胡壮麟 语言学教程 第六章Language Processing in Mind

胡壮麟 语言学教程 第六章Language Processing in Mind Chapter 6 Language Processing in Mind 6.1 introduction Why study language? ... LANGUAGE is a mirror of the mind in a deep and significant sense. It is a product of human intelligence, created anew in each individual by operations that lie far beyond the reach of will or consciousness. (Chomsky, Reflections on Language, 1975) The study of "language and mind" aims to model the workings of the mind in relation to language. Since structures and connections in the mind are inevitably unobservable, researchers put forward hypotheses based on fragmentary clues. This accounts for the high degree of controversy which surrounds almost all areas of the subject. The label most usually given to the study of "language and mind" is psycholinguistics, a term which is often perceived as being trendy. It has therefore been somewhat overused in recent years, and can be found applied to just about any linguistic topic. PSYCHOLINGUISTICS "proper" can perhaps be glossed as the storage, comprehension, production and acquisition of language in any medium (Spoken or written). Why study psycholinguistics? PSYCHOLINGUISTICS is concerned primarily with investigating the psychological reality of linguistic structures. Less modestly, it sometimes also produces findings which make their own mark on linguistic research, leading to the modification of theoretical ideas. If we view the task of psycholinguistics in the light of the general paradigm of cognitive psychology, it becomes identified with the search for behavioral manifestations of linguistic constructs. As psychology, in its modern guise, uses an experimental methodology, it also follows that such behavioral manifestations will be studied under experimental conditions, and it is reasonable to see the last thirty years or so of psycholinguistics in terms of the construction of ingenious laboratory techniques for pursuing this goal. It is perhaps useful to distinguish psycholinguistics from the PSYCHOLOGY OF LANGUAGE, which deals with more general topics such as the extent to which language shapes thought, and from the PSYCHOLOGY OF COMMUNICATION, which includes non-verbal communication such as gestures and facial expressions. A possible divide within psycholinguistics is of those who style themselves COGNITIVE PSYCHOLINGUISTS as opposed to EXPERIMENTAL PSYCHOLINGUISTS. The form are concerned above all with making inferences about the content of the human mind, Whereas the latter are somewhat more concerned with empirical matters, such as speed of response to a particular word. In practice the two schools often overlap, but extreme supporters of each way of thinking sometimes perceive the gap as being a large one. 6.1.1 Evidence Psycholinguistics attracts adherents from both linguistics and psychology, though these often have somewhat different approaches, particularly with regard to methodology. Linguists tend to favor descriptions of spontaneous speech as their main source of evidence, whereas psychologists mostly prefer experimental studies. This divide highlights the fact that investigators face an unsolvable paradox: the more naturalistic a study, the greater the number of uncontrolled variables; the more rigidly the situation is controlled, the greater the likelihood that the responses obtained will be untypical of real speech situations. Care must therefore be taken to approach topics from different angles, in the hope that the results will coincide. The subjects of psycholinguistic investigation are normal adults and children on the one hand, and aphasics--people with speech disorders--on the other. The primary assumption with regard to aphasic patient that a breakdown in some part of language could lead to an understanding of which components might be independent of others. 6.1.2 Current issues In spite of the varied approaches found in modern psycholinguistics, a number of general trends and crucial issues can be identified. A major point of agreement among various researchers is that the human language system is likely to be "modular", in the sense of being constituted out of a number of separate but interacting components. A considerable amount of recent work has attempted to elucidate this possible insight, although the number and nature of these modules is far from clear. The realization that language organization is likely to be modular has, however, led to a major controversy concerning the integration of the modules, as to whether they remain separate with links between them, or lead to an overall central organizer which contains more abstract representations. For example, it is clear that at some level written and spoken representations of words must be kept separate. One can therefore argue for an approach which contains two separate lexicons, one for written speech, the other for spoken, with links between them. On the other hand, one could suggest that these separate lexicons lead ultimately to an abstract "master-lexicon" in which differences between the various outputs are conflated. The issue is still undecided. More recently, the question of "encapsulation" has become dominant, the extent to which each module works automatically and independently, with its content sealed off from that of other modules (Fodor 1983, 1985). A further problem is the relationship between STRUCTURE and PROCESS. It is generally agreed that the mind is likely to contain certain linguistic structures which are utilized in the course of various "processes", such as comprehending or producing speech. Some researchers have argued that structures and processes are linked only indirectly, others that the connection is a close one. This debate is often phrased in terms of the relationship between a linguist's grammar and a human "grammar", and the extent to which the former has PSYCHOLOGICAL REALITY. Those who believe that the relationship between structure and process is weak tend to accept the idea that a linguist's grammar may have "psychological reality" even though there is no way in which it seems to be directly used in the processing of speech. Chomsky (1980) for example, has argued that any model which represents the "best guess" as to the linguistic structures in the mind must be regarded as "psychologically real" until superseded by a better model, even though it has no relevance for comprehension or production. Other researchers, however, have argued for a closer relationship between structure and process, suggesting that linguists' grammars ought to have a clear relationship to linguistic processing. This controversy is unlikely to be solved in the near future. The three major strands of psycholinguistic research are the COMPREHENSION, PRODUCTION, and ACQUISITION language. That is, (1) How do people use their knowledge of language, and how do they understand what they hear or read? (2) How do they produce messages that others can understand in turn? (3) How language is represented in the mind and how language is acquired? Owing to the space limitation, this chapter will mainly introduce the processes of language comprehension and production. 