Corps de l’article

This world is in love with its own ignorance.

Sri Aurobindo, Savitri, Book 6, Canto 2[1]

1. Introduction

An often maintained view of professional interpreters’ expert performance suggests that they should “produce the same effect on [listeners] as the original speech does on the speaker’s audience” as well as “the same cognitive content” (Déjean Le Feal 1990: 155). Moreover, this production is to be “presented with equal clarity and precision in the same type of language if not better, given that we are professional communicators, while many speakers are not, and sometimes even have to express themselves in languages other than their own” (Déjean Le Feal 1990: 155, my emphasis). If interpreters are to meet these high expectations, their target language production must at times outperform original speakers’ output, especially if it is in a foreign language. As English has become the global lingua franca, conference speakers’ “sometimes” having to “express themselves in languages other than their own” has become the rule rather than the exception, imposing extra pressure on interpreters to do “better” than the speaker (Reithofer 2010: 151). There is indeed preliminary evidence that interpreters “may improve on a deficient NNS [non-native speaker] source text” (Reithofer 2010: 151) or that “[a]nticipation and conscientious guesswork may even remedy some of the shortcomings of the [non-native English speaker (NNES)] original and make the interpreted version better understandable than the source text” (Kurz and Basel 2009: 193). Better comprehension results among those listening to an interpreter than those listening to the original NNES speaker have been demonstrated in Reithofer’s studies (2010; 2013). At the same time, ITELF-related studies, especially those based on self-report, suggest that this may come with the cost of extra cognitive effort (Albl-Mikasa 2022).

The manner in which interpreters handle ever more prevalent ELF speaker input is investigated in the context of ITELF (Interpreting, Translation and English as a lingua franca), an interdiscipline between TIS (Translation and Interpreting Studies) and ELF research, that has evolved over the last 15 years. According to ITELF studies, non-L1 use of English generates a number of vagaries for interpreters and translators with consequences for their performance and processing (Albl-Mikasa 2018; Albl-Mikasa and Ehrensberger-Dow 2019), due to “the variety and unpredictability of language parameters: interlocutors’ accents, transfer features, and proficiency levels” (Mauranen 2012: 7). While interpreters generally strive for high target speech standards, some, according to interviews among professional conference interpreters, are tempted to adopt a “garbage in—garbage out” strategy when confronted with non-L1 English speakers (Albl-Mikasa 2014: 296). And indeed, in ELF settings, interpreters are particularly concerned about non-L1 use of English, which is also reflected in ELF research efforts: “ELF does not exclude NSs [native speakers] of English, but they are not included in data collection, and when they take part in ELF interactions, they do not represent a linguistic reference point” (Jenkins 2007: 3). In the following, as in ITELF research in general, the term ELF speaker is therefore used to refer to the non-L1 speaker of English who provides the input to be rendered by the interpreter for a target (language) audience (Albl-Mikasa and House 2020: 171-172).

This raises the question as to what interpreters actually do when faced with ELF input as opposed to more conventional “Standard English”[2] input? Do they actually improve ELF speakers’ source texts and, if so, how do they go about it? What does “making the source text better” consist of? Relevance Theory (RT) seems particularly suited to answering these questions. Conceiving of communication as both ostensive and inferential (see Section 2 below), it is geared towards both speaker and hearer, two roles interpreters enter into. This is because, although they are not the addressees of the speaker’s message, interpreters are both (interim) recipient (of the ELF source speech) and producer (of the target speech). Moreover, RT also covers the original speaker. It can thus provide a framework for explaining interpreting under ELF conditions, in that it captures both the interpreter dimension, that is how interpreters understand ELF speaker input and make amendments to enhance that input for the target audience, as well as the ELF speaker dimension, that is how and why their ELF input may be suboptimal and require optimisation measures. In addition to the conceptual framework, RT provides the toolkit (in terms of enrichment processes underlying the recovery of explicatures and implicatures, see Section 3 below) to analyse in detail how interpreters choose to intervene and to compare source and target propositions.[3]

This paper sets out to shed light on the ways in which interpreters intervene in their effort to maximize ELF source speech comprehension and relevance for their audience. After exploring the explanatory potential of RT for an understanding of the communicative consequences of ELF on interpreting, it looks at 56 renditions by 28 professional conference interpreters. The data was collected within the CLINT (Cognitive Load in Interpreting and Translation) project, which uses a mixed-method design to investigate questions of processing problems, coping strategies, cognitive load and general challenges associated with the processing of non-L1 ELF as opposed to more standard English input (see Albl-Mikasa, Ehrensberger-Dow, et al. 2020; Ehrensberger-Dow, Albl-Mikasa, et al. 2020).

2. Conceptual framework

As “a general theory of human cognition and communication” (Carston 1988: 156), RT seems particularly suited to addressing the above questions regarding possible improvement measures on the part of the interpreter. As a cognitive psychological theory, it “treats utterance interpretation as a cognitive process” (Wilson and Sperber 2016: 278). At the same time, it goes beyond the general cognitive-psychological insights of the construction of mental representations in search for coherence and pragmatic acceptability (see Albl-Mikasa 2008) in that it proceeds from the fundamental and highly plausible assumption that the search for relevance is a basic feature of human cognition:

As a result of constant selection pressure towards increasing efficiency, the human cognitive system has developed in such a way that our perceptual mechanisms tend automatically to pick out potentially relevant stimuli, our memory retrieval mechanisms tend automatically to activate potentially relevant assumptions, and our inferential mechanisms tend spontaneously to process them in the most productive way.

