The three input variables

Three input variables are analyzed in this study: structural difference in the language pair of translation or interpretation, mode of language transfer (translation or interpretation) and sentence complexity. This section looks at research into difficulty associated with each of those three input variables.

2.2.1 Structural difference and difficulty

2.2.1.1 The branching direction of subordinate clauses

Two attached clauses can be in a relation of syntactic coordination or subordination. There are three broad categories of subordinate clause generally investigated cross-linguistically: relative, complement and adverbial clauses (Gast and Diessel 2012: 1). In addition to coordination and subordination, Spronck and Nikitina (2019) argue for a third type of clause attachment – reported speech or thought – based on a number of cross-linguistic features.

A typology of languages in terms of the typical branching direction of relative clauses is given by Dryer (2013a) in the World Atlas of Language Structures. Dryer (2007) also provides a typology of languages in terms of the typical branching direction of relative and adverbial clauses. Schmidtke-Bode and Diessel (2017) classify languages in terms of the typical branching direction of complement clauses. Diessel (2001) does the same for the typical branching direction of adverbial clauses.

According to the above classifications, Indo-European and Semitic languages are in one group, with typically right-branching relative and complement clauses, and adverbial clauses which typically branch either way. Finno-Ugric languages are in a second group, with typically right-branching complement clauses and relative and adverbial clauses which typically branch either way. Sino-Tibetan languages are in a third group, with typically right-branching complement clauses and typically left-branching relative and adverbial clauses. Other languages, like Japanese, Korean and Turkish, are in a fourth group, with typically left-branching structures for all three major types of subordinate clause.

2.2.1.2 SVO-SOV

A smaller-scale typological distinction is Greenberg’s (1963) widely used six-way classification of typical word order in languages as “SVO” (Subject – Verb – Object), “SOV” (Subject – Object – Verb), VSO, etc. That distinction involves the typical order of the major elements of a clause, with most languages classified as SVO or SOV. Dryer (1997) argues for a typology based on two binary parameters: OV vs VO and SV vs VS. Those typological distinctions operate on a more local scale than the branching direction distinction in this study. Distinctions like SVO‑SOV or OV-VO involve internal clause structure, whereas the branching distinction used here involves the typical linear placement of a subordinate clause in relation to its parent.

In terms of internal clause structure, many German and Dutch subordinate clauses are typically verb-final (SOV or OV). After describing SVO and SOV languages, Dryer (2013b) explains: “A third subtype of language lacking a dominant order consists of languages in which different word orders occur but the choice is syntactically determined. For example, in German and Dutch, the dominant order is SVO in main clauses lacking an auxiliary and SOV in subordinate clauses and clauses containing an auxiliary…. Because this results in both orders being common, neither order is considered dominant here.”

There are some Indo-European languages, like Indo-Iranian languages and Armenian, where the verb can come at the end of a main clause too. Such head-final clauses are an exception to the head-initial structure characteristic of other large phrase types in Indo-European languages. One explanation for that exception is that verbs “face an additional challenger for initial position, because they alone of the lexical categories take specifiers” (subjects), which also claim first place in a clause (Baker 2003: 61). Rather than relegating a verb to second position after its subject – so the argument goes – languages like German (in subordinate clauses) or Hindi (in all clauses at formal register) prefer to keep a phrase head in extreme position and so send a verb to the end of its clause. In any case, the verb-final rules in such languages apply only at the level of an individual clause, not of an entire complex sentence. In contrast, left-branching languages like Japanese, Korean and Turkish not only have the main verb (along with indicators of tense, modality and negation) at the end of the main clause. They also place that main clause at the end of a complex sentence, after any subordinate clauses, in inverse order to that of Indo-European languages.

Despite this major typological difference, both Hindi and Japanese (and sometimes Hungarian or even German) are sometimes grouped together as “SOV” or “OV” languages, because of the typical position of a verb in its clause. But in terms of the branching direction of relative and complement clauses, Indo-European languages (including German) are all typically right-branching; Japanese, Korean and Turkish are left-branching; and the branching direction of Sinitic languages like Mandarin is mixed. This typological distinction lies at the heart of difficulty in translation or interpretation associated with structural difference in a language pair, as highlighted in this study.

2.2.1.3 Structural difficulty in translation

Section 2.1.1 above mentioned several studies indicating that certain features of a text can make that text inherently difficult to read and difficult to translate into any language. There is also some research suggesting that structural differences between can be a separate source of translation difficulty, independently of the difficulty associated with a specific text.