6.2 Language Comprehension Most of psycholinguistic research and theories are concerned with the comprehension of text, but the assumption is that they are also relevant to SPEECH COMPREHENSION. If we assume that the processes involved in TEXT COMPREHENSION are by and large the same processes as those involved in speech comprehension, then it should follow that those who have good reading skills should also tend to have good speech comprehension skills, whereas those who are poor at reading should also tend to be poor at understanding speech. This does, in fact, seem to be the case. Reading and speech comprehension both involve several processes. It is important for the reader or listener to take account of the grammatical structure and the meaning of what is being presented. Meaning is extracted from text or speech by relating what is presented to information stored in long-term memory. There have been several attempts to explain how our previous knowledge is applied to text and speech, and some of these attempts will be discussed shortly. 6.2.1 Word recognition An initial step in understanding any message is the RECOGNITION OF WORDS. As we have seen, the meaning of a sentence is determined in part from the meanings of the words in it. Additionally, the recognition of a word provides information that determines the syntactic structure of the rest of the sentence. For instance, identifying a word also identifies what category it is, thus determining what range of phrase structures it can occur in. And when we recognize a verb like put, as well as accessing its meaning, we also expect certain types of complements to follow it; that is, it subcategorizes for an object noun phrase and a locative phrase. When we hear the word arrive, we know that it does not take an object. This kind of information is made available to speakers by recognizing a word and accessing the information that they have in their memory concerning it. Identifying words is such an effortless task under most conditions that we don' t realize how difficult it really is; designing a computer system to accomplish this task, for instance, has proven virtually impossible. First recall the, sheer number of words that any given person knows: the average six-year-old knows already about 14,000 words. A different kind of difficulty is presented by ambiguity, where two words with different meanings sound the same. Despite these difficulties, the subjects are able to identify what they hear quite quickly. One source of evidence that this is the case comes from "shadowing experiments" in which people repeat aloud what they have just heard over a tape recorder. Ss can fluently and accurately repeat what they have heard with a lag of considerably less than a second. They cannot do this if what they hear is nonsense or a foreign language. How do we go about recognizing words then? One common sense view that receives a lot of support from experimental evidence is that as soon as people hear speech, they start narrowing down the possible words that they may be hearing. If the first sound that they hear is /s/, that eliminates all words beginning with other sounds; if the next sound is/p/, many other possibilities are eliminated. A word is identified as soon as there is only one possibility left. This account is referred to as the COHORT THEORY and hypothesizes that auditory word recognition begins with the formation of a group of words at the perception of the initial sound and proceeds sound by sound with the cohort of words decreasing as more sounds are perceived. This theory can be expanded to deal with written materials as welt. Several experiments have supported this view of word recognition. One obvious prediction of this model is that if the beginning sound or letter of a word is missing, recognition will be much more difficult, perhaps even impossible. As early as 1900, experiments showed that word recognition is much more impaired by the mispronunciation of the initial letter of a word than by the mispronunciation of the final letter. This suggests that this model is correct; if the end of the word is missing, it can be predicted based on the initial portion, while it is much more difficult to use the end to predict the early part of the word. One of the most important factors that affects word recognition is how frequently the word is used in a given discourse or context. This FREQUENCY EFFECT describes the additional ease with which a word is accessed due to its more frequent usage in the language. For example, words like better or TV occur more often than debtor or mortgage. This effect is not easy to explain assuming the beginning-to-end word recognition approach sketched above. One possible explanation of the frequency effect is that the lexicon is partially organized by frequency rather than simply in terms of the sounds in beginning-to-end order. People also recognize a word faster when they have just heard it or read it than when they have not recently encountered it. Frequent words are likely to have been encountered more recently than infrequent words, so it is possible to explain the frequency effect as a recently effect and reduce the number of separate effects that have to be explained. RECENCY EFFECTS describe the additional ease with which a word is accessed due to its repeated occurrence in the discourse or context. However, again, the recently effect does not gibe with the simple beginning-to-end identification algorithm sketched above. Another factor that is involved in word recognition is COTEXT. People recognize a word more readily when the preceding words provide an appropriate context for it. For example, in the sentence This is the aorta, people are not given any context that helps identify the word aorta. But in the sentence The heart surgeon carefully cut into the wall of the right aorta, many people would find that the cue of the heart surgeon helps them to identify the word more quickly. One mechanism that has been proposed to account for this kind of context effect is a SEMANTIC ASSOCIATION NETWORK. This network represents