Wilson and Sperber 2016: 253

This means that while RT, too, premises the construction of mental representations which undergo inferential computations for the understanding of utterances, it adds a decisive dimension. To illustrate this, reference assignment is a case in point. From a cognitive perspective, rather than simply identifying an appropriate object or event, hearers and listeners access a mental representation which uniquely identifies the intended referent. This is further specified through RT in that, “according to the RT framework, a pragmatically acceptable interpretation is the first interpretation consistent with the principle of relevance” (Blakemore 1992: 73).

Reference to relevance is what makes RT particularly interesting for CTIS (Cognitive Translation and Interpreting Studies) and ITELF because it is explanatory of both the speaker and hearer sides, and also the way speakers make propositional content explicit. According to RT, communication is ostensive-inferential in the sense that it is for the communicator to make ostensibly manifest what s/he wants to communicate and for the hearer to infer the communicative intention. According to the first principle, the Cognitive Principle of Relevance, which states that “[h]uman cognition tends to be geared to the maximization of relevance” (Wilson and Sperber 2016: 254), “what makes an input worth picking out from the mass of competing stimuli is […] that it is more relevant than any alternative input available” (Wilson and Sperber 2016: 252). An utterance is more relevant the more cognitive effects it generates and the less effort for processing it induces (Carston 1988: 168). A positive cognitive effect is a “worthwhile difference to the individual’s representation of the world” (Wilson and Sperber 2016: 251). According to the second principle, the Communicative Principle of Relevance, which stipulates that “[e]very ostensive stimulus conveys a presumption of its own optimal relevance” (Wilson and Sperber 2016: 251), a stimulus is ostensive in that it attracts an audience’s attention. Since cognition is geared towards the maximisation of relevance, an audience will only pay attention to a stimulus that seems relevant enough. A communicator trying to attract attention will therefore produce stimuli that are the most relevant (that is yielding the greatest effects in return for the smallest processing cost) that s/he is willing and able to produce, thus creating expectations of relevance. The audience, in turn, will presume that the input is relevant enough to be worth processing (that is testing interpretive hypotheses) and will stop processing when his or her expectations of relevance are satisfied (Wilson and Sperber 2016: 259), that is when the least costly interpretation provides an adequate range of cognitive (or contextual) effects against the context of pre-existing assumptions (Carston 1988: 168-169).

Cognitive effects or changes to an individual’s set of assumptions include the strengthening, revision or abandonment of available assumptions, which constitute the hearer’s “cognitive environment” or “representation of the world.” They result from the processing of input in the context of previously held assumptions. Utterance interpretations can only be derived from contextual knowledge and not from input or context alone (Wilson and Sperber 2016: 251). Defined “in psychological terms as a subset of the hearer’s beliefs and assumptions about the world” (Blakemore 1992: 18), context encompasses the interpretation of the preceding discourse as well as information from the physical environment or stored memory. From the enormous amount of background information, hearers select or construct the assumptions for the occasion that are the most accessible and minimise processing cost in yielding the intended interpretation (Blakemore 1992: 18, 88).

Looking at the other side of the coin, according to the Communicative Principle of Relevance, hearers can take it for granted that “[a]iming for relevance, any speaker must make assumptions about the hearer’s processing abilities and contextual resources, and these assumptions will be reflected in the form of his utterance” (Blakemore 1992: 52, my emphasis). Again, this goes to show that RT is concerned not only with the (inferential) comprehension side of utterance processing, but also with its production side, more specifically, with the input generated to enable efficient relevance-geared comprehension. This means that, as producers of target output, interpreters have to phrase their renditions in such a way that their audiences can derive the cognitive effect intended by the original speakers at minimal cost. As listeners/recipients, interpreters can proceed from the default assumption that the input they receive will guide them towards relevance (that is maximum cognitive effect or informational gain at minimum processing cost).

In ELF conference settings, interpreters are faced with ELF speakers, whose second or foreign language English may not provide them with the linguistic means of expression to allow them to choose, according to RT, “the linguistic form of utterance to guide the interpretation process” or the “linguistically specified [lexical, syntactic or intonational] devices to constrain the hearer’s choice of context” (Blakemore 1992: 176, 177). As a result, the following basic premise of communication may be undermined: “the expectations of relevance raised by an utterance are precise enough, and predictable enough, to guide the hearer towards the speaker’s meaning” (Wilson and Sperber 2016: 250). It seems especially questionable that the ELF speaker aiming at optimal relevance will be in a position to “not only try to give the hearer adequate contextual effects, but also try to give these effects for the minimal necessary processing effort [to] put the hearer to no unjustifiable effort in obtaining these effects” (Blakemore 1992: 74). Judging from the following ELF and ITELF research findings, it is plausible to assume that ELF speakers do not always have the linguistic means of expression available that will ensure such optimal relevance and minimal processing effort. As found by (often corpus-based) ELF research, characteristics of ELF communication include pronounced proficiency level variations and extensive cross-linguistic influence from ELF speakers’ L1, innovative (non-standard) patterns, ad hoc creations and so-called approximations to collocations in ELF usage (Mauranen 2012). ELF-induced phenomena in the areas of processing, accent, lexis, grammar, cohesion and argumentation have been shown to deviate from expected standards to a considerable degree. This makes them unpredictable and may trigger interpreting difficulties and disruptions in comprehension and processing as evidenced by interpreters’ self-reports (Albl-Mikasa 2018; Albl-Mikasa 2022; Albl-Mikasa and Gieshoff [forthcoming]).

