• 5.1 The Kintsch and van Dijk model for discourse processing
• 5.2 Text linguistic methods in translation and interpretation research
• 5.2.1 Description of argumentative texts (Tirkkonen-Condit)
• 5.2.1.1 Illocutionary and interactional structure
• 5.2.1.1.1 Speech acts and illocutions
• 5.2.1.2 Problem-solution analysis
• 5.2.1.3 Macrostructure analysis
• 5.2.1.4 Textual analysis of an interpreted event: a dialogic approach
• 5.2.2 Inadequacy in translation (Vehmas-Lehto)
• 5.2.3 Shifts in cohesive elements in simultaneous interpretation (Shlesinger)
• 5.2.4 Component processes in simultaneous interpreting (Dillinger)
• 5.3 Applications of the Kintch and van Dijk model in simultaneous interpreting research
• 5.3.1 Mackintosh (1985)
• 5.3.2 Lambert (1988)

5 Text linguistic models
5.1 The Kintsch and van Dijk model for discourse processing	In 1978, Walter Kintsch and Teun van Dijk launched a theoretical framework for the study of the ability of language users to partially reproduce and summarise previously acquired information from discourse. This theory has subsequently evolved into a dynamic, process-oriented, "strategic" model of discourse comprehension and production (van Dijk & Kintsch 1983).  The theory presented in 1978 consists of several sub-theories (van Dijk & Kintsch 1978:67): (i) a theory of discourse, consisting of a) a grammar of discourse, with at least  a theory of semantic representations (propositions) for sentences and sequences of sentences (micro-structure); a theory of semantic representations for global discourse structures (macro-structures); a theory relating micro-structures with macro-structures. b) a more general theory of (non-linguistic) discourse structures, with specific theories for different kind of discourse (ii) a theory or model of discourse structure processing, in particular of semantic information, i.e. for comprehension/interpretation, storage in memory, memory transformations, retrieval, and (re-)production and use/application. (iii) a more general theory for complex cognitive information processing, in which the ability to process discourse is related to our ability to perceive/interpret and memorise complex events and actions after visual input, and to plan or organise and execute complex actions, both bodily and mental (reasoning, problem solving). The semantic structure of discourse is the formal reconstruction of the "information" or "content" of a discourse (van Dijk & Kintsch 1978:67). At the micro-level the semantics assigns sequences of propositions to the sequence of sentences of the discourse. Propositions combine in compound propositions and sequences of propositions, which are parities connected. Connection conditions are based on relations between facts and relative to a topic of discourse (van Dijk 1977). Connection is thus a specific kind of coherence, defined over sequences of propositions, not only in terms of relations between facts and relative to a topic of discourse, but also in terms of intensional and extensional relations between "parts" of propositions (quantifiers, predicates, arguments, etc.) (van Dijk & Kintsch 1978:68).  The set of factors (knowledge, beliefs, opinions, wishes, attitudes, or tasks) that in a particular context of action or discourse processing influences macrostructures is called the cognitive set of a language user or participant (van Dijk 1980). The theory distinguishes between an implicit and an explicit text base underlying discourse. The first one is the one actually expressed in discourse, whereas the latter is a theoretical construction, containing also those postulated propositions which are necessary to establish coherence. These interpolated propositions are those which are "presupposed" by the propositions actually expressed in the discourse.  According to the Kintsch and van Dijk model, the surface structure of a discourse is interpreted as a set of micropropositions. Some of the propositions are present in the surface structure and the others are inferred on the basis of prior knowledge, stored in long-term memory. The micropropositions are then processed by the working memory in order to establish coherence with the propositions already stored in short-term memory, i.e. the previous segment that has been processed. Short-term memory acts as a buffer while the working memory searches for argument overlap between incoming propositions are those already stored. If there is no overlap, the working memory searches long-term memory. If an overlap is established, directly or through inferencing, the proposition in working memory enters short-term memory.  