• 7.1 Chernov’s "probability prediction mechanism"
• 7.1.1 Redundancy
• 7.1.2 Distribution of redundancy in texts
• 7.1.3 Probability prediction model
• 7.2 Cognitive problems: an example
• 7.3 Alexieva’s model
• 7.3.1 Pronunciation errors

7 Cognitive models
7.1 Chernov’s "probability prediction mechanism"
7.1.1 Redundancy	In communication and information theory redundancy is described as essential to combat noise, to assure reliability and to maintain a communication channel. English writing is estimated to be 50 % redundant which accounts for the ability of native speakers to detect and correct typing errors. The amount of information actually transmitted is not increased (Krippendorff 1986). Chernov (1979, 1985) has pointed out that simultaneous interpreting is impossible without redundancy, and it plays a significant role in his model for simultaneous interpreting. According to Chernov (1985) our understanding of language is based upon the human ability to make inferences. We are able to extract the meaning of a message after only receiving part of it by making linguistic, cognitive, deictic and pragmatic inferences. For this to be possible in such a complicated activity as simultaneous interpreting, there has to be a certain level of redundancy in the message, and this level is higher than in written translation. Another important factor in our comprehension process is our inherent ability to make prognoses, our "probability prediction mechanism" which helps us to adjust instantly to changes in the environment. This has to do with how the nervous system works: "the human central nervous system developed as a mechanism of maximal anticipation of sequential and iterative phenomena of the outside world at the greatest possible speed" (Anokhin 1978:19, quoted in Chernov 1994:145).  Chernov distinguishes between objective and subjective redundancy. Objective redundancy consists of iteration of message components and their interdependence, and these factors are independent of the message recipient. But a communication may also be subjectively redundant for the message recipient. This redundancy is built up by inferences by the hearer about the meaning of the utterance and the part of discourse already produced. Inferences can be broken down into linguistic, cognitive, situational, and pragmatic. Linguistic inferences are drawn about the verbal form of the message, and about the referential component of the semantic structure of the discourse. Both syntactic and semantic rules apply as sources of linguistic inference. Thus, a semantic constraint, as an objective factor of redundancy, becomes a subjective factor of linguistic inference for a given hearer, depending on, and drawn on the basis of, his/her knowledge of the language spoken (for a discussion, cf. Chernov 1994:142).[12] Cognitive inferences are made when the utterance makes sense, i.e. when the semantic components already produced interact with the listener’s background knowledge. E.g., to understand the utterance "he studied at Eton", one must have the appropriate background knowledge about Eton College in the United Kingdom (Chernov 1994:143). Situational inference has the communicative situation or situational context as source of inference. E.g., a speaker’s address "Mr. President..." could allude to the president of a country, or of a company, or the president of the United Nations, etc. Chernov has identified eight factors involved in simultaneous interpreting situations: 1. characteristics of the source message, or speaker (S), obtained from a reply to the question "Who is speaking?" 2. theme of the message (Th): "What is he talking about?" 3. relation of the act of speech to the event that provoked it (E): "In what connection is he speaking?" 4. message recipient, or audience (A): "Whom is he addressing?" 5. place, or forum (F): "Where is he speaking?" 6. time (T): "When is he speaking?" 7. purpose of the communication (P): "What is he aiming at?" 8. motive (M): "Why is he speaking?"  (Chernov 1975, 1978, 1987, 1994). Pragmatic inference is made when the hearer draws conclusions about the speaker and his/her social role, on the basis of the semantic contents of the utterance, the background assumptions of the hearer, and the hearer’s knowledge of the factors of the communicative situation.
7.1.2 Distribution of redundancy in texts	Information content or redundancy are not evenly distributed throughout the communication. According to Chernov, redundancy is concentrated in the thematic, or topical, part of utterances, whereas the greatest density of information is at the rheme of the utterance. It is thus possible to compress speech in various ways in the thematic part by reducing the number of syllables, words or semantic components, and by simplifying syntactical structures. An example, given by Chernov: a chairman’s announcement: "I now give the floor to the distinguished delegate of the United Republic of Tanzania!" can be interpreted by a simple "Tanzania!". Here, only the rheme of the utterance remains after the act of compression. Since humans identify figures through the perception of stretches of maximum curvature, not of straight line, and that moving objects are given priority in perception to stationary objects, Chernov argues that comprehending meaning and sense is subject to a similar law of perception. Attention is thus primarily focused on semantic components that carry new information. This means that the interpreter’s attention is focused on the rheme of the utterance. Misperceptions or losses of items in the thematic, redundant, portion of the communication can be restored, but the loss of a rhematic item may easily result in mistranslation (Chernov 1994:147).