In demonstrating how linguistic devices may be misleading when not used in line with L1 speaker conventions, so-called “procedurals” seem to be an excellent point in case. As is illustrated in the following example from Blakemore (1992: 136), they are designed to guide utterance interpretation. They may, however, fail to fulfil this function in the context of ELF.

Contrary to representational expressions, which help hearers identify the mental propositional representations to construct (for example that Barbara and David are not in town), procedural expressions such as so, after all or however “do not contribute to a propositional representation, but simply encode instructions for processing propositional representations” (Blakemore 1992: 151). They help constrain the interpretation to be recovered by the hearer by constraining the hearer’s choice of context. Discourse connectives like so or therefore, for instance, introduce contextual implications (for example that David will not be in town if Barbara is not); discourse connectives such as after all strengthen existing assumptions (for example that David is indeed not in town if Barbara is not); and discourse connectives such as however introduce denials contradicting an existing assumption (for example that David is not in town despite his normally being in town when Barbara is not). If the two sentences were presented without connectives (“Barbara isn’t in town. David isn’t here.”), any of the interpretations in the three examples would be possible and it would take more processing effort to access further contextual assumptions in order to decide which of the interpretations is actually intended. Context selection is thus ensured by these procedural expressions at minimal processing costs along with the contextual effects on and changes to the hearer’s cognitive environment. By instructing the hearer how to manipulate the inferential computations and process the proposition expressed in a particular context, these procedural devices are “effective means for constraining the interpretation of utterances in accordance with the principle of relevance” (Blakemore 1992: 137).

Given the crucial nature of this guiding function, problems can be expected to arise when procedural devices cannot fulfil it in ELF interpreting contexts. Even for non-ELF contexts, procedurals have been hypothesised in the context of interpreting to “correlate with the least readily ‘translatable’ items in language,” contrasting with “technical and single-referent terms like ‘dehumidifier’ or ‘NATO’” (Setton 2005: 81). This has been confirmed for the translation process, where processing effort has been found to increase for segments conveying procedurally encoded information as opposed to those rendering conceptually encoded information (Alves, Gonçalves et al. 2014). In ELF contexts, this may be aggravated because procedural expressions are among the expressions which can be particularly hard for ELF speakers to use in line with L1 speaker conventions. ELF studies find that ELF speakers tend to “not only use different discourse markers but also attribute different functions to them” (Cogo and House 2018: 215). For instance, “[a]s opposed to uses of so as an interpersonal marker, ELF users realize it to express self-attention, as a discourse-structurer and a ‘fumble’ to overcome formulation problems” (Cogo and House 2018: 216). This is thus a situation in which procedural expressions, on account of their crucial function of pointing the way towards particular contexts of the unfolding discourse, “must be more intricately intertwined with the syntactic, lexical and prosodic fabric of a language than any other items” (Setton 2005: 80). Alas, they are probably less entrenched in non-L1 speakers who learned English explicitly rather than acquiring it implicitly (Williams 2009: 318). This suggests that discourse connectives in ELF settings may fail to raise precise and predictable enough expectations of relevance. As a result, rather than reduce the effort by guiding the process of deriving cognitive effects, they may actually increase it. Moreover, further effort may be required from interpreters in that they may have to replace or add procedurals in the target speech in order to restore their relevance-guiding function for the listeners.

Against this backdrop, ELF speaker input may be found to be wanting in relevance-theoretic terms. An ELF speaker’s chosen expressions may at least partially fail to fulfil their guiding function towards relevance and towards providing the basis for listeners to infer the speaker’s communicative intentions. To put it more concretely: An ELF speaker in an ELF setting may use linguistic means of expression that require more processing effort for the recovery of the intended meaning (or produce no additional benefit) because of a lack of “pragmatic fluency” (House 1999: 86). Interpreters can therefore be expected to require greater comprehension effort in order to infer the explicated and implicated intentions from the unreliable input. On top of that, they will have to produce amended target renditions which compensate for the lack of relevance guidance in the ELF source input and restore that guidance so that the audience is enabled to optimally infer relevant (that is easily accessible) content as intended by ELF speakers. In their capacity as target speech producers, interpreters will have to “make a decision about what to make explicit and what to leave implicit” (Blakemore 1992: 173), which may involve reorganising, at additional cost, the source text’s explicit-implicit pattern for the target text with a view to optimising relevance for the audience. Moreover, when interpreting for non-L1 English audiences, interpreters may have to make yet another extra effort to assess, adapt to and accommodate the estimated contextual resources of their audience in order to maximise relevance for them.

Based on cognitive-psychological RT premises, interpreters’ default expectations are standard assumptions of optimally relevant input. Faced with ELF speaker input that is suboptimal in its guiding functions, interpreters may come under pressure, requiring additional resources for the recovery of relevant cognitive effects on their part and compensation and restoration of such effects for the target audience. The RT framework thus provides a solid foundation for a better understanding and demonstration of why interpreters might find it necessary to intervene and optimise ELF source input.

