectures, papers, instructions, etc) Discourse coherence: how the sentences "stick together" (i.e. cohere or be cohesive). Two mechanisms to establish coherence: LOCAL structure: relationships between individual sentences in a discourse. GLOBAL structure: the overall framework of a discourse (e.g. the schema) 1.2 Local structure can be detected through many different linguistic tricks that establish cohesion (see Carroll, p. 158). One linguistic cue for discourse coherence: (co)reference 1.2.1 Anaphoric reference: refers back to an antecedent. "Mary laughed at John, and he was pretty upset." ("He" refers back to the antecedent "John") Within a sentence, there are very strict constraints on anaphoric reference (at least in English!) (a) Bill asked John to tell Fred about him. (b) Bill asked John to tell Fred about himself. In (a), "him" can only refer to "Bill" (or some fourth unmentioned person), NOT "John". In (b), "himself" can only refer to "John", or possibly "Fred" -- definitely NOT "Bill." (Why? It has to do with general principles of human language structure that are actually pretty complicated, and vary somewhat from language to language.) Across sentences (i.e. in discourse), the constraints are not as clear: (c) Bill saw his old friend John. He smiled. (d) Bill saw his old friend John. He had gained some weight. In (c), "he" can refer to "Bill" or "John". In (d), "he" can refer to "Bill" or "John", but somehow it makes more sense if "he" refers to "John". 1.2.2 Cataphoric reference: refers to something coming later in the discourse: "This is what we'll do. I'll rob the bank and you drive the getaway car." ("This" refers to the following sentence.) "I didn't see him clearly at first, but when the monster finally walked into the bright light, I screamed." ("him" refers ahead to "monster") 1.3 QUESTION: Do sentence-processing and discourse processing behave differently? Two arguments so far: Anaphoric reference within a sentence is strictly constrained, but across sentences it isn't. Cataphoric reference is not uncommon across sentences or clauses, but it is often disallowed within sentences (e.g. in "He asked John to talk to Bill", "he" cannot refer to "John" or to "Bill".) So maybe sentence comprehension involves automatic, on-line, restricted processes, while discourse processing involves controlled, off-line, unrestricted processes...? This is a controversial issue (see below). 1.4 Basic strategy of forming discourse coherence at the local level: try to link new information with the given (old) information. (Clark and Haviland 1977) 1.4.1 Given (old, assumed) vs. new information is sometimes indicated directly by syntax or morphology. In English, notice the effect of "the" (indicates old information) vs. "a" and "there" (indicate new information): [The man] [lived in New York] OLD NEW [A man lived in New York] NEW There was [a man living in New York] NEW *There was [[the man] living in New York] In Chinese, you can get the same effects with 這 vs. 一、有 Two Chinese devices used to mark the given/new distinction (among many others): Chinese sentences tend to have a clear topic-comment structure, where the topic is usually a noun phrase that provides old information, and the rest of the sentence comments on it: 【大象】，鼻子很長。 【那個人】我真的很討厭。 The given/new distinction can also be marked by particular morphemes, for example whether Chinese speakers choose a general classifier (i.e. 個) versus a specific classifier (e.g. 條, 張, 台, etc). Thus in Mandarin, speakers tend to use a specific classifier for a noun the first time it's mentioned in a discourse (new topic), while later references often use a general classifier or no classifier (Erbaugh 2000). 1.4.2 Three steps in using new and given information: 1. Identify given and new information in current sentence. 2. Find antecedent for the given information in memory. 3. Attach the new information to this spot in memory. Reaction times for sentence comprehension are increased if any one of these steps is difficult. For example, if antecedent is "far away" and thus hard to find, this slows down step 2: Mary is sick. [Mary in foreground] We are worried about her. Mary is sick. [Mary in foreground] John is sick too. [John in foreground] We are worried about her. Another way step 2 can be slowed down is if antecedent is implicit rather than explicit (see discussion below on "inferences"). 2. Discourse and memory. 2.1 Discourse is remembered in terms of abstract propositions, not in terms of syntactic structure. 2.1.1 Kintsch and Keenan (1973): Subjects were asked to read sentences that varied in the number of propositions: "Cleopatra's downfall lay in her foolish trust in the fickle political figures of the Roman world." 8 propositions, including downfall(Cleopatra), lie(downfall,trust), foolish(trust), etc. "Romulus, the legendary founder of Rome, took the women of the Sabine by force." Only 4 propositions, including founder(Romulus,Rome), legendary(founder), etc. Results: sentence with more propositions took longer to read. Obvious (I hope) objection: the number of propositions is confounded with syntactic complexity! So these results could merely be due to syntactic processing, not to the processing of propositions. 