The global meaning of a discourse is represented by semantic macrostructures. Since these will be represented as propositions, semantic mappings, called macrorules are used to relate microstructures with macrostructures. Their function is to reduce and organise information, i.e. they delete and combine sequences of propositions. Macrorules are entailed by the sequence of propositions in the discourse, and due to this recursive nature, macrorules generate not only one, but several macro-structures at increasingly more global levels of semantic representation. No proposition may be deleted which is a presupposition for a subsequent (macro)proposition in the discourse. Microstructures are processed into macrostructures by application of the following macrorules (Kintsch & van Dijk 1978; Renkema 1993): Macrorule 1: Deletion Of a sequence of propositions we may delete all those denoting an accidental property of a discourse referent (NB the general constraint: if not necessary for the interpretation of following propositions). (1) A girl in a yellow dress passed by. 1. A girl passed by. 2. She was wearing a dress. 3. The dress was yellow. Propositions 2. and 3. can be eliminated. Actually, the deletion rule can be used reversely, as a selection rule for those propositions that are necessary for the interpretation of other propositions: (2) John is sick today. He will not be going to the meeting. The microproposition "John is sick" is relevant for the interpretation of the following sentence. But if the theme of John’s illness is not continued, then this proposition is irrelevant at the macrolevel and can be deleted. Macrorule 2: Generalisation Of a sequence of propositions we may substitute any subsequence by a proposition defining the immediate superconcept of the micropropositions. (3) Mary was drawing a picture. Sally was jumping rope and Daniel was building something with Lego blocks. 1. The children were playing. Specific predicates and arguments in a series of propositions are replaced by more general terms so that one propositions suffices. Macrorule 3: Construction Of a sequence of propositions we may substitute each subsequence by a proposition if they denote normal conditions, components or consequences of the macroproposition substituting them.  (4) John went to the station. He bought a ticket, started running when he saw what time it was and was forced to conclude that his watch was wrong when he reached the platform. 1. John missed the train. Note that neither "train" nor "missed" are mentioned in (4). The proposition is constructed on the basis of general knowledge. In Macro-rules 1 and 2 the information is irrecoverably lost. In rule 3 information is partly recoverable, inductively, by general knowledge of postulates and frame knowledge concerning normal conditions, components and consequences.  Superstructures are conventional schemas which provide the global form for the macrostructural content of discourse. In other words, macrostructures deal with the content and superstructure with the form. Renkema (1993) points out that the discourse form stands above the content in some sense. The superstructure of a lecture or of a scientific article are two examples of discourse forms that can be used with a specific content.  Kintsch and van Dijk have subsequently developed their model from a primarily structural/hierarchical one to a more dynamic, process-oriented on-line model which they refer to as strategic (van Dijk & Kintsch 1983). They now assume, inter alia, that understanding takes place successively, while incoming data is processed at once and not after all the data has been received.
5.2 Text linguistic methods in translation and interpretation research	Text linguistic approaches and models have been used to some extent in research on translation and interpretation. Two examples from the area of Finnish are Tirkkonen-Condit’s studies of translations from English into Finnish (Tirkkonen-Condit 1985 and 1986) and Vehmas-Lehto’s studies of translations of Russian newspaper articles into Finnish (Vehmas-Lehto 1989).  Shlesinger (1995) has studied shifts in cohesive elements in English–Hebrew simultaneous interpretation. Dillinger (1989) has investigated, with the use of text-structure variables, the component processes specific to simultaneous interpreting and common to interpreting and listening.