3. Methodological toolkit

Just as the conceptual RT framework offers explanatory insights into interpreter processing of ELF input, so the methodological tools provided by RT are ideally suited for the analysis of interpreter renditions of ELF input because they take into account both decoding and inferencing processes in explaining utterance interpretation. Setton speaks of RT as a “code-plus-inference model of linguistic communication” (2005: 70). While cognitive-constructivist discourse analysis approaches hinge on the gap left by the underdeterminacy of linguistic expressions and how this is bridged by the inferential derivation of speakers’ thoughts and communicative intentions (Brown and Yule 1983), the RT framework also addresses the linguistic input and introduces procedures for the recovery of explicitly communicated content as well as the derivation of contextual implications. In Carston’s words:

Pragmatists have given much attention to what an utterance can convey implicitly […]. Equally important, however, is the proposition explicitly expressed by the utterance of linguistic expressions, what is said, in Grice’s terms […] [s]ince what is said (the explicit) and what is implicated (the implicit) exhaust the (propositional) significance of the utterance.

1988: 155

This seems essential for interpreting because faithful rendition as expected from interpreters involves a focus on the explicitly communicated and even some consideration of the linguistic input. Complete and accurate rendering of source utterances presupposes processing along the narrower lines of the (micro-)propositions expressed because it is at that level of cognitive processing that accuracy and detail in recall and completeness of source message rendition can be secured while also keeping processing costs low (Schnotz 1994: 155, 180, 201; Albl-Mikasa 2017: 93). At the same time, some attention has to be directed to the linguistic surface structures because, on top of conveying propositional content and illocutionary function, “interpreting is essentially an (idiomatic) re-lexicalization and re-textualization task that requires terminological precision and some degree of retention and getting across of routine expressions (cf. Ilg 1980)” (Albl-Mikasa 2017: 95). Consequently, in addition to the (explicated and implicated) propositional content of the source utterance, the linguistic input needs, to some extent, to be catered to. In RT, this is covered by a fundamental distinction between the explicitly communicated, that is the explicature, which is derived via decoding and inferencing and the implicitly communicated, that is the implicature, which is derived via inferencing alone (Sperber and Wilson 1986/1995: 182). The decoding part is crucial in that it caters to interpreters’ actual and precise reconstruction of propositional content and the amendments they make to the source input. According to RT, linguistic input is decoded and the resulting logical form or semantic representation developed or “fleshed out” (Carston 1988: 164-167) by “the building in of extra information of one sort or another” (Carston 1988: 317). This is done by means of pragmatic enrichment processes, such as “reference assignment, disambiguation, specification of vague terms, supplying of ellipsed material, building in certain [causal and temporal] relations between events and states, supplying empty grammatical categories with conceptual content” (Carston 1988: 317). The resulting explicature may be a basis for the derived implicatures. However, hypotheses or assumptions are not constructed in a sequential but in a parallel manner so that hearers may arrive at the implicature right away (Wilson and Sperber 2016: 261).

The “relevance-theoretic comprehension procedure [that] applies in the same way to the resolution of linguistic underdeterminacies at both explicit and implicit levels” (Wilson and Sperber 2016: 264) is illustrated by the following example from Carston (1988: 155):

As a hearer of B’s utterance, A combines the proposition expressed with other contextual assumptions to construct hypotheses and come to an interpretation, such as the following (italics are used to indicate propositions/assumptions/thoughts/interpretations as opposed to utterances/natural language expressions):

This example illustrates the different types of enrichments or enrichment processes undertaken to arrive at the explicature or explicit content of the utterance. The explicature differs from the implicated assumptions (implicatures) in that it is based on both decoding and inference, and therefore overlaps in content and form with the utterance, that is explicitly reflecting some of the elements of the utterance. By contrast, these remain invisible in the fully inferred implicatures, which are independent assumptions (Carston 1988: 157). The contextual implication or implicated conclusion is the most important type of cognitive effect a hearer may gain (as indeed the implicature Jane is not feeling happy is what the hearer, who asked the question in A above, had wanted to be informed about). However, in the context of interpreting, the interesting part is the explicature or, more precisely, the enrichment or completion processes that turn the decoded logical form of the utterance, which served as a template for the development of the propositional form, into the explicature. This is because the explicature or full proposition as intended by the speaker is what the interpreter is supposed to convey, leaving the derivation of implied content or implicatures to the primary (target) audience.