the relationships between various semantically related words. Word recognition is thought to be faster when other members of the association network are provided in the discourse. It is obvious that the meaning of a word is tied to our understanding and general knowledge of the concept to which it refers. Thus, it is not unreason- able to suppose that hearing the words heart surgeon not only activates the direct meaning of the words heart surgeon, but also makes a number of associated concepts that are more available to the hearer, such as those involved in the physiology of the heart, modern surgical procedures, and so on. These concepts are in turn linked to the words that are used to refer to them. 6.2.2 Lexical ambiguity Much research has centered on how ambiguous words such as bug or rose are understood. There are two main theories: (1) all the meanings associated with the word are accessed, and (2) only one meaning is accessed initially. Support for the first position comes from experiments such as the following. When people are asked to finish a sentence, they take longer when the fragment to be finished contains an ambiguous word than when the ambiguous word is replaced by an unambiguous term, as in the following sentences: ex. 6-1 a. After taking the right turn at the intersection….. ( right is ambiguous: correct vs. rightward) b. After taking the left turn at the intersection… (left is unambiguous) What this delay suggests is that all meanings of ambiguous words are accessed and that time has to be taken to decide among them. However, other experiments suggest that under some circumstances, only one meaning is initially accessed. Two of the effects mentioned above have been shown to be important here: the frequency and context effects. First, if one of the meanings is much more frequent than the other, people tend to assume that the word has the more frequent meaning. The word chair, for example, has at least two meanings--an object to sit on and the head of a department or committee--but the former occurs much more often in speech than the latter, and people often appear to recognize only the more frequent meaning soon after the word chair is presented to them. This seems to suggest that only one meaning is initially considered, at least for words whose various meanings differ markedly in frequency of occurrence. The semantic context effect also plays a significant part in which meaning is the most appropriate. When a word like bug is seen in the context of spy, it is reliably identified as meaning "a listening device"; but in the context of spiders and roaches it is identified as meaning "an insect". 6. 2.3 Syntactic processing Once a word has been identified, it is used to construct a syntactic structure. In some cases this is quite straightforward. Psycholinguists generally assume that the syntactic structure is built as soon as possible rather than waiting to see what the whole string of words is before deciding what structure it has. The reason for this assumption is that people normally already know what the beginning of the sentence means before they hear the whole thing; this must be the case given the common experience we all have of being able to fill in a word that someone else we have been listening to is groping for. The shadowing task mentioned earlier proves the same point. Sometimes when people are shadowing, they mistake what they hear and say a word that is not on the script that they are repeating aloud. This word is generally syntactically and semantically reasonable given the context. If all the words in a sentence are unambiguous or have only one possible category and those categories fit only one phrase structure, it is not hard to understand how people can build the correct structure and interpret it so quickly. For example, in the jealous woman went away, the word the can only be a determiner, woman is a noun, and the combination forms a noun phrase or nominal word group; went is a past tense verb, away is a particle, and the combination of the two makes a verb phrase; finally, the combination of a noun phrase and a verb phrase makes a sentence. From this example there is no reason to think that the process of determining the structure of a sentence is difficult. However, as always, there are complications due to the ambiguity of individual words and to the different possible ways that words can be fit into phrases. Sometimes there is no way to determine which structure and meaning a sentence has. For example, there are two possible structures associated with the sentence The cop saw the spy with the binoculars. The ambiguity lies in how the prepositional phrase with the binoculars fits into the rest of the sentence; it can be a complement of the verb see, in which case it means that the cop employed binoculars in order to see the spy, or it can be part of the noun phrase, in which case it specifies that the spy has binoculars. Some ambiguities are due to the ambiguous category of some of the words in the sentence. In the expression the desert trains, should desert be taken as the subject of the verb trains, or is desert a modifier of the noun trains ? If the sentence continues men to be hardy, the first structure is correct, but if it continues seldom run on time, the latter is appropriate. One interesting phenomenon concerning certain ambiguous sentences is called the "garden path", GARDEN PATH sentences are sentences that are initially interpreted with a different structure than they actually have. It typically takes quite a long time to figure out what the other structure is if the first choice turns out to be incorrect. Sometimes people never figure it out. They have been "led up the garden path," fooled into thinking the sentence has a different structure than it has. Reduced relative clauses quite frequently cause this feeling of having been garden-pathed. For example, The horse raced past the barn fell means "the horse that was raced past the barn fell," but even when this is explained, many people have trouble figuring out how it fits together. As with lexical ambiguity, an important question in sentence processing is how people decide which structure an ambiguous sentence has. The alternatives are that people either consider all possibilities and decide which is the best, or else they use some strategy to decide which structure to consider first. If that structure does not work out, they may reconsider. This garden path phenomenon suggests that for at least some ambiguities, people try one analysis of an ambiguous sequence of words first, and only become aware of the other possibility