5.2.1 Description of argumentative texts (Tirkkonen-Condit)	Tirkkonen-Condit has developed a method for describing argumentative texts using text linguistic tools. The method contains problem-solution (PS) analysis, interactional and illocutionary (I&I) analysis, and macrostructure analysis. The purpose of the study was to describe two authentic texts in order to develop a method for the description of argumentative text structure in general, and at the same time contribute to the battery of text type criteria, and to shed light on text comprehension and interpretation, and, ultimately, translation. The need to describe the structure of texts derives from problems in translator training, especially in the area of argumentative texts, which according to Tirkkonen-Condit have turned out to be the most difficult to translate of all factual prose text types. The problems are not related to linguistic proficiency, but rather to the comprehension and interpretation of this kind of texts in general. In shedding light on aspects of comprehension, Tirkkonen-Condit’s study also aims at developing tools for translation assessment and marking, and for the selection of text extracts for translation exercises and examinations. In a separate study (1986), Tirkkonen-Condit implements her ideas on translation assessment, using the text linguistic tools developed in the previous study. The text description used by Tirkkonen-Condit (1985) is a combination of three models for discourse analysis: 5.2.1.1 Illocutionary and interactional structure The first is an analysis of the illocutionary and interactional (I & I) structure of the argumentative text (1985:42 ff.). This analysis reveals the hierarchical structure of the text (general vs. particular, superordinate vs. subordinate) as well as interactional aspects, e.g. the illocutions prevailing in sentences and groups of sentences. 5.2.1.1.1 Speech acts and illocutions According to Austin (1976) all expressions of language must be viewed as acts. There are three kinds of action in each utterance: 1. locution, the physical act of producing an utterance; 2. illocution, the act which is committed by the utterance; and 3. perlocution, the act of producing an effect through locution and illocution. In speech act theory, illocution is the main focus of attention. Certain minimum requirements must be met if an illocution is to be successful. Searle (1969) has defined four "felicity conditions" (see below) which illocutions must meet.  There are certain requirements which the production of a form, i.e. the locution, must meet to ensure that the illocution takes place. This illocution, then, serves as a prerequisite for the achievement of the intended perlocution. Here follows an example in the form of an interrogative, quoted from Renkema (1993:26). Can you stop by in a minute? Why is this interrogative generally interpreted as a request? A request can be identified by the following felicity conditions: a. the propositional content The content must refer to a future act, X, which is to be carried out by the addressee. b. the preparatory condition (circumstances that are essential for the uptake of an illocution as the intended illocution) 1. The addressee is capable of executing X and the speaker believes that the addressee is capable of doing it. 2. It is obvious to both conversational participants that the addressee will not perform the act without being asked. c. the sincerity condition The speaker actually wants the addressee to do what had been requested. d. the essential condition (the condition that separates the illocution in question from other illocutions) The utterance serves as an attempt to persuade the addressee to execute X. On the basis of these rules, the interrogative "Can you stop by in a minute?" does possess the illocutionary intent of a request. But it does not explain why it must be interpreted as an order when it is uttered by a supervisor to a subordinate. In this case the situation is not self-explanatory and a knowledge of the surrounding environment is required (Renkema 1993:26). In this context Renkema (1993:28) points out: The analysis of illocutions makes it clear that in the research into the relationship between form and function, form by itself cannot provide a definitive answer. Clearly, other factors, such as the co-operative principle[7] [of Grice] and knowledge of the world (...) will have to be taken into account as well. Illocutions can function as text type markers in the following way (Tirkkonen-Condit 1985:150):   Text type Dominant illocution Subsidiary illocution Descriptive Narrative Expository Argumentative Instructive Statement " " Assertion Directive - Directive - Directive - Figure 5-1 Illocutions as text type markers 5.2.1.2 Problem-solution analysis The second mode of description is problem-solution (PS) analysis which aims at describing the superstructure of the text (cf. van Dijk & Kintsch op.cit.). PS analysis is part of the I & I analysis (see section 5.2.1.1). The text is described as a sequence