In non-mediated communication, the inferential enrichment processes a listener undertakes in order to develop and flesh out the logical form of an utterance to arrive at the explicature remains hidden in this primary listener’s mental blackbox. There is no access to what this listener understood. In interpreter-mediated communication the enrichment processes and developments or completions become manifest and observable in the interim productions the interpreter, as the secondary listener, supplies for an audience. According to RT “every utterance is an interpretive representation of a thought—namely, the thought that the speaker wishes to communicate” (Blakemore 1992: 161). Similarly, as Gutt (1991) emphasises, translation is a matter of interpretive resemblance, in that “an utterance can be used to represent another utterance” (Blakemore 1992: 161), so that “the rendered text is an interpretation of the translator’s/interpreter’s thoughts which are themselves an interpretation of the thoughts of the original speaker” (Blakemore and Gallai 2014: 107). Manifested in interpreters’ renditions, the enrichments and completions speak to what the interpreters understood of the original speaker’s utterances and to what interventions they chose to carry out to re-represent these utterances in their enactment of speaker fidelity. While an utterance is an interpretive representation of thought or, as explained above, also of an utterance “to the extent that it resembles it in semantic and logical properties” (Blakemore 1992: 161)—so that source and target utterance must share those properties—, the closest in resemblance may not be the optimally relevant representation. The degree of faithfulness or interpretive resemblance varies from situation to situation and the proposition expressed does not have to be identical to the thought or utterance it represents. For instance, a strictly true interpretive representation may be less relevant in the sense that it may entail extra processing cost. To take another example from Blakemore (1992: 109), telling a friend that one earns 900 pounds a month (while, in a tax return, one would put down the exact salary figure of 897 pounds) would be the most relevant utterance, in the sense that it provides the information as intended by the speaker or the informational gain for the hearer at less processing cost than the exact, “true” figure. Similarly, in translation, providing a target text rendition with a view to satisfying the informational needs of the audience at an acceptable cost may be more relevant than an extremely close resemblance of source and target utterances.

This may involve a “reorganization of the explicatures and implicatures conveyed by the source text into a target text counterpart” (Alves 2007: 62). In other words, assumptions implicated in the source text may have to be made explicit in the target text (and vice versa) in order to achieve interpretative resemblance in line with the principle of relevance. A case in point is explicitation, when translators or interpreters make explicit propositional content left implicit by the source text or speaker. In his application of RT to translation, Alves (2007) uses the RT framework to demonstrate how the extra effort allocated for explicitations may, in some cases, lead to additional information gains for the audience and, thus, contribute to “the maximisation of relevance” (Alves 2007: 73), while, in others, constraining readers’ inferential processing in a contextually inadequate manner. The RT procedures described above allow us to gauge interpretive resemblance between source input and target output and to see how interventions and amendments, while differing from the source input, may actually heighten (logical-semantic) resemblance. As such, the enrichment processes as defined by RT become a toolkit for analysing the various types of enrichments or extra information interpreters chose to build into their renditions. Analysis of the enrichment procedures will serve to determine a) what interpreters understood the source utterance to express and b) what they felt they should make explicit to maximise relevance for the target audience. In ELF contexts, this makes it possible to determine possible comprehension problems induced by ELF speaker input, the extent to which resemblance can be produced from ELF input, the actual optimisation efforts undertaken by the interpreters and the kind of resemblance they deemed to be adequate. RT procedures thus allow us to monitor the interventions and amendments carried out by interpreters, pointing to three plausible scenarios for interpreting under ELF conditions:

In those cases in which interpreters receive non-optimal cues from ELF speaker input, they will set out to provide “better cues” with a view to supporting relevance for the target audience, thus optimising relevance for them. In RT terms, this suggests that interpreters’ output may generate more readily derivable cognitive effects for listeners than the original ELF speaker output (which was the case in Reithofer’s [2013] study above). In their effort to create interpretive resemblance and shared assumptions (including shared contextual implications), interpreters will develop the logical form of the ELF input in a way that makes their target output more relevant (for example less costly in terms of processing while remaining adequately informative) for the audience. In this way, the higher processing cost endured by interpreters during the recovery of the source input may not be passed on to the listeners of the target rendition. As applied below, the RT framework provides the instruments to analyse whether, to what extent and how interpreters treat ELF source speeches in their effort to maximise relevance for the target audience. The following presents an empirical RT-based study that showcases how precisely professional interpreters go about this endeavour. By comparing the renditions of an ELF speech with those of the edited version of that same speech, differences in the type and number of enrichments and completions serve to illustrate the actual steps taken and interventions carried out by the interpreters to maximise relevance.

4. Empirical investigation

In the following, I present an RT-based analysis that compares interpreter processing of ELF and non-ELF input. It proceeds from the above-mentioned assumption that interpreting ELF input requires more interventions than rendering “Standard English” input[4] because of the potential failure of ELF input to maximise relevance for the hearer/listener/interpreter. Against the backdrop of the notion of interpretive resemblance, the comparison is based on the manifestation of interpreters’ comprehension results and production choices in the form of the target speech. The corresponding source and target items reflect the interpreters’ assumptions on how the ELF speaker’s communicative intentions should be understood and conveyed. In the detailed analysis, the various enrichments (such as reference assignment, disambiguation, specification, closing of elliptic gaps and general enrichment), as manifested in the target rendition, act as indicators of the interpreters’ decision-making and intervention choices.

4.1 Project methodology

For the purpose of this paper, a detailed RT analysis was conducted of 56 interpretations of two speeches, both available in an original authentic ELF version and a version edited to conform to expectable “Standard English” conventions. The first speech is of a more general (Text 1) and the second of a more technical (Text 2) nature. The data is part of the larger-scale interdisciplinary Sinergia project CLINT (for details see Albl-Mikasa, Ehrensberger-Dow, et al. 2020), which examines how interpreters and translators with different levels of expertise cope with (spoken and written) ELF input compared with multilinguals with no training in translation or interpreting (see Ehrensberger-Dow, Albl-Mikasa, et al. 2020). The RT analysis is conducted exclusively on the interpreting data and only on the professional interpreters’ renditions (not on those of the 24 student interpreters also participating in the project).