Some such research involves translation between different European languages. Referring to English-German translation, Nord (2005) analyzes various types of translation problem, including “linguistic translation problems,” which result from structural differences in the language pair of translation rather than from the content of any specific text. Experiments reported by Vanroy (2021) associate linear and hierarchical differences in corresponding word groups with difficulty in English-to-Dutch translation, as measured by eye-tracking and key-logging data. The author concludes that “diverging syntactic properties between a source and target unit cause increased translation difficulty” (p. 155). In studies of translation from English into Danish, German and Spanish, Bangalore et al. (2015; 2016) find that differently ordered syntax is associated with higher cognitive load, as reflected in reading time per source word, response time and total translation time.

In a study of Arabic-English translation, AlBzour and AlBzour (2015: 24) describe translation problems caused by “differences between the source language and the target language in terms of syntactic structures and semantic relations,” resulting in translation solutions that are “fully ungrammatical, odd and even absurd … where there is an abyss of syntactic and semantic differences between these two systems.”

Other analyses of translation difficulty involve translation between European and Asian languages. Carl and Schaeffer (2017: 55) find much higher degrees of syntactic variation between English and Japanese or Hindi than between English and Danish, Spanish or German. They say this makes the translation process much “more difficult and time-consuming.” Zou  (2016: 190) finds that the most difficult aspect of English-Mandarin translation is the translation of long, complex sentences, due to “difference in phrases and sentence structures.” In a study involving Mandarin and seven European langugages, Wong (2006: 124) concludes that “translating between the European languages is much easier than translating between Chinese and any one of the European languages,” with structural difference being a greater factor of difficulty than differences in vocabulary or culture. In his view, “this is because the translator is, during the translation process, constantly dealing with syntax in two directions: the syntax of the source language on the one hand and the syntax of the target language on the other.” Philippi (1989: 682) colorfully describes Japanese-English translation as “a process of radical demolition, in which a confused-looking mass of raw materials is dynamited and then reassigned, after considerable shifting of positions … to neat, new sentences.”

Some researchers propose special strategies to avoid distortion of meaning due to structural difference. Such strategies are examined by Chesterman (2000) in a general analysis of translation between structurally different languages. Muñoz Martín (2012) discusses mental load and coping strategies in translation. Lian (2006) describes such strategies in English-to-Mandarin translation. Others confirm such techniques as applied to a specific translation task: Yang (2010) for a work translated from Finnish into Mandarin, and Ikbal et al. (2016) for a translation from English into Arabic. These strategies generally involve changing the hierarchical structure of sentences to ensure smoother reading and better rhetorical effect. That suggests there may be a trade-off between structural accuracy and readability.

The studies cited above consider structural difficulty in translation involving Indo-European languages (or languages with similar complex sentence structure, like Arabic). A few studies involve European languages and Mandarin or Japanese. I’m not aware of any empirical research to date involving parallel translations into languages with a representative range of complex sentence structure, which is at the heart of the structural difficulty investigated in this study.

2.2.1.4 Structural difficulty in interpretation

Difficulty associated with structural difference between languages has been an increasing topic of research into simultaneous interpretation, because of the added burden which some authors see structural difference as placing on an interpreter’s working memory.

Nowhere is that difficulty more strongly questioned than in Seleskovitch and Lederer’s (1989) influential théorie du sens or “interpretive theory.” Central to that theory is the notion of “deverbalizing” – processing a message through a language-free stage between understanding and reformulation. The interpretive theory maintains that simultaneous interpretation “hardly differs from one language pair to the next” (p. 137). This view is supported by pointing to interpretation between German and other European languages: “The success of simultaneous interpretation [from German into French] shows the validity of the interpretive method applied to a syntactically very different language pair” (p. 149).

The interpretive theory maintains that observations of interpretation between languages like English, French and German can be extrapolated to interpretation in any language pair. But other researchers disagree. They argue that major differences in the linear order of clauses may make complex sentences in some language pairs resistant to structurally accurate and coherent translation or interpretation. It’s this aspect of discourse, the “constraint of linearity,” which, in their view, the interpretive theory disregards (Gumul and Łyda 2007; Shlesinger 2014).

In that view, non-linguistic processing, though useful as a conceptual tip for interpreters, may not be able to obviate major differences in the order in which information is presented in a sentence, especially if the sentence has many subordinate clauses. Chomsky (2000) sees the clause as constituting a unit of logical processing, or “phase.” In a study of interpreters’ ability to recall the form of a sentence they’ve just interpreted, Isham (1994: 195) finds that interpreters have a greater tendency to process sentence information clause by clause than sentence by sentence, suggesting that “interpreters use the clause as their default unit of processing.” This supports findings by Goldman-Eisler (1972) in a study of input segmentation in interpretation, and by Davidson (1992) in a study of Japanese-English interpretation.