2.1.2 McKoon and Ratliff (1980): Subjects read passages (see example in Carroll, p. 167). Afterwards, they were given a verification task: "Is this sentence true or false"? The verification task involved primes and targets, which were always actual sentences taken from the passage. In one case the prime and target were "close" in terms of the SURFACE STRUCTURE of the passage (i.e. near each other in the actual passage), e.g.: "Land ownership was not a problem." "The fur traders were nomads." In the other case the prime and target were "close" in terms of the PROPOSITIONAL STRUCTURE of the passage. "Circulation among the Indians was vital." "The fur traders were nomads." Results: More priming by "propositionally close" primes. This implies that people process discourse in terms of propositions, not in terms of a sequence of sentences. Remember this if you want to learn how to write clearly! 2.2 The role of working memory 2.2.1 The model of Kintsch and van Dijk (1978): Discourse is processed in units of about 6 to 12 propositions (i.e. all that can fit into working memory at one time) Priority in processing is given to recent and high-level propositions: e.g. in the passage discussed above, the highest level is "French settlements" Higher level propositions are recalled better than lower level propositions because they are called into working memory more often. That is, each time a low-level proposition is put into working memory, it is linked to the relevant higher level propositions, which are therefore reactivated. 2.2.2 Implications for individual differences in discourse processing: People can vary in their working memory capacity, and this affects how they process discourse. Daneman and Carpenter (1980): They first gave a task to measure individual variation in working memory capacity. Subjects had to remember the final word of each sentence in a list. Some could remember when there were 5 sentences, some only when there were no more than 4, 3 or 2. Subjects were then asked to identify the antecedent of a pronoun separated from it by 2, 3, 4, 5, 6, or 7 sentences in a story. There was a correlation between subjects' performance on the two tasks, since both depended on the same factor: working memory capacity. This behavior also correlates with scores on the SAT! (Standard Aptitude Test, required to get into American universities). The results support the Kintsch and Van Dijk model, but the SAT effect implies something else as well (in my opinion): there may be no such thing as general "intelligence"; you may be very smart, but you may not do well on the SAT merely if you have a small working memory capacity! 2.2.3 Another theory (not necessarily contradictory to the previous ones): Situational models. These are mental models of the world described by a discourse. They are presumably even less "linguistic" than propositions, since they may involve spatial representations or other very un-language-like kinds of knowledge. Bruder (1995) Issue: Does the "deictic center" play a role in story comprehension? Deictic center: the primary perspective in a story or portion of story, where your mind's eye is located when mentally "viewing" the action of the story. [related to DEIXIS: linguistic "pointing", e.g. "this" vs. "that"] Subjects are asked to read passages, and then given a statement to evaluate for truth or falsehood along a scale of certainty (off-line, no reaction time measurements): 0 = certainly FALSE 7 = certainly TRUE Two kinds of passages: with "came" (i.e. moving towards the deictic center) vs. with "went" (i.e. moving away from the deictic center). [Passages based on the classic short 1947 novel "The Pearl" by John Steinbeck] [Juana and Kino were inside the brush house.] Juana was preparing a fire. She broke little pieces of brush. Now Kino got up and wrapped his blanket about his head and nose and shoulders. He slipped his feet into his sandals and CAME/WENT outside to watch the dawn. Kino squatted down and gathered the blanket ends about his knees. "came": deictic center follows Kino outside. "went": deictic center stays inside with Juana. Test: T or F? "Juana is inside the brush house." Results: Subjects give higher TRUE scores for the "went" passages, implying that the reader's "mental eye" was sensitive to the deictic center! 3. Inferences 3.1 An INFERENCE is a proposition that is not explicitly or overtly stated, but which implied by the propositions that are overty stated in the discourse, often only with help of additional background knowledge in the mind of the listener/reader. This knowledge is of an incredible complexity and depth: "Mary and Joan bought two apples, and Joan ate both of them. Mary was not happy." Possible inferences (written as sentences for clarity): Mary's unhappiness was caused by Joan's action. Mary wanted to eat an apple too. The two apples were supposed to be shared (one for Mary, one for Joan). Mary and Joan are equals (e.g. sisters or friends, not e.g. mother and daughter). ... 3.2 Inferences are stored in memory the same way as propositions derived from explicit sentences: 3.2.1 Bransford, Barclay and Franks (1972) Subjects were presented with one of two kind of sentences: "Three turtles rested ON a floating log, and a fish swam beneath them." [implies that the fish swam beneath log too] "Three turtles rested BESIDE a floating log, and a fish swam beneath them." [no such implication] Memory quiz: Subjects who had received the "on" sentence were asked which of the following they had gotten earlier: "Three turtles rested on a floating log, and a fish swam beneath THEM/IT." Likewise for subjects who had received the "beside" sentence: "Three turtles rested beside a floating log, and a fish swam beneath THEM/IT." Results: "On" subjects more likely to falsely recall "it" than the "beside" subjects. Interpretation of results: "On" subjects were storing the following implicit proposition in memory just as if it had been presented explicitly: "The fish swam beneath the floating log." (Note: the memory quiz was a surprise, and was given long after the memory for the sentence forms had begun to fade.) 3.2.2 Kintsch (1974) Subjects were presented with stories that either presented an implication explicitly, or where the implication remained only implicit. "Explicit inference" story: "A carelessly discarded burning cigarette started a fire." "The fire destroyed many acres of virgin forest." Therefore: "A discarded cigarette started a fire." [The reader is explicitly told this.] "Implicit inference" story: "A burning cigarette was carelessly discarded." "The fire destroyed many acres of virgin forest." Therefore: "A discarded cigarette started a fire." [not explicit, but only inferred from combination of both sentences] Task: verification task involving Inference sentence; RT was measured. Results: When tested immediately after story, RT was faster for Explicit inference story. When tested 15 minutes after story, RT was the same for Explicit and Implicit stories. Interpretation of results: At first subjects still have syntactic form of story sentences available, and so the sentences in the "Explicit inference" story help them make decisions about propositions quickly. After the memory for syntax fades, all that's left is propositions. Then the inferred propositions are treated the same way as propositions derived from actual sentences in the story. 3.3 What sort of inferences are drawn? Only ones that are NECESSARY to make the discourse understandable and which are EASY to draw from explicit statements or general knowledge (McKoon and Ratliff 1992). That is, are inference "restricted"? 3.3.1 The sentence-comprehension gap-priming studies (last time) show a kind of on-line inference of a very restricted sort: as soon as we hear a verb, we know immediately what all the arguments are for that verb. 3.3.2 But what about more truly propositional inferences? They ARE made, as the above memory studies show, but WHEN? Are they made immediately, on-line, automatically? Or are they made more slowly, off-line, using semi-conscious "controlled" processes? 3.3.3 Trabasso and Suh (1993) think that on-line inference- making is NOT restricted in this sense; Perfetti (1993) thinks that it probably IS. 4. Schemata (plural of schema) We've focussed mainly on LOCAL structure. How does GLOBAL structure help in the comprehension of discourse? 4.1 A SCHEMA is "a structure in semantic memory that specifies the general or expected arrangement of a body of information" (Carroll, p. 175) This concept is not only used for studying discourse comprehension, but any situation where people are absorbing complex information. 4.2 Bartlett (1932): A truly classic experiment! Issue: How do people remember a story if they don't have a familiar schema to follow? British subjects read an Eskimo folk story; Bartlett cataloged the kinds of changes they made when they recalled the story later. Original version: (maybe copyrighted, but I'm not the only one who's posted it on the web!!) Title: The War of the Ghosts One night two young men from Egulac went down to the river to hunt seals, and while they were there it became foggy and calm. Then they heard war-cries, and they thought: "Maybe this is a war-party." They escaped to the shore, and hid behind a log. Now canoes came up, and they heard the noise of paddles, and saw one canoe coming up to them. There were five men in the canoe, and they said: "What do you think? We wish to take you along. We are going up the river to make war on the people." One of the young men said: "I have no arrows." "Arrows are in the canoe," they said. "I will not go along. I might be killed. My relatives do not know where I have gone. But you," he said, turning to the other, "may go with them." So one of the young men went, but the other returned home. And the warriors went on up the river to a town on the other side of Kalama. The people came down to the water, and they began to fight, and many were killed. But presently the young man heard one of the warriors say: "Quick, let us go home: that Indian has been hit." Now he thought: "Oh, they are ghosts." He did not feel sick, but they said he had been shot. So the canoes went back to Egulac, and the young man went ashore to his house, and made a fire. And he told everybody and said: "Behold I accompanied the ghosts, and we went to fight. Many of our fellows were killed, and many of those who attacked us were killed. They said I was hit, and I did not feel sick." He told it all, and then he became quiet. When the sun rose he fell down. Something black came out of his mouth. His face became contorted. The people jumped up and cried. He was dead. Version as remembered 2.5 years later by one British subject: Title: [none] Some warriors went to wage war against the ghosts. They fought all day and one of their number was wounded. They returned home in the evening, bearing their sick comrade. As the day drew to a close, he became rapidly worse and the villagers came round him. At sunset he sighed: something black came out of his mouth. He was dead. Some of the results: Subjects dropped information that didn't fit their expectations (e.g. of how a story is supposed to go): In remembered version, there is only one warrior (irrelevant character dropped) Information sometimes changed to make the story more consistent with British-style stories or expectations: In revision, the warrior fought AGAINST the ghosts, rather than WITH them as in original. In revision, wounded warrior was carried rather than walking himself. In revision, death occurs at sunset, not sunrise. 4.3 Dooling & Lachman (1971), Sulin & Dooling (1974): Issue: Can a schema (global structure) help with the understanding of discourse that is complex in its local structure? Passages written with low-frequency words and uncommon syntax are much easier to understand if we know what the schema is. [passage in Carroll, p. 175] 4.4 "Grammars" for schemata contain rules and structures that are analogous to syntax, but instead have units like "episode", "complex reaction", "goal path" etc. 4.4.1 Evidence for the basic unit, the EPISODE: a story unit that contains a beginning, a development, and an end: Black & Bower (1979): Episodes are recalled as whole chunks (i.e. subjects don't recall just the beginning and forget the end). The length of one episode does not affect the recall of another (i.e. each episode is processed separately in working memory). Haberlandt, Berian and Sandson (1980): Stories were presented sentence by sentence on a computer screen. Subjects took longer to read sentences that complete old episodes or start new ones compared to all other sentences. 4.4.2 Do "story grammars" have universal elements? Mandler, Scribner, Cole and DeForest (1980): American kids and adults and Liberian (African) kids and adults all show the same patterns of recall for the major parts of a story, in spite of differences in experience, education and cultural expectations. REFERENCES Bartlett, F. C. (1932). Remembering. Cambridge University Press. Black, J. B., & Bower, G. H. (1979). Episodes as chunks in narrative memory. Journal of Verbal Learning and Verbal Behavior, 18, 309-318. Bower, G. H., & Morrow, D. G. (1990). Mental models in narrative comprehension. Science, 247, 44-48. Bransford, J. D., Barclay, J. R., & Franks, J. J. (1972). Sentence memory: a constructive versus interpretive approach. Cognitive Psychology, 3, 193-209. Bruder, G. A. (1995). Psychological evidence that linguistic devices are used by readers to understand spatial deixis in narrative text. In J. F. Duchan, G. A. Bruder, & L. E. Hewitt (Eds.) Deixis in narrative: A cognitive science perspective (pp. 243-260). Lawrence Erlbaum. Clark, H. H., & Haviland, S. E. (1977). Comprehension and the given- new contract. In R. O. Freedle (Ed.) Discourse Production and Comprehension (pp. 1-40). Ablex. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450-466. Dooling, D. J., & Lachman, R. (1971). Effects of comprehension on retention of prose. Journal of Experimental Psychology, 88, 216-222. Erbaugh, M. (2000). Classifiers are for specification: complementary functions for sortal and general classifiers in Cantonese and Mandarin. Proceedings of the 33rd meeting of the International Conference on Sino-Tibetan Languages and Linguistics, Bangkok, October 2-6. Haberlandt, K., Berian, C., & Sandson, J. (1980). The episode schema in story processing. Journal of Verbal Learning and Verbal Behavior, 19, 635-650. Kintsch, W. (1974). The representation of meaning in memory. Erlbaum. Kintsch, W., & Keenan, J. (1973). Reading rate and retention as a function of the number of propositions in the base structure of sentences. Cognitive Psychology, 5, 257-274. Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychology Review, 85, 363-394. Mandler, J. M., Scribner, S., Cole, M., & DeForest, M. (1980). Cross- cultural invariance in story recall. Child Development, 51, 19-26. McKoon, G., & Ratliff, R. (1980). Priming in item recognition: the organization of propositions in memory for text. Journal of Verbal Learning and Verbal Behavior, 19, 369-386. McKoon, G., & Ratliff, R. (1992). Inference during reading. Psychological Review, 99, 440-466. Perfetti, C. A. (1993). Why inferences might be restricted. Discourse Processes, 16, 181-192. Sulin, R. A., & Dooling, D. J. (1974). Intrusion of a thematic idea in retention of prose. Journal of Experimental Psychology 103, 255-262. Trabasso, T., & Suh, S. (1993). Understanding text: Achieving explanatory coherence through on-line inferences and mental operations in working memory. Discourse Processes, 16, 3-34.