The renditions of 28 professional interpreters were subjected to the RT analysis. Their age ranges from 28 to 64 years (MD = 49) with an interpreting experience of 2 to 38 years (MD = 19). All participants had a degree in conference interpreting with German as their A and English as their B (19) or C (9) language. Most participants had French and/or Italian in their language combination. All had given their informed consent to participate in the study after receiving information about the procedure and the data collection, were compensated for their participation and informed that they could opt out of the experiment at any moment.

The source speeches were authentic presentations recorded at real-life conferences in an ELF setting. The first one was held in English by an Italian L1 speaker on energy-related matters in Switzerland, the second one by a Chinese L1 speaker on demand forecasting methods in China. The first one can be described as fairly generic, with 96% of all words among the 5000 most frequent words in American English[5]. The second text was more technical, with 89% of all words amongst the most common ones.

Both speeches were recorded, transcribed and edited to conform to the conventions of L1 English, with hesitations and processing phenomena moved to places more conventional for L1 speech. The edited version did not iron out vague expressions or poor logic but reflected the diction and speaking style of an English L1-speaker and had typical ELF phenomena removed, as can be seen in the following example.

The transcripts of both the original (ELF) and the edited (EdE) versions were read out by the same professional female Northern American speaker with a rather neutral accent and recorded on video. Eliminating the confound of accent and speaker-related properties by recording both versions with the same speaker allowed us to enhance comparability and, with a view to the overall project, to enable comparison of spoken and written ELF in the interpreting and translation tasks. Moreover, accent is associated with both L1 and non-L1 speech and has been studied more intensely than non-accent-related ELF features elsewhere (for example Cheung 2013; McAllister 2000). All source speeches were made available in the form of videos, each approximately 12 minutes in length. The delivery speed was around 125 words per minute for both versions of Text 1. The ELF version of Text 2 registered 108 words per minute and the respective edited version came in at 118 words per minute. The slower pace of the ELF version of Text 2 was due to a great number of pauses in the ELF original, which, together with slower delivery pace, is rather common for (lower proficiency) ELF speakers. Participants were tested individually at the ZHAW laboratory in a simulated conference interpreting workplace setting, that is one equipped with a console, mike, etc. After a five minute warm-up interpretation of a very general text, half of the participants interpreted the authentic ELF version of Text 1 and, after a break, the EdE version of Text 2, the other half started with the EdE version of Text 1 followed by the ELF version of Text 2, so that each participant produced two renditions.

4.2 Enrichment categories as a basis for RT analysis

The 56 renditions by 28 professional interpreters (14 of Text-1-ELF, 14 of Text-1-EdE, 14 of Text-2-ELF and 14 of Text-2-EdE) were tagged for the following enrichments and completions taken from the RT literature[6], using MAXQDA software for qualitative analysis.

Table 1

Enrichment categories from RT literature

Enrichment categories from RT literature

-> Voir la liste des tableaux

Reference assignment, disambiguation and enrichment general are the main completion categories featuring in the RT literature. However, the more detailed categories of filling elliptic gaps, supplementing cohesive elements such as conjunctions and adverbs and specifying vague expressions, are also dealt with, especially by Blakemore (1987). The category of neutralisation (a sub-category of disambiguation) is taken from Setton (1999) and of particular interest for the analysis of non-L1 speaker input as can be gleaned from the following quotation:

Two types of context function as disambiguators [of sentence ambiguity in actual discourse]: linguistic context disambiguates by suppressing virtual polysemy, while ‘cognitive context,’ accumulating through discourse (medium-term memory) and life (long-term memory) neutralises the effect of deviations such as ungrammatical and infelicitous expression (e.g. with non-native Speakers).

Setton 1999: 40, my emphasis

The category of enrichment generally covers all completions that do not fall under any of the more specific categories. It should be noted that the only rendition items/enrichments tagged were those that were successfully inferred, that is accurately rendered, due to the great many uncertainties associated with a qualitative assessment and analysis of inadequately rendered segments.

4.3 Enrichment-tagging of both versions of both texts

The following provides examples of a target rendition paragraph of each text version (Text-1-ELF, Text-1-EdE, Text-2-ELF, Text-2-EdE), illustrating the completions supplied by one of the 28 professional interpreters (IntPro).

In this paragraph of the rendition of ELF source Text 1, the interpreter (IntPro04) sets right the number agreement (this technologies—> diese Technologie) and supplies the appropriate preposition (mit) which was missing in the ELF original (combine without with) in segment 9. Apart from those two neutralisations s/he adds the enrichment andere [others] as well as the cohesive adverbial insbesondere [in particular] in that same segment. In segment 10, s/he specifies people into Bevölkerung [population] and, in segment 11, adds another enrichment damit die Leute wissen [so that people know, replacing so] as well as the corresponding referent sie [they] which helps to alleviate reference tracking and level out the register shift from third person [the people] to second person address [you]. Finally, a third enrichment (das wollten wir zuerst erfassen) [this is what we wanted to register first] is made in segment 12, which may either help the interpreter buy time and fill a reflection gap or may represent a felt need to make things clearer for the listener because of the somewhat unmotivated register shift in the original (We wanted… you… we wanted). Whatever the interpreters’ reasons behind supplying these completions when interpreting from the ELF version, it can be observed that, in the renditions of the edited version of the same paragraph, a much smaller number of completions, namely two, was felt necessary or appropriate (Figure 2): That is was expanded to wenn wir ihnen zeigen [when we demonstrate to them] in segment 11 and invite was disambiguated into auffordern [call upon] in segment 12. For what may be good reasons or bad, the ELF version clearly seems to have triggered more completion efforts on the part of the interpreter (compare Figure 2).