of minitexts composed of situation, problem, solutions, and evaluation. By using information from the I & I analysis it is possible to detect the hierarchical and interactional relations between the minitexts. It is possible to show how minitexts can be embedded in other minitexts and also show the function of one minitext in relation to another. The PS structure of an argumentative text can be illustrated as a Chinese box diagram, which gives hierarchy information in that superordinate sequences literally include the subordinate sequences. In the following diagram (after Tirkkonen-Condit 1985:126), we see that a minitext with the components situation, problem and solution has three minitexts subordinated to its problem component. The three minitexts are composed of situation + problem, situation + problem, and situation + problem + solution, respectively.       Sit Pr Sit Pr Sit Pr Sol     Sit P r o b l e m Sol Figure 5-2 PS structure of an argumentative text 5.2.1.3 Macrostructure analysis The third mode of description is the macrostructure analysis which reveals the semantic structure ("macrostructure" in van Dijk’s terms) of the text. The macrostructure analysis derives information from the I & I and PS analyses and rearranges it in such a way as to turn out summaries of the text with varying degree of specificity. The summaries are the concrete representatives of the levels of macrostructure. The macrostructure analysis makes use of the hierarchical distinctions from the I & I analysis. It also uses information from the PS analysis, in that the summaries manifest four types of macropropositions which are relatable to the PS components of situation, problem, solution, and evaluation. 5.2.1.4 Textual analysis of an interpreted event: a dialogic approach Seen from the point of view of dialogicity, a speech can be seen as a sequence of speech acts produced by the speaker in answer to imaginary questions of the audience (cf. Tirkkonen-Condit 1985:47). Of course at conferences the audience is often allowed to ask questions at some stage or another during or after the speech, but certain speech situations or types of speech do not allow such interruption. In the following table we use the format used by Tirkkonen-Condit (1985:48) to show how the dialogical approach to the text helps in mapping the structure of the text in terms of I & I and PS description. The text is from the first minitext (which could be called "Setting the scene") of an inaugural speech taken from our corpus (see appendix 1).   The dialogue I & I description PS description Speaker and imaginary audience (in brackets) Interactional role Illocutionary value   Speaker: I feel privileged to be here (sect. 2, 3) Situation Statement Situation component (Audience: Can you elaborate on this?)       Speaker: I'm not worthy to stand in the pulpit (4) Negative evaluation Assertion Problem component (Audience: On what grounds are you asserting this?)       Speaker: I'm here only because of my readiness to serve science (sect. 5) Justification Statement  - " - (Audience: What should be done about it?)       Speaker: Best place would be in the middle (sect. 6) Solution Directive: recommendation  Solution component Figure 5-3 Implicit dialogue in text (example from our corpus)
5.2.2 Inadequacy in translation (Vehmas-Lehto)	Vehmas-Lehto (1989) has studied Finnish translations of Russian newspaper articles from the aspect of adequacy as an inherent part of translation quality. Adequacy is defined as the compliance of a translation with the norms of the target language, and, to a certain extent, with those of the target culture The study is based on the assumption that in order to be adequate, Finnish translations of Russian journalistic texts should resemble the corresponding Finnish functional style, i.e. Finnish journalistic language (cf. the textuality standard "acceptability", section 3.1).[8] The text corpus used in Vehmas-Lehto's study consists of Russian journalistic texts, their published translations, alternative translations serving as suggestions for improvement, and authentic Finnish journalistic texts used as material for comparison. Vehmas-Lehto’s experiments gave the following results: 1) the published translations are usually identifiable as translations, primarily because of the general impression they make of clumsiness and obscurity; 2) the published translations are experienced as unpleasant, uninteresting, and apt to arouse suspicion; 3) there is an (unacceptable) abundance of emotive elements in the vocabulary;  4) the published translations are difficult to comprehend. Textual comparisons revealed a number of deviations from the usage and recommendable norms of Finnish journalistic language. These errors are basically due to interference from the source texts. They are mostly quantitative: e.g. an abundance of coordinated constructions, emotive words, and clichés, a high average length of sentences, clauses, and noun phrases, a high frequency of nouns and adjectives, and a low frequency of connectives.