Setton (1993: 238) is critical of the interpretive theory, which he says has come in for increasing criticism, “especially from the Japanese sphere…. Apart from … the perceived ‘naively empirical’ nature of the theory, … cultural and linguistic factors are … swept aside by (largely … uninformed) dogma in support of the theory’s universality.” He also finds that structural factors “amplify certain difficulties beyond the point where they can be satisfactorily solved by strategies proposed hitherto” (p. 252). The same author (1999: 54) says: “Outside the [interpretive theory], almost all writers … with the relevant experience consider [interpretation from a left-branching language into a right-branching one] to present particular problems.” Setton and Motta (2007: 205) consider that “the famous notion of ‘deverbalization’ … has never been formulated with enough precision to satisfy everybody, or perhaps to be properly tested.” According to Gile (2009: 198): “While the relevance of language-specificity in interpreting has not been demonstrated empirically, arguments in favour of the hypothesis are strong, especially with respect to the implications of syntactic differences between the source language and the target language in simultaneous interpreting.”

In a review of empirical studies on difficulty in simultaneous interpretation, Christoffels and Groot (2005) suggest: “It is possible that the particular language combination involved influences the difficulty of interpreting: The more the two languages involved deviate from one another on the lexical, morphological, syntactic, semantic and pragmatic levels, the more difficult [interpretation] is likely to be.” The authors also hypothesize that “specific difficulties with certain language combinations can be expected for similar reasons. For example, syntactic differences between source and target language that force an interpreter to wait before formulating the target utterance tend to increase the load on the memory effort.”

As explained above, German is typologically similar to other Indo‑European languages in terms of the typical branching direction of subordinate clauses. Despite that large-scale structural similarity, many studies explore specific difficulties associated with interpretation between German and other European languages.

In an overview of German-English simultaneous interpretation, Wilss (1978: 343) suggests that “languages with predominantly parallel syntactic patterning, e.g. English and French, demand less syntactic restructuring than do languages which differ considerably in structure, e.g. German and English.” The author concludes that “transfer on the basis of parallel syntactic structures can … be regarded as easier to accomplish.”

Jörg (1997: 218-219) analyzes the tactic of anticipation in German-English interpretation, where he summarizes the specific challenge of that language pair as follows: “The underlying problem in German-English [interpretation], which often compels interpreters to resort to verb anticipation, is syntactic divergences between the two languages…. In German, the semantically relevant element of the verb phrase is often in end position … [so] waiting for the main verb may overtax the interpreter’s short-term memory and result in a loss of information…. One way out of this dilemma is the anticipation of the verbal component at the end of the sentence.”

In an overview of simultaneous interpretation strategies, including ones specific to certain language pairs, Riccardi (1999: 173) finds that “a synthetic language like German will involve greater difficulties in transfer to an analytic language like Italian, as more syntactic reformulation is required…. Syntactic restructuring is generally accompanied by greater cognitive load, unless automatic strategies for reformulating particular source language structures are acquired” [my translation].

In an analysis of data from German-French simultaneous interpretation published by Lederer, Van Besien (1999) finds anticipation to be a frequently used tactic. He summarizes his findings by saying (p. 251): “The fact that so many verbs were anticipated suggests that anticipation is a language-specific phenomenon. The material also contains cases of structural anticipation, a strategy which enables the interpreter to postpone the moment at which s/he has to produce a verb.”

In a study comparing strategies used by students of English-Italian and German-Italian interpretation, Donato (2003: 128-130) finds that “fewer restructuring operations are required in [interpretation] from English to Italian than in [interpretation] from German to Italian, where structural dissimilarities are deeper,” and that “the subjects’ ability to follow the [source language] structure … is indeed facilitated by closer morphosyntactic similarities between [source language] and [target language].” She describes various interpretation strategies such as “morphosyntactic transformation,” which she finds her subjects to use almost twice as often in interpreting from German to Italian as in interpreting from English to Italian.

In an empirical study of German-to-English interpretation, Seeber and Kerzel (2012: 238) find that “cognitive load during simultaneous interpreting of syntactically asymmetrical structures increased. These results are at odds with a universalist view of interpreting, according to which structural differences of the languages involved are irrelevant to the process.”