Figure 1

IntPro04 rendition of text-1-ELF

-> Voir la liste des figures

Figure 2

IntPro06 rendition of text-1-EdE

-> Voir la liste des figures

The interpreter’s greater activity of fleshing out the ELF source text for the target listener is even more evident for the more technical source speech (Text 2) in Figure 3. Reference is assigned in segment 33 (Methode/method substitutes one); it’s similar is enriched into wird ähnlich gearbeitet [a similar approach is taken] and thus made clearer (segment 34); the gap in segment 35 (and the….) is filled by supplying es gibt auch noch [there is also]; another enrichment is added for clarification in segment 36 (als zusätzlicher Parameter [as an additional parameter]); and in segment 37 the grammatical infelicity can be a exponential smoothing is neutralised into durch eine exponentielle Glättung ausgeglichen [levelled out by means of an exponential smoothening]. These five completions compare with a single one in the edited version of the same paragraph (Figure 4), namely the adding of the enrichment zum Zeitpunkt t [at time t] in segment 37, a simple repetition from segment 36. The greater number of enrichments may not necessarily be indicative of greater cognitive demands associated with the interpretation of the ELF version. Some may come automatically or not add to a better understanding by the target audience. All that can be said is that the interpreter clearly took more steps and engaged in more extensive completion activity, which may or may not require additional resources. In this context, it is interesting to note that, in the ELF version, two omissions can be observed in segment 36 (when demand occurs at time t) and segment 37 (and the estimation of probability can be another exponential smoothing), which is not the case in the rendition of the EdE version. Again, this may be indicative of higher cognitive demands, but need not necessarily be.

Figure 3

IntPro06 rendition of text-2-ELF

-> Voir la liste des figures

Figure 4

IntPro5 rendition of text-2-EdE

-> Voir la liste des figures

Even if the completions supplied in the renditions of the ELF and EdE versions of the two source speeches as analysed above cannot necessarily be taken to be indicators of extra effort on the part of the interpreter, they clearly illustrate the nature of the enrichments (and optimisation measures) made by the interpreters and provide first insights into the differences in completion numbers between the two versions. For a fuller picture, the following is a quantitative presentation of the results across all 28 interpreters and 56 renditions.

4.4 Enrichment results across all 28 interpreters

The following table (Table 2) lists the aggregated results of all 56 renditions of the more general text (14 each for ELF and EdE) as well as of the more technical one (again, 14 each for ELF and EdE).

What stands out in Table 2 and may also have been expected is the high number of neutralisations in the ELF renditions (overall 414) as compared with none in the EdE versions where infelicitous ELF-induced irregularities had been edited out. While it can be assumed that, on the basis of their multilingual competence, interpreters will neutralise singular/plural disagreements more or less automatically, other infelicities such as unconventional tense constructions or inaccurate lexical choices may not be ironed out as effortlessly. For instance, in Text 1, the conditional “would have done” re-occurred several times and seemed to have been used to indicate indirect speech or refer to thoughts and deliberations of the project team in the past. This turned out to be confusing: a participant mentioned in the retrospective comments that one never quite knew whether or not the project had actually been implemented. Moreover, approximations (such as “the people we were not looking at” for “looking for” in the sense of “target group”) seemed clear to some participants but not to others. Whether straightforward or not, the very number of these neutralisations (237 in Text 1 and 177 in Text 2 as opposed to none from the EdE source) makes it plausible to suggest that some amount of additional capacity may have been invested in these operations during ELF input processing.

Table 2

Number of all enrichments across all 28 interpreters in ELF and EdE versions of texts 1 and 2

Number of all enrichments across all 28 interpreters in ELF and EdE versions of texts 1 and 2

-> Voir la liste des tableaux

Table 2 also illustrates that the interpreters engage in a higher overall number of enrichments when rendering the ELF rather than EdE source text (a total of 962 in Text 1 and 586 in Text 2 as opposed to 645 and 303, respectively). This may indicate that the interpreters perceived the input as not clear enough, felt a need to make the communicative intention more explicit for the audience and that a greater decision-making effort was involved. The greater number of cohesive enrichments (178 vs. 127 in Text 1 and 116 vs. 57, that is double the number, in Text 2) also indicates that cohesive ties and (to a lesser extent) referential connections were not always deemed clear enough and sufficiently supportive when interpreting from the ELF source. Similarly, the higher number of filled elliptical gaps from the ELF versions (48 vs. 28 — almost twice as many as from the edited version of Text 1 — and 28 comparing to 0 in Text 2) points to a need for clarification. What is less obvious is the greater number of specifications and disambiguations in the general EdE (183 specifications and 25 disambiguations) as opposed to the ELF (153 and 12) texts. In the case of the technical Text 2, this reverse order is only true for the number of specifications (74 from the ELF and 79 from the EdE version), not for that of the disambiguations (19 from the ELF vs. 10 from the EdE version). It could be assumed that, when interpreting from the ELF texts, participants were too busy ironing out irregularities and infelicities and building cohesion to allot any resources to specifications. After all, it is fully legitimate for an interpreter to leave a source text as vague as is. As for disambiguations, this may be true for the general text, while for the technical text, disambiguation may well have been indispensable in conveying the message. However, it would take analysis of the self-report input to further interpret the results.