when the one tried first does not work out. In the example above, people realize that there is something wrong with their initial analysis of the sentence because fell cannot be fit into the structure that they have settled on initially. Several suggestions have been made about how people decide which analysis to try first. One guess is that there is a strong tendency to build as little structure as possible. For example, if the first word in sentence is the, an infinite number of structures could potentially follow; the is a determiner and unambiguously initiates a noun phrase, but that noun phrase can serve as a subject of a sentence, as a possessive phrase modifying the subject noun, as a possessive phrase modifying a possessive noun, etc. Because of recursion, there are an infinite number of possibilities. It would be inefficient for people to assume all these infinite structures until they get some positive evidence for one of them. And if they arbitrarily choose one of the possibilities, they are most likely to choose the simplest. The idea is that people initially construct the simplest (or least complex) syntactic structure when interpreting the structure of sentences. This is called the MINIMAL ATTACHMENT THEORY. There are other types of information that people can use to choose the appropriate analysis for an ambiguous sentence as well. As with words, the semantic interpretation sometimes determines which is the most likely interpretation. Returning to the ambiguity above, it is equally likely that someone can employ binoculars to see a spy or that a spy should have binoculars. But if the word binoculars is replaced with him by Galileo, the great Italian the word revolver, the situation changes. A spy with a revolver is likely enough to be seen, but it isn't possible to see using a revolver. Therefore, the choice between the possibilities can be made based on the pragmatic plausibility of the two interpretations. 6.2.4 Semantics and sentence memory One of the dearest demonstrations that memory representations are not syntactic under certain conditions arises when we present subjects with short stories, each of which contains a key sentence. At some point after the subject hears this sentence. He is presented with a test sentence and has to decide whether it is identical in all respects with any sentence he has heard in the story. The test sentence is con- strutted in such a way as to bear one of a number of relationships to the key sentence and, of course, on any trial subject does not know which sentence in the story is the key sentence. To illustrate, one story which has been used in this sort of study described the invention of the telescope and included ex. 6-2a as its key sentence. ex. 6-2a Galileo, the great Italian scientist, sent a letter about it to him. The last sentence, which could immediately follow the key sentence or follow it after an interval of 60 or 120 syllables, could either be identical with (a), related to (a) via PASSIVE, related to (a) via another syntactic rule not involving any change in meaning, or related to (a) via switching the subject and indirect object NPs--a switch which involves a marked change in meaning. These last three possibilities are illustrated in (b)---(d). b. A letter about it was sent to him by Galileo, the great Italian scientist. c. Galileo, the great Italian scientist, sent him a letter about it. d .He sent a letter about it to Galileo, the great Italian scientist. The results of the study were that if the test sentence was presented immediately after the key sentence, the subjects would recognize any change, syntactic or semantic. After as little as 60 syllables, however, Ss were almost as likely to respond that they had heard (b) or (c) in the story as they were to respond positively to (a) itself, i.e. after this short period they were not capable of recognizing syntactic changes which did not also involve a change of meaning. For (d) subjects' performance was much more accurate and, even after 120 syllables, they were detecting semantic changes with almost 100 percent accuracy. This seems to indicate that the syntactic details of linguistic material are not usually stored for very long and that it is a representation of a sentence's meaning which a subject has available in his memory under normal circumstances. Note that this does not demand that syntactic details have no role in memory. In fact, the results of this study are consistent with such details being available to memory for a short time and it may be that under certain conditions the experimentalist can gain access to this level of representation. It is, perhaps, not insignificant that the present 'study uses materials in structured story form, whereas most of the earlier memory studies we have mentioned use sentences in isolation as stimuli. Rather more interesting work than the above has taken place within the framework of ASSIMILATION THEORY. This is a rather difficult position to define but it differs from approaches using semantic features in its emphasis on the importance of background knowledge in "normal" situations where we might memorize linguistic material. There is evidence to indicate that, for some levels of memory representation, this background knowledge constitutes a necessary condition, i.e. without the knowledge the appropriate representation cannot be set up. This evidence arises when subjects are presented with structured texts and are instructed to comprehend them and remember as much of their content as possible. The texts are chosen to describe unlikely states of affairs but the lexical items and syntactic structures they use are familiar enough. One group of subjects simply hear the text, a second hear the text and are then shown a picture providing an appropriate context for the text and a third group are shown the picture before they hear the text. The results are quite spectacular, with the third group remembering the content of the text much more fully than either of the other two, and this would seem to require that subjects in this group managed to build up a level of representation which they could use in this task. This level of representation was denied by the first two groups, despite the familiarity with the lexical items and syntax of the text. Assimilation Theory has opened up new areas of memory research over the last few years and it seems likely that it will direct investigation for some time in the future. Its major weakness is that it has said little about the details of semantic representations for sentences beyond the fact that such representations must allow the processes briefly described here to take place. In particular, it says nothing about the semantic representation of words and it is self-evident that the understanding process involves "looking up" the meanings of particular words in some sort of psychological dictionary or lexicon. 