5.2.3 Shifts in cohesive elements in simultaneous interpretation (Shlesinger)	Shlesinger (1995) has examined the number and type of shifts in cohesive elements of an English-language text undergoing simultaneous interpretation into Hebrew. The productions of 13 advanced interpreting students working from an 11-minute impromptu speech were analysed, and the results revealed a regular occurrence of shifts in all types of cohesive devices, particularly those perceived by the interpreter as nonessential, the most common shift-type being complete omission. Shifts occurred with higher frequency at the beginning of texts. Omission/error rates decreased when interpreters benefited from prior exposure to the source text.[9] Shlesinger (1995) points out that cohesive devices serve a crucial function in text interpretation in that they define links and relationships between primary textual elements. Failure to reproduce these links in a translation can significantly alter text reception and meaning. Three intrinsic constraints have appreciable impact on an interpreter's ability to convey true meaning: (1) the speed of source-text delivery; (2) text linearity, which forces interpreters to work on smaller, incomplete language units; (3) assumptions by the speaker as to the level of subject knowledge available to the audience and/or the interpreter.
5.2.4 Component processes in simultaneous interpreting (Dillinger)	Dillinger (1989) has investigated the component processes specific to simultaneous interpreting and common to interpreting and listening. Experienced conference interpreters and inexperienced bilinguals performed simultaneous interpreting from English into French and then gave a free recall immediately afterwards. A comparison group of bilinguals performed a simple listening task with the same materials.  The texts were on an unfamiliar topic (positron emission tomography) and differed only with respect to frame type. Experience showed a main effect on interpreting measures, (experienced interpreters performed more accurately), and interacted with text-structure variables that indexed proposition generation, but did not affect recall. Task did not have a main effect on recall and interacted weakly with text-structure variables. Text and Text-structure variables had very strong effects both for the interpreting and the recall measures. The results were viewed as evidence that interpreting involves the same component processes as normal listening comprehension rather than constituting a specialised comprehension skill. Analyses of text-structure variables provided evidence for influence of high-level conceptual processing and other component processes both on line and off line. Since there was no evidence that interpreting interfered with comprehension, the qualitative on-line measures possible in the interpreting task appear to be generalisable to comprehension under more usual circumstances.  However, as Dillinger himself stresses (Dillinger 1989:89) it would be misleading to draw the conclusion from his study that there are no differences at all between expert and novice interpreters. Dillinger's study concentrates on the comprehension processes, and expert interpreters may very well differ from novices with respect to their production processes, which Dillinger has not studied: It is possible that experienced interpreters will show more independence in their production; that is, the novices will tend to follow the surface features of the original, whereas the experts will produce target-language texts whose formal features are nearly independent of those of the original. The present study suggests precisely that this difference would not be due to problems in comprehension, but to differences in production ability. (Dillinger 1989:89) Dillinger also points out (1989:88) that his results refer to simultaneous interpreting of prepared texts in conference settings, and may not be generalisable to interpreting more spontaneous dialogue or debates. Since texts of the latter type are generally less explicit and less predictable, they make greater demands on prior knowledge and inference generation. He cites Frederiksen (1989) who argues that the processing of different text types is independent of general comprehension skill. It is therefore possible that an interpreter may work well in the booth with well-prepared texts, but not perform so well with conversational dialogue, or vice versa.