Another language sometimes considered to pose language-specific difficulties for interpretation from or into European languages is Arabic. Arabic and other Semitic languages are in the same typological group as Indo-European languages in terms of the branching direction of subordinate clauses (with typically right-branching relative and complement clauses and adverbial clauses which typically branch either way). One structural challenge of Arabic is the tendency for a verb to be placed before its subject in some types of clause. Like verb-final clauses in German, this feature operates on a local scale, involving the order of elements within a clause, rather than the order of clauses in a complex sentence.

In a study of simultaneous interpretation between English and Arabic, Al-Rubai’i (2004: 262) concludes that non‑parallel syntax between those two languages can “have an adverse effect on the quality of simultaneous interpreters’ performance.” In another study of English-Arabic interpretation, El-Zawawy (2022: 30) finds that “simultaneous interpreters operating from English into Arabic deal with complex structures in different ways by adopting and adjusting their strategies. The dominant strategy is linearization, which is triggered by the need to translate under time pressure. This strategy also unloads the simultaneous interpreter’s cognitive burden.”

The works cited above suggest that simultaneous interpretation of comparable speeches may be more difficult in language pairs like English-German, Italian-German or English-Arabic than in language pairs like English-Italian or English-French, for reasons specific to the structural difference of those language pairs. But that structural difference mostly involves the order of elements within a clause. So it operates on a much more local scale than the major differences which divide languages like Indo-European and Semitic ones from languages in several other families in terms of the typical branching direction of subordinate clauses. If local differences in the internal structure of a clause may be associated with increased production difficulty in translation or interpretation, as suggested in the above studies, it seems reasonable to suspect that much larger-scale typological differences are associated with correspondingly greater degrees of production difficulty.

To date, little empirical research has been done on difficulty associated with structural difference in simultaneous interpretation between languages with major differences in complex sentence structure.

Gile (2011: 34) reports on a comparative study analyzing interpretation of an English speech into French, German and Japanese. He finds that “there were more errors and omissions in the Japanese renderings than in either the German or French renderings.” The author concludes that this is “consistent with the tightrope hypothesis, according to which interpreters tend to work close to cognitive saturation, which also makes language-specific and language-pair-specific idiosyncrasies relevant parameters in the interpreting process.”

Ahn (2005) finds that complex sentences can’t be interpreted with sustained accuracy and coherence from Korean into English. She analyzes two types of syntactic management tactic which distort “perspective coherence.”

In a study of English-Mandarin interpretation, Wang and Gu (2016) find a high frequency of unnatural pauses, errors and inaccuracies, which they see as indicators of difficulty associated with “structural asymmetry” in that language pair. They conclude: “While waiting helps to earn more time for the interpreters to process and restructure the right-branching sentences, it is employed with a high risk of cognitive saturation which more often than not causes such problems as information loss or errors in their interpretations” (p. 13).

Yang (2002, as cited in Chen et al., 2015) proposes the principle of “syntactic linearity” as a specific interpretation tactic for dealing with the structural difference between English and Mandarin. That tactic consists in interpreting the elements of a complex sentence in more or less parallel order to the order in which they appear in the source language for as long as possible, then breaking to start a new sentence. This echoes the technique of “translating with the flow,” proposed by the “Beijing school” for interpreting from Mandarin into English. Zhuang (1991, as cited in Setton, 1999: 51), describes that technique as “flexibly selecting English words and phrases to follow closely the order of the Mandarin original; this sometimes sounds awkward, but it is still tolerable.”

2.2.1.5 Managing separate syntactic representations

Outside of translation and interpretation, evidence for cognitive difficulty associated with structural difference between languages is provided by studies suggesting that bilinguals process syntactic structures in their two languages more easily when those structures are similarly ordered in both languages.

Jacob et al. (2016) find that German speakers of English as a second language have easier mental access to English structures with the same constituent order and embedding level as in German than they do to structures where constituents are ordered differently or embedded at different levels in each language. A study of English-Spanish bilinguals by Hartsuiker et al. (2004: 204) suggests that, in the language mechanism of a bilingual person, “rules that are the same in the two languages are represented once.” A study of Dutch-English, Dutch-French and Dutch-German bilinguals by Hartsuiker et al. (2016) provides further evidence that bilinguals have “shared syntactic representations” of similarly ordered structures and different syntactic representations of differently ordered ones.