6.2.5 Basic processes in reading Eye movement. A useful way of investigating some of the processes involved in reading is to study the EYE MOVEMENTS of people reading. While we feel that our eyes move smoothly across a page of text, the reality is quite different. Our eyes actually make a series of rapid movements known as "saccades", and between saccades there are fixation periods lasting for approximately 250 milliseconds. A point towards the beginning of a word is usually fixated, and there is a distance of approximately eight letters or spaces between successive fixations. While most fixations typically move forwards in the text, around 10 or 15 % of them involve the eyes fixating an earlier part of the text than the previous fixation. Of particular importance, information is obtained from the text only during fixations and not at all during saccades. The perceptual span. How much information is extracted from a single fixation? Various methods have been used to measure what is known as the PERCEPTUAL SPAN, which is the range of letters from which useful information is extracted. Not surprisingly, the perceptual span varies depending on factors such as the size of the print, the complexity of the text, and so on. It is typically the case, however, that the perceptual span encompasses about three or four letters to the left of fixation and some fifteen letters to the right of fixation. The opposite pattern is found in readers of Hebrew, who read from right to left. What appears to be happening is that it is more valuable to look ahead in the text rather than to look backwards to words which have already been processed. The fact that the perceptual span covers almost 20 letters means that some of the letters included in it do not fall within the focal region of the eye, which is the area of high acuity. What information is extracted from the area lying outside the fovea? Fairly complex studies have revealed that meaning is not extracted, but that information about the identity of the letters is obtained. The immediacy assumption. At what point is meaning extracted from the words in a text? The reader is supposed to carry out the processes required to understand each word and its relationship to previous words in the sentence as soon as that word is encountered; this is known as the IMMEDIACY ASSUMPTION. Carpenter and Daneman (1981) obtained some evidence consistent with the immediacy assumption. They included the following two sentences in an account of a fishing contest: ex. 6-3 Tomorrow was the annual, one-day fishing contest and fishermen would invade the place. Some of the best bass guitarists in the country would come to this spot. These sentences were written specifically to confuse the subjects. It is natural to interpret bass as referring to a kind of fish, but the following word guitarists makes it clear that the appropriate meaning of bass is its musical one. Most readers fixated an unusually long time on the word guitarists, which suggests that the ambiguity and its resolution were noticed almost immediately. There is other evidence which indicates that the immediacy assumption is over-simplified. The fact that most readers sometimes move their eyes back to earlier parts of a text suggests that the meaning of text is not always extracted immediately. In addition, readers sometimes fixate the same word more than once, which is contrary to the spirit of the immediacy assumption. 6.3 Discourse/text interpretations Discourse serves as a context, affecting sentence and word-level interpretation, tipping the interpretation of what would otherwise be ambiguous words or phrases in a certain direction. For example, some sentences can be interpreted literally (according to standard usage of the words) or figuratively (by deviating from what we understand to be the standard significance of the words for some special meaning or effect). Context can prompt readers to engage in one or the other. When we are trying to understand a sentence, we often make use of information that is not contained directly within the sentence itself. This is known as contextual information, and we can distinguish between two kinds of context: general and specific. GENERAL CONTEXT EFFECTS occur when our general knowledge about the world influences language comprehension. SPECIFIC CONTEXT EFFECTS involve information obtained from earlier parts of a discourse. General context effects occur all the time, because a crucial aspect of language comprehension involves making use of any relevant general knowledge that we possess. This can be demonstrated at an anecdotal level. When the author first visited the United States, he was baffled by the commentaries on baseball games. He could understand each word and sentence at some level, but Full comprehension was impossible because of his imperfect knowledge of the rules of baseball. Specific context effects can operate in a similar fashion, as was shown by John Bransford and Marcia Johnson (1972). They gave their subjects the passage in the panel below: The procedure is actually quite simple. First you arrange items into different groups. Of course one pile may be sufficient depending on how much there is to do. If you have to go somewhere else due to lack of facilities, that is the next step; otherwise, you are pretty well set. It is important not to overdo things. That is, it is better to do too few things at once rather than too many. In the short run this may not seem important but complications can easily arise. A mistake can be expensive as well. At first, the whole procedure will seem complicated. Soon, however, it will become just another facet of life. It is difficult to foresee any end to the necessity for this task in the immediate future, but then, one never can tell. After the procedure is completed one arranges the materials into their appropriate places. Eventually, they will be used once more and the whole cycle will then have to be repeated. However, that is part of life. Subjects who were given this passage on its own found it (as you probably did) difficult to comprehend. Those who were provided with an appropriate context in the form of the title, "Washing clothes", on the other hand, found the same passage reasonably easy to understand. 