Any statement regarding effort invested by the interpreters remains speculative at this stage. What can be stated, based on the quantitative analysis, is that, in the renditions of the given case study, interpreters did resort to an overall rather large number of enrichments, thus making the content and intentions communicated by the speaker more explicit. In the technical text, almost double the number of enrichments were made when interpreting from the ELF as opposed to the EdE text and, in the general one, there were around 1.5 times as many enrichments from the ELF text. It needs to be noted in this context that the edited version was left vague where the original was vague and poor logic was retained (see editing process in section 4.1 above) which may have led to a higher number of enrichments also in the renditions of the edited versions, while the difference might otherwise have been even more pronounced. This is supported by our MAXQDA sub-tagging for enrichments that seemed to be clear candidates for producing a potential cognitive effect in listeners, which was carried out with a view to an upcoming/planned qualitative analysis. According to preliminary results for the general text, 1.5 times as many such enrichments were tagged for the ELF (338) as for the EdE (222) renditions. This suggests that a very real need for clarification and greater explicitness was felt on the part of the professionals when interpreting from the ELF source speech and that they consciously endeavoured to provide additional information for their audience.

In principle, RT does have some potential to indicate differences in effort in the recovery of ELF speech:

[R]elevance theory does not provide an absolute measure of mental effort or cognitive effect, and it does not assume that such a measure is available to the spontaneous workings of the mind. What it does assume is that the actual or expected relevance of two inputs can quite often be compared. These possibilities of comparison help individuals to allocate their cognitive resources, and communicators to predict and influence the cognitive processes of others. They also make it possible for researchers to manipulate the effect and effort factors in experimental situations.

Wilson and Sperber 2016: 278

In our case, a triangulation of the results from the comparison between ELF and edited versions with those from other more cognitive demand-geared methods (applied in the project) may prove interesting.

5. Conclusion

As part of the larger-scale CLINT project, this paper looks at the renditions of two source texts (a more general and a more technical one), both in an original ELF and an edited version. The enrichments and completions made by 28 professional interpreters in their (56) target renditions are analysed based on a relevance-theoretic approach, and the renditions based on the ELF and EdE versions compared. The objective was to look into the way interpreters are said (see the introduction above) to improve or optimise source texts, especially those produced by non-L1 speakers (of English). By analysing, completion by completion, how exactly interpreters flesh out source input in order to recover the speaker’s intended message and to enhance its relevance for the target listeners, interpreters’ augmentation efforts can be illustrated. This is particularly relevant in the context of ELF where the (non-standard) linguistic structures are not only underdetermined (as is the case with any language contribution), but may be incomplete, unconventional or insufficient in a number of ways. The results show that interpreters do engage in a considerable number of enrichments and completions and that this number is significantly higher in the renditions based on the ELF as opposed to the EdE versions of the two texts (up to 100% in the technical and around 50% in the general text).

The significantly higher number of enrichments in the renditions of the ELF as opposed to the edited versions of authentic conference speeches suggests that interpreters strive to optimise ELF source input and to maximise relevance for their listeners. Whether this extra completion effort on the part of the interpreters actually adds to cognitive effort invested can only (and will) be corroborated by triangulation with other methods that are used in the CLINT project for the measurement of the cognitive effort expended by the interpreters (see Ehrensberger-Dow, Albl-Mikasa, et al. 2020). Whether or not this optimises a target version as compared to its source can only be determined by an in-depth quality rating of the target texts and may require the cognitive effects obtained by target audience listeners to be tested, for example in the form of Reithofer’s (2013) comprehension tests (see above). However, since we are dealing here with professional interpreters with a vested interest and self-imposed drive to enable smooth communication, it can be assumed that the enrichments expressed are attempts at providing optimal target renditions and, in the case of ELF input, optimising the partly incomplete and unsatisfying source—as is also suggested by sub-tagging for enrichments that seem to harbour obvious potential for producing a cognitive effect in listeners (these tags are to be subjected to qualitative analysis as part of a follow-up analysis of the data). In other words, the enrichments can be construed as an attempt at maximising relevance for the audience of the ELF speech, potentially at the expense of additional resources.

The following outcomes of the analyses presented seem notable: (1) The principles and mechanisms of fleshing out the underdetermined linguistic input of a source speech as provided by RT are a particularly useful instrument for a demonstration of interpreters’ actual completion activity (reference assignment, disambiguation, closing of elliptical gaps, specifications, general enrichments, etc.) in the target text and supply preliminary cues as to the potential effort involved. (2) RT instruments can be fruitfully applied to the analysis of interpreter renditions in general and renditions of ELF source input in particular. Describing enrichments as manifestations of interpreters’ cognitive processing provides insights into the processing of ELF and into why and how recovery of explicatures from ELF speech input may be less straightforward. (3) The RT framework provides a conceptual foundation upon which to explain what may be different about ELF input processing. It also allows for informed guesses as to the cognitive effects interpreters may be able to generate in listeners.