It seems reasonable to extend these findings to bilinguals whose languages are more structurally different than any two Indo-European languages. A translator or interpreter working between two languages with similarly ordered complex sentence structures is likely to use a shared syntactic representation in transferring content between those structures. In contrast, a translator or interpreter working between languages where subordinate clauses branch in opposite directions may be faced not only with problems of working memory, but also with the need to manage two different syntactic representations in reproducing the relations between differently ordered clauses.

In a sense, saying that a speaker of two languages with similarly ordered complex sentence structures has a shared syntactic representation of those structures is stating the obvious. It’s  like saying that a speaker of two languages with versions of the same alphabet has a shared representation of the writing system of both languages. Finnish is in a different language family from most other European languages. But a speaker of a European language learning Finnish is likely to learn easily that “että” is the Finnish word for “that” (as in “I know that he’s here”). That’s because they already have a mental representation of a parallel structure with an initial lexical complementizer (like English “that,” French “que,” German “dass,” etc.). In contrast, a speaker of a European language learning Turkish is likely to have a much harder time learning the corresponding Turkish structure. That’s because complementizing morphology in Turkish is neither initial nor lexical, instead involving nominalization, formation of the genitive, various possessive forms, plus other case endings depending on the type of complement taken by the verb in the main clause. And that main clauses follows its complement clause, in reverse order to all European languages.

Such differences in difficulty because of the existence or lack of shared syntactic representations suggest that a translator or interpreter working between languages with major differences in complex sentence structure isn’t just dealing with the task of reordering, changing nested structures or preserving semantic relations. At the same time, they’re also likely to be managing two different syntactic representations of each differently ordered relation between clauses.

2.2.2 Mode of language transfer and difficulty

The last section reviewed literature on the first of the three input variables analyzed in this study: structural difference in the language pair of translation or interpretation. This section reviews literature on the second of those input variables: mode of language transfer. Three modes of language transfer were considered in this study: legal translation, subtitle translation and simultaneous interpretation. The reasons for choosing those modes and research on each one are briefly discussed below.

2.2.2.1 Legal translation

As a genre of standard written translation, this study has chosen to focus on legal translation, as opposed to other genres such as literature or magazine articles. According to Verified Market Research (2024), legal translation accounts for a larger share of the global translation market than any other type of standard written translation, at 37% of that market in 2020.

One reason for choosing legal translation in this study is that legal texts often have long, complex sentences, which is where the translation difficulties highlighted here are most likely to appear.

In a descriptive corpus-based study of UN translations of English legal texts into Arabic, Abu‑Ssaydeh and Jarad (2016: 100-101) note: “An essential feature of English legislative writing is the high frequency of complex sentences; through the use of coordination and subordination, legislative English is capable of producing long, complex patterns which represent bafflingly intricate patterns that many translators find extremely challenging.” The  authors analyze various types of complex structure, including nominalization and syntactic discontinuity, which they consider typical of legal texts.

Lin et al. (2023) report on an empirical study comparing the syntactic complexity of UK company law texts and translations of Hong Kong and mainland Chinese laws from Mandarin into English. They find that “complex nominal and hypotactic structures [in source texts and in translations] result in a high number of propositions per sentence, placing a high demand on the cognitive processing abilities of those who read and understand the text” (p. 1).

Another reason the present study has chosen to focus on legal translation is that, in addition to difficulty in the translation process, output-related issues, such as distortions of meaning or coherence and comprehension difficulty for the reader, can have major consequences in legal translation. Each of the three legal texts analyzed here is prescriptive, is aimed at a large international readership and has far-reaching political and social consequences. So distortions and lack of clarity in their translations can cause serious problems for large numbers of people. In a discussion of issues surrounding equivalence and fidelity in legal translation in various legal and cultural contexts, Leung (2019: 157) cautions: “The risk is that a failure to achieve translation equivalence compromises legal certainty.”

In addition to sentence complexity, another common source of difficulty in legal translation is terminology. This is especially true of translations spanning different legal practices and cultural traditions. Ramos and Cerutti (2021) report on a comprehensive corpus-based empirical study of difficulty in legal translation for large international organizations. Text units were rated for difficulty by teams of experienced professional translators. Less common special terminology was identified as the greatest source of translation difficulty in all types of legal text considered. This type of difficulty isn’t examined in the present study, which doesn’t analyze word choice or phrasing within clauses.

Another characteristic constraint of legal translation is the requirement for accuracy and fidelity to the content and structure of the original text. Translators in most fields have relative freedom in recasting sentence structure, including sentence divisions, between the source and target language. But legal translators are generally instructed not to do so.