7.1.3 Probability prediction model	The "probability prediction machinery" in simultaneous interpretation works in four tiers: (a) Sound patterns (syllables encoding phonemes, intonation, stress, and other prosodic features) (b) Grammatical (syntactical) and categorical semantic features (c) Semantic tier per se (d) Sense tier per se The levels of the "probability prediction machinery" are based on redundancy from the level of the syllable to that of the word, phrase, utterance, communication (text) and situational context. These levels are distributed among the tiers as follows:  Tier Redundancy level (a) prosodic: sound patterns syllable, word, phrase, utterance (b) syntactic: grammatical features and semantic categories phrase, utterance (c) semantic per se phrase, utterance, text (d) implicational: sense per se utterance, text, situational context Figure 7-1 Chernov's redundancy levels A close co-operation between the various levels starts at the moment, or even before, the speaker is given the floor. If the speaker is known to the interpreter, the interpreter starts immediately to work out a probability prognosis for the semantic structure and sense of the incoming message. This is possible due to the interpreter’s previous knowledge about other factors in the interpreting situation. The process can be described as a top-down prognosis made at the highest tier. The next step in the prognosis is effectuated at the acoustic/prosodic (a)-tier, from the bottom and upwards, immediately involving the syntactic and semantic tiers (b) and (c).  If the speaker is unknown to the interpreter, and the situation in general is unfamiliar, probability prediction begins in the bottom-to-top direction. As the message develops, which normally happens during the first few sentences, a probability prognosis for the whole message (or its first thematic part) is starting to take shape in the mind of the interpreter, with the interaction of all the levels described. If this interaction does not take place, interpretation errors and omissions appear. As the meaning and semantic structure of the message develop, the forecast of semantic and even purely linguistic features of the text narrows down at times to certainty (probability = 1). But as soon as the speaker starts on a new topic, the probability prediction process starts all over again. Probability prediction is facilitated by the fact that the feed-back process is carried out subconsciously as automatic operations according to internal programmes, plans of the utterances. Only when difficulties of perception occur, is attention fully switched over to perception and comprehension. Such disturbances can be 1) noise or rapid rate of speech, resulting in low perceptional redundancy at lower levels of the mechanism, 2) complicated syntactic structures or syntactic errors in the source language utterance, resulting in low redundancy at the syntactical tier, 3) unknown terminology, semantic gaps or reference to unknown facts or events, resulting in low redundancy at semantic and sense tiers. In such cases, the interpreter becomes unconscious of his/her own target language communication, and if errors and omissions occur, they are not corrected.
7.2 Cognitive problems: an example	In the following example from our study, the interpreter has a problem with the Swedish term "matrilinjal" (= Eng. matrilineal), which the speaker introduces — with some problem. At first the interpreter does not interpret the term at all, but the next time the speaker uses it, the interpreter introduces the non-existing word "*matriaalinen" (Eng. approx. *matrial) and the last time the speaker uses the term, the interpreter decides to choose the existing word "matriarkaalinen" (= Eng. matriarchal), but which is also wrong. Swedish original Translation of original Finnish interpretation Translation of interpretation ...han beskriver den prehellenistiska världen // som ju va en värld / eeh en matria_ matrilinjal värld / en värld en matriar_ en värld / som ni vet / som man tror / dyrkade den stora gudinnan / ...  (P 19-20) ...he describes the prehellenistic world // which was a world / eh a matria_ matrilineal world / a world a matriar_ a world / which as you know / as we believe / worshipped the great goddess / ...hän kuvaa tätä prehellenististä ai_ maailmaa / maailmaa / eeh jossa / maailma / kuten te tiedätte / tai ainakin / kuvitellaan näin / siinä palvettiin / eeh suurta jumalatartanta /[13] ...he describes this prehellenistic ti_ world / a world / eh where / a world / as you know / or at least / so we believe / there they worshipped / eh the great goddess / ......men de gick på den matrilinjala linjen och / och mannen flyttade ju till kvinnans hem / allt de här som ni säkert känner till å / å de e ju etnografiska fakta att / att den matrilinjala eeh / världen den finns ju fortfarande kvar på sina håll  (P 21) ... but