6.3.1 Schemata and inference drawing It is believed that SCHEMATA, meaning packets of stored knowledge, play an important rote in language processing. The futures of schemata are as follows: (1) Schemata can vary considerably in the information they contain, from the very simple to the very complex. (2) Schemata are frequently organized hierarchically; for example, in addition to a rather general restaurant schema or script, we probably also have more specific restaurant schemata for different kinds of restaurant ( e. g. fast-food places, up-market French restaurants, and so on ). (3) Schemata operate in a top-down or conceptually driven way to facilitate interpretation of environmental stimuli. There are various characteristics of language processing which indicate the key role played by schematic and other stored knowledge. For example, language comprehension frequently requires us to go far beyond the literal meanings of the sentences we read or hear. Essential information is often only implied, so that it is necessary to draw inferences in order to understand fully what is intended. You might think that only rarely do inferences need to be drawn to fill in the gaps in discourse. Consider the following sentence: Three turtles rested on a floating log, and a fish swam beneath them. Most of us would use our stored knowledge of spatial relationships to draw the inference that the fish swam beneath the log as well as beneath the three turtles. This sort of inference is drawn so effortlessly that we are generally unaware that we have drawn an inference at all. How do we know that someone has drawn a particular inference? According to Bransford, Barclay, and Franks (1972), the inferences which people draw are stored in long-term memory along with information about the sentences actually presented. As a result, they will sometimes mistakenly believe on a subsequent memory test that they previously heard or saw an inference. Bransford et al. (1972) presented their subjects with sentences such as the one above about the fish and the turtles. Later on, they were given a test of recognition memory. Subjects were confident that they had previously been presented with sentences which they had actually heard before (i. e. Three turtles rested on a floating log, and a fish swam beneath them ). They were equally confident, however, that they had heard sentences which involved an inference from what they had heard (e. g. Three turtles rested on a floating log, and a fish swam beneath it, although in the actual sentence, the fish swam beneath them ). This indicates that spontaneous spatial inferences are made. No complete theory of what is involved in inference drawing is available. However, schemata or scripts presumably play a part. In discussion of schemata, we believe that some of the errors observed in memory seem to reflect the use of schemata. For example, people made use of a restaurant schema or script to understand a story about Jack eating out at a restaurant. Subsequently, they falsely recognized sentences fitting into the schema but not actually included in the story (e. g. Jack sat down at the table. ). Schemata thus lead people to draw schema-relevant inferences which facilitate comprehension but which may impair memory. An important theoretical issue is whether schematic knowledge is always used at the time of comprehension and storage or whether it is sometimes used at the time of retrieval. So far as inferences are concerned, there is reasonably strong evidence that many of them are drawn during the comprehension process. For example, in a study the length of time that the eyes fixated on the various words in a text was measured. The fixation time on a word (e. g. knife ) was less if the same word had been presented before than if a general word had been used earlier in the text (e. g. knife rather than weapon ). Presumably this happened because it took the subjects some time to work out that a knife was the weapon referred to previously. This long fixation of knife following weapon did not occur, however, if the reference to weapon had included sufficient information for the reader to make the inference that the weapon was indeed a knife. The implication is that the inference was drawn during comprehension of the word weapon. There is also evidence that retrieval processes can be systematically affected by prior knowledge in the form of schemata. Anderson and Pichert (1978) asked subjects to read a story about a house from the perspective of a prospective buyer or of a prospective burglar. After the subjects had recalled the story, they were then asked to recall the story again on the basis of the alternative perspective. The key finding was that subjects recalled more information relevant to the alternative perspective on the second recall than on the first, presumably because the relevant schemata guided the recall process. As one of the subjects reported: "When he gave me the homebuyer perspective, I remembered the end of the story, you know, about the leak in the roof. The first lime through I knew there was an ending, but I couldn't remember what it was. But it just popped into my mind when I thought about the story from the homebuyer perspective" (Anderson and Pichert, 1978:10). 6.3.2 Story structure Our comprehension of and memory for stories are highly "selective'', in the sense that we focus on the central theme of the story rather than on the relatively unimportant details. This was. demonstrated convincingly by Gomulicki (1956). He asked one group of subjects to write a summary of a story that was visible in front of them. He asked a second group to read the story and then to recall it from memory. The third group of subjects were shown the summaries and the recalls, and were generally unable to tell which were which. These findings indicate that what is extracted from a story and then remembered closely resembles a summary in its emphasis on the main theme of that story. Van Dijk and Kintsch (1983) have proposed a theory of story processing. They argued that a story is first of all processed so that the individual propositions (i. e. assertions that may be true or false) are extracted. Evidence that propositions play a major role in sentence comprehension was obtained by Kintsch (1973). They presented sentences which contained approximately the same number of words, but which varied in terms of the number of propositions. For example, the sentence, Cleopatra' s downfall lay in her foolish trust of the fickle political figures of the Roman world, contains twice as many propositions as the sentence, Romulus, the legendary founder of Rome, took the women of the Sabine by force. Kintsch (1973) discovered