Two of the three legal texts considered in this study were produced by United Nations bodies. The UN Instructions for translators stipulates that “fidelity to the original text must be the first consideration” (Šarčević 1997: 16). Poon (2005) discusses cultural influences in a number of case studies of English-Mandarin legal translation. She advises legal translators to “endeavour to give a faithful translation closest to the meaning of the source text” by producing “a semantically and syntactically literal translation” (p. 322).

One reason legal translators are generally advised not to change sentence structure is to ensure faithful and accurate transfer of content. Another reason is to make it possible for readers of different language versions of a legal text to discuss the text using common references. For example, with an EU legal text, it’s important for people negotiating or interpreting different official language versions of the text to be able to cite, for example, the third sentence in section 3.2 and know they’re all referring to the same sentence.

A strong tendency to preserve sentence organization in translation is characteristic of texts produced by other organizations as well. In their study of techniques used in Arabic legal translation at the UN, Abu-Ssaydeh and Jarad (2016) find that “preference is given to maintaining the syntax of the source language rather than utilizing any alternative techniques such as movement, splitting the complex sentence into two or three simpler sentences or making any other syntactic changes.”

2.2.2.2 Subtitle translation

For subtitle translation, this study has chosen to analyze five different TED talks, as opposed to other types of subtitle translation, such as translation of subtitles for films or entertainment series. According to the TED Translate webpage, TED talks are currently translated into 115 languages by 50,856 translators. 238,200 translations of TED talks have been produced to date.

One reason for choosing TED talks in this study is that other types of subtitle translation tend to involve a lot of dialogue consisting of simple sentences, where the translation difficulties highlighted here are less likely to appear. In contrast, lectures by single speakers who are experts in their fields tend to have more complex sentences. Among online lecture platforms, TED is probably the most widely watched, with hundreds of millions of views.

Another reason this study has chosen to focus on TED talks is that the TED website offers a unique repository of parallel translations of single source texts into a large range of languages. Translation and interpretation corpora are often bilingual. If they’re multilingual, they tend to involve a much smaller number of languages, like the six official UN languages. Or they tend to involve only European languages with similar complex sentence structure, like the 24 official EU languages, which aren’t representative of the structural difficulty considered here. I’m not aware of any other translation or interpretation corpus comparable in size to the TED repository, with parallel translations of single source texts into many different languages and language families.

The downside of TED talks for research into structural difficulty is that they’re generally delivered in a simple, conversational style. That makes for effective, engaging communication. But it also means that the talks consist of shorter and simpler sentences than are characteristic of many genres of written text or more formal speech. Still, sentences in TED talks do tend to be longer and more complex than those typically found in other sources of subtitle translation, like films and entertainment series.

One of the main difficulties of subtitling for films and entertainment series is the appropriate transfer of culturally specific language. Such content typically shows scenes from various situations experienced by a community in one part of the world. Making those situations understandable and relatable to a broad public in another part of the world with different cultural norms and references can be challenging.

Another characteristic constraint of subtitling is the limitation on space. The TED Translate Guidelines instruct translators to keep subtitle segments to a maximum of two lines and reading speed to a maximum of 21 characters per second. Fast, dense speech has to be condensed to fit that requirement. There’s also next to no room for explanations by the translator, as a translator of a standard written text can provide in a footnote or even a simultaneous interpreter can sometimes provide as a quick aside comment.

Several instructional guides to subtitling, such as the comprehensive work by Cintas and Remael (2020), provide useful advice on navigating those challenges.

In addition to the limit on segment length, there’s also a pragmatic structural constraint in subtitle translation between languages with very different structure, which I’ve seen in practice but haven’t seen described in research. Subtitle segments need to be timed so that they match the image on the screen. That can be a problem if a complex sentence in the original version of a talk contains two or more segments, each of which matches what the speaker is doing in the original version, but those segments should be ordered differently according to the syntax of the subtitle language.

For example, an English-speaker might start a sentence with a main clause describing how they gave each child in a classroom a piece of paper, while at the same time showing the audience a piece of paper. They might continue the same sentence with a subordinate clause of purpose – “for them each to draw a funny shape on” – while at the same time picking up a colored pencil and starting to draw on the paper. Then they might pause, show the audience the shape they’ve just drawn, and end the same sentence … “like this!” A structurally faithful Japanese translation of that same sentence would have to start with “like this” and end with “give.” But that structurally faithful translation would no longer match the video image. Instead, the subtitler may need to change the ordinary structure of a Japanese sentence to get their subtitles to match the speaker’s actions.

I’ve often seen this sort of problem