it followed the matrilineal line and / the man moved to the woman’s house / all this that you probably know and / and these are ethnographic facts that / that the matrilineal eh / world it still does exist in some places / ....mutta kaikki kulki tässä matriaalisella linjalla ja / eeh ja mies muutti naisen kotiin ja niin edelleen / eeh se tämä maailmahan on edelleenkin olemassa / tietyillä alueilla / ... but everything followed the matrial line and / eh and the man moved to the woman’s house and so on / eh it this world does still exist / in some areas / ....stadier i övergången från modergudinnan å till dom manliga gudarna å från hela den organisationsform som e / de matrilinjala / systemet till de patri_ / p_ patrikala / patriarkala / (P 23-24) ... stages in the transition from the mother goddess and to the male gods and from all of this form of organisation which is / the matrilineal / system to the patri_ / p_patrichal / ....vaiheista jossa siirrytään äitijumalasta / miesjumaliin / ja koko tästä matriarkaalisesta eeh järjestelmästä / tähän patriarkaaliseen järjestelmään / .... stages where they change from the mother god / to male gods / and from the whole matriarchal system / to this patriarchal system / bold italics the term 'matrilinjal' and its translations Figure 7-2 The term "matrilineal" in various shapes Note that the speaker in the beginning has obvious problems with the term . She begins ’ — after some hesitation — with the false start ’matria-’, immediately followed by ’matrilineal’, and the rest of the passage shows continued uncertainty: ’a world a matriar_ a world / which as you know / as we believe / worshipped the great goddess’. Irrespective of whether the interpreter is familiar with the term ’matrilineal’ or not, her omission of the term in the first passage may be a strategic choice: waiting for the term to be stabilised by the speaker.[14] Anyway, as a result of these dramatic events, the interpreter’s output is affected on the phonological and morphological level (cf. next section). Chernov’s model presupposes concurrent operation of the probability prediction model at several levels at each given moment in time; multichannel information processing; and heuristic interplay of levels from bottom to top and from the top down. It is thus a dynamic model like the one by Kintsch and van Dijk.
7.3 Alexieva’s model	Another dynamic model of the comprehension and production processes in simultaneous interpreting has been developed by Bistra Alexieva (1985, 1988, 1994). She refers to her model as "semantic" (1988), although she emphasises the necessity of studying units of text larger than the phrase, such as paragraphs and whole texts.  Alexieva has defined four levels for the simultaneous interpreting process: A-level: the temporal properties of the interpretation (speed, number of pauses, the length and function of pauses); B-level: other "oral" factors: pronunciation and prosody, para-linguistic and extralinguistic features, e.g. variations in pitch, loudness, diction; C-level: a linguistic level, which can be divided into three parts: morphological, syntactic, and lexical; D-level: semantic entities above word level, i.e. phrases, paragraphs, utterances and suprasentential segments in the text produced.  Phenomena that are noticed at a lower level, may be explained by analysing the interpretation at the highest level.
7.3.1 Pronunciation errors	In this sample, which is part of the example in the previous section, the interpreter makes two errors in pronunciation of two quite usual Finnish words:  Swedish original Translation of original Finnish interpretation Translation of interpretation ... värld / en värld en matriar_ en värld / som ni vet / som man tror / dyrkade den stora gudinnan / ...  (P 20) ... a world a world a matriar _a world / which as you know / as we believe / worshipped the great goddess / ... / maailma / kuten te tiedätte / tai ainakin / kuvitellaan näin / siinä palvettiin / eeh suurta jumalatartanta / ... a world / as you know / or at least / so we believe / there they [worshipped] / eh the great [goddess] / bold: mispronunced words Figure 7-3 Pronunciation error due to cognitive overload Note the interpreter's *palvettiin instead of palvottiin ’was/were worshipped’, and *jumalatartanta instead of jumalatarta = partitive singular of jumalatar ’goddess’. In the terms of Alexieva’s model these ’neologisms’ can be described as morphological problems on the C-level (or possibly, in the case of *palvettiin, a phonological feature on level B). They are signs of a broader "cognitive overload" problem on the D-level: the interpreter is processing the unclear utterance of the speaker (see previous example) and the normal self-correcting mechanism is disabled. Alexieva’s semantic model as a tool for the construction of a text typology of interpreted events is described in section 4.4.1.