that reading time for such sentences increased by approximately one second for each additional proposition. According to van Dijk and Kintsch (1983), the propositions of a story enter into a short-term working buffer of limited capacity, which is similar to the working memory system. The working memory is termed as a combined temporary memory and mental work space in which recent stimuli are briefly held, either for rehearsal and recall or for meaningful integration with other knowledge. When the buffer contains a number of propositions, the reader or listener tries to relate them to each other in a coherent fashion. In general terms, subsequent ability to remember the propositions depends on the length of time they spend in the working buffer. Those propositions which are highly relevant to the main theme of a story tend to be stored for a relatively long time in the working buffer. The reason is that such propositions are generally well connected to other propositions in the buffer, -and so retaining them in the buffer facilitates the task of making coherent sense of the story. It follows from these theoretical assumptions that thematic information should be better remembered than non-thematic information. Van Dijk and Kintsch (1983) argued that there were additional processes involved in the task of understanding the gist or MACROSTRUCTURE of a story. More specifically, they claimed that readers or listeners make extensive use of their general knowledge to work out the major theme of a story. This leads to the production of MACROPROPOSITIONS, which are general propositions used to form an overall macrostructure of the story. There are substantial differences in the kinds of knowledge that individual people can bring to bear on story understanding. As a result, the way in which the working buffer is used and the story macrostructure which is formed will also differ from one person to the next. The theory proposed by van Dijk and Kintsch (1983), even though it has enhanced our understanding of story processing, suffers from the limitation that it is not precisely enough formulated to account in detail for such individual differences. 6.4 Language production More is known about language comprehension than about language production. This is perhaps because it is generally easier for an experimenter to exercise control over the comprehension material than to constrain a subject's language production. Furthermore, language production cannot be considered purely from the perspective of a theory of language. LANGUAGE PRODUCTION is very definitely a goal-directed activity, in the sense that people speak and write in order to make friends, influence people, convey information, and so on. The two forms of language production considered in this section are speech production and writing. Speech production has been investigated more thoroughly than writing, but the reasons for this are not clear. 6.4.1 Speech production The usual approach in cognitive psychology is to set the subject a task, and then to assess how accurately or efficiently that task has been performed. This approach doesn't work, however, when we are investigating speech production. If we tell our subjects what we want them to say, then the spontaneity of normal speech is completely lost. If, on the other hand, we leave our subjects free to say whatever they like on a given topic, then we have very little experimental control over what is said. One approach is to ask people to speak on a particular topic, and to make a tape recording of what they say. Another, more useful approach is to ask people to make a collection of the speech errors which they make in everyday speech. At a theoretical level, a promising start has been made by Garrett(1976; 1984). He argued that producing speech is a much more complex matter than it might appear to be from our everyday experience. According to his model, there are altogether five different levels of representation involved in speaking a sentence, and they occur in the following sequence: (1) The message-level representation: this is an abstract, pre-linguistic representation of the idea or ideas that the speaker wants to communicate. (2) The functional-level representation: this is an outline of the proposed utterance having grammatical structure; in other words, the slots for nouns, adjectives, and so on are allocated, but there are no actual words to fill the slots. (3) The positional-level representation: this differs from the functional level representation in that it incorporates the words of the sentence that is to be produced. (4) The phonetic-level representation: of the necessary information about the ways the intended sentence are pronounced. (5) The articulatory-level representation: this is the final representation, and contains a set of instructions for articulating the words in the sentence in the correct order. This complex theory of speech production has not as yet been tested thoroughly. However, there is support for some of its major assumptions. In essence, Garret ( 1984 ) claimed that the speaker engages in reasonably elaborate planning before beginning to speak. One way of testing this notion is to consider the sorts of error that people make while talking. If, for example, sounds or words from the end of a sentence intrude into the early part of a sentence, then this provides evidence for the notion of forward planning. The classic error of this type is the SPOONERISM (or SLIP OF THE TONGUE), where the initial letter or letters of two words are transposed. Other errors also demonstrate the existence of forward planning. An ANTICIPATION ERROR occurs when a word is spoken earlier in the sentence than it should be (e. g. The school is at school. ) A similar type of error is the EXCHANGE ERROR, in which two items within a sentence are swapped (e. g. This is the happiest life of my day. ) In general terms, the demands of speech production are so great that production of speech and planning for the next utterance are difficult to combine satisfactorily. Pauses are used in an attempt to ease these processing demands, but even with pauses, spontaneous speech is usually characterized by a variety of errors. The position is rather different when people have prepared the content (but not the wording) of what they intend to say (e.g. before a public lecture). Such prepared speech exhibits many fewer grammatical and stylistic errors than does spontaneous speech. According to Garrett (1976; 1984 ), speakers decide on the grammatical structure of a proposed utterance in the functional-level representation, and then select the appropriate words to fit into that structure in the subsequent position-level representation. Given this sequence, it would be possible for the grammatical structure of a spoken sentence to be correct even though some of the words were incorrectly positioned. Precisely this is found with MORPHEME-EXCHANGE ERRORS, in which the roots or basic forms of two words are switched leaving the grammatical structure unchanged (e. g. He has already trunked two packs. ) Why is human speech production so prone to error? According to Dell (1986), it is the price we pay for having such a flexible speech- production system. Its flexibility has the great advantage of allowing us to produce novel sentences. Indeed, most speech errors involve novelty, but simply novelty of an unwanted kind. If we had a very rigid speech-production system, it might prevent errors from occurring, but we would suffer the disadvantage of very stereotyped utterances. 