5.3 Applications of the Kintch and van Dijk model in simultaneous interpreting research
5.3.1 Mackintosh (1985)	Mackintosh (1985) has made an attempt to verify the applicability of the Kintsch and van Dijk model to conference interpreting. Mackintosh distinguishes between consecutive and simultaneous interpreting in the possible applicability of the model. In consecutive interpreting, interpreters notes down the essential features of the message and its structure. Mackintosh’s contention is that this schematic notation of the semantic features of the discourse is the result of the interpreter’s application of the macrorules to the micropropositions of the original message, and that the interpreter’s notes reproduce the resultant macropropositions. When reconstituting the message in the target language, the interpreter applies the macrorules once again, but this time in the inverse direction, in order to derive the micropropositions.  In simultaneous interpretation, the process is more complicated. When the interpreter hears an incoming segment of discourse, s/he starts processing it phonologically and semantically in working memory. At the same time, the previous segment is produced in the target language after being checked for relevance and coherence and stored in short-term memory for matching with subsequent incoming segments. (The number of micropropositions that a segment contains varies according to the interpreter’s processing strategy, the microstructures of the discourse etc.) This implies that the processing load of a simultaneous interpreter is greater that that of a consecutive interpreter.  Mackintosh (1985) claims that observation of the macrostructural processing in simultaneous interpreting is not directly observable; in consecutive the process of note-taking and message-analysis can be described in terms of macrostructures. Mackintosh suggests that a recall test after interpreting would give evidence about the formation of macrostructures and macropropositions. If scores in the interpreting protocol correlate with scores in the recall protocol, message integrity in simultaneous interpreting could be seen as a function of macrostructural processing. The schemata within which the interpreter construes meaning play an important part in organising the application of the macrorules and restricting the lexicosemantic choices the interpreter has to make. (Mackintosh 1985:40). To test the model, Mackintosh analysed protocols from a consecutive relay experiment and a similar simultaneous experiment. In addition to interpreter groups who interpreted between two languages, a control group relayed from English into English. All subjects interpreted into their A language except one in each group who interpreted into their B language.  The results of these studies show that the standard of L1 suffers under some conditions. Assuming that the extent to which the level of L1 deviates from the standard level is an indication of how demanding the task is for the interpreter, then self-cued recall from L1 into L1 at the interpreter’s own pace, i.e. consecutive from English into English, was the least demanding. When the interpreter works in two languages, there are many more departures from standard English. These show in consecutive as awkward formulations, and in simultaneous as clumsy expressions and grammatical errors. The protocols indicate that simultaneous interpreting imposes the heaviest processing load, on the basis of departures from standard English.  In the simultaneous experiment, most of the micropropositions of the original are present in the target language text. An interesting indication that simultaneous interpreters also process the discourse according to the model is "soit plus de 13 milliards de centimes" which is incorrectly rendered by "which is about 13 times more than they are getting at the moment". This error can be explained by the fact that the interpreter had to explain an unknown concept for which she had to find a meaningful equivalent; in searching for argument overlap, under a severe time constraint, coherence was considered of overriding importance, and the statement of the interpreter is in fact coherent with the overall schema of the text and its macrostructures. The need to operate three cognitive sets simultaneously (the one the interpreter assumes for the speaker, the interpreter’s own, and the hypothetical set of the listeners) can increase the processing load to the extent that part of the message is forfeited. Decisions about what is not to be forfeited are informed by the macrostructures of the text (Mackintosh 1985:42).
5.3.2 Lambert (1988)	Lambert (1988) has studied the recall of interpreters of four texts after simultaneous interpretation of the texts, consecutive interpreting and shadowing, and a straightforward listening task. The test was divided into two parts, one recall test, and following that, three recognition test of lexical, semantic, and syntactic recognition. For recall, the input was in subjects’ "passive" language, with recall in mother tongue (L2 into L1) and for recognition, both input and recognition was in subjects’ "passive" language (L2 into L2). The text passages were broken down into a structured list of propositions following the methodology proposed by Kintsch and van Dijk (1978). (In the context of conference interpreting, "passive" languages are the interpreters' B or C languages; the "active" language, A language, is usually the mother tongue.) By examining and comparing the amount and quality of retention following each processing type, it was hoped to gain a better understanding of what is meant by depth of processing, how deeply each type of message input is processed, and which type requires the greatest or the least amount of effort and attention on the part of the interpreter. By weighing the retention scores, it appeared that deeper processing of incoming material occurs during listening and consecutive interpretation, followed by simultaneous interpretation and shadowing.  It is hypothesised that the concurrent vocal activity on the part of the interpreter which takes place in both simultaneous interpreting and shadowing may be a source of conflict that prevents the interpreter to process the material to any greater extent. This problem does not arise in listening or consecutive interpreting. In the light of this hypothesis, Lambert questions Mackintosh’s (1985) conclusion that simultaneous interpreting imposes a heavier processing load than consecutive interpreting. Lambert claims that the greater number of departures during simultaneous interpretation may have been due to the simultaneity of listening, translation and speaking — conflicting activities which prevent the interpreter from processing material as deeply as under consecutive interpreting conditions (Lambert 1988:386).[10]