6.4.2 Written language Written language differs from spoken language in a number of ways. The hesitations, grammatical errors, and interchanges of words characteristic of spoken language are largely, or entirely, absent from written language. It should therefore come as no surprise to discover that theories of the processes involved in writing differ significantly from those put forward to explain speech production. One of the most detailed theories of the writing process was proposed by Hayes and Flower ( 1986 ). According to them, writing essentially consists of three inter-related processes: ( 1 ) The planning process, which involves producing ideas and arranging them into a writing plan appropriate to the writer ' s goals. ( 2 ) The sentence generation process, which translates writing plan into actual sentences that can be written down. ( 3 ) The revision process, which involves an evaluation what has been written so far; this evaluation can encompass individual words at one extreme or the overall structure of the writing at the other extreme. In general, the processes operate in the order: planning, sentence generation, and revision; however it is quite common for the writer to return to the planning process after several sentences have been generated and revised, especially with longer pieces of writing. Writing plans are obviously much influenced by the relevant knowledge that the writer possesses about the topic to be written about. Another less obvious factor determining the quality of the writing plan is STRATEGIC KNOWLEDGE, which is knowledge of the methods used in constructing a writing plan in order to make it coherent and well-organized. Children often lack such strategic knowledge--they tend to make use of a KNOWLEDGE-TELLING STRATEGY, where they simply write down everything they can think of that is relevant to a topic without organizing the information in any way (Scardamalia and Bereiter, 1987). Skilled writers, in contrast, possess strategic knowledge, which enables them to make use of a KNOWLEDGE-TRANSFORMING STRATEGY. This involves focusing on potential problems within the planning process (e. g. "Are the main points arranged in the most logical order?"). The writing plan constructed during the planning process is generally much shorter and sketchier than the written story or essay itself. Kaufer, Hayes, and Flower (1986) asked writers to write down an outline of their essays corresponding to the writing plan. They discovered that the subsequent essay was always at least eight times longer than the outline. Kaufer et al. (1986) discovered that expert writers produced larger sentence fragments than less skilled writers: 11.2 words versus 7.3 words on average. However, the similarities outweighed the differences. Both groups of writers wrote down approximately 75 % of the sentence fragments they produced in their verbal protocols. They also found that when sentence parts were altered, it was almost always the last part to be produced that was altered. One of the interesting and perhaps surprising findings about the revision process is that expert writers typically spend longer than non-expert writers on revision. There are two main reasons for this. First, expert writers are more skilled at detecting errors that require revision. For example, in a study by Hayes and others (1985), expert writers detected approximately 60% more problems in a writers. Second, expert writers tend to focus on the overall coherence and structure of what has been written, whereas non-expert writers concentrate more on individual words and phrases. It takes much more time to change the structure of a piece of writing than simply to alter some of the words. How adequate is the theoretical approach to writing advocated by Hayes and Flower (1986)? On the positive side, they have been reasonably successful in identifying some of the major differences between strategies used by expert and non-expert writers. This is of value in terms of suggesting ways in which poor writers could improve their writing skills. Probably the major limitation of their approach is the emphasis they place on verbal protocols – i.e. the spoken thoughts of writers engaged in writing. While there may be conscious awareness of many of the process involved in sentence generation and revision, it seems probable that much of the planning process occurs below the conscious level and thus cannot be verbalized. Further Reading Aitchison, J.1990. Language and Mind: Psycholinguistics, in N. E. Collinge ( ed. ) An Encyclopaedia of Language, Routledge, 333-370. Aitchison, J.1998. The Articulate Mammal: An Introduction to Psycholinguistics. Routledge. ndCarrol, D.V. 1994/1999. Psychology of Language ( 2 edition).Monterey, CA: Brooks/Cole Publishing Co Dijkstra, T & De Smelt, K. (Eds.)1996. Computational Psycholinguistics. London: Taylor & Francis Ltd, 3-23 Eysenck, M. W. 1993. Principles of Cognitive Psychology. LEA, Hillsdale, 102-130. Foss, D., & Hakes, D. 1978. Psycholinguistics : An Introduction to the Psychology of Language. Englewood Cliffs, NJ: Prentice-Hall, 61 - 229. Garman, Mi. 1990. Psycholinguistics . Cambridge: Cambridge University Press. Gleason, J.B. 1998. Psycholinguistics (second edition ). Harcourt Brace College Publishers, 1 - 150, 223 - 339 and 409 - 435. Harley, T. 1995. The Psychology of Language: From Data to Theory. East Sussex: Erlbaup (UK) Taylor & Francis, 31- 100, 139 - 174 and 207 - 277. Kess, J. F. 1992. Psycholinguistics: Psychology, Linguistics and the Study of Natural Language. Amsterdam: John Benjamins, 29 - 222. Steinberg, D.D. 1993. An Introduction to Psycholinguistics. London: Longman, 5- 132. 桂诗春，2000。《新编心理语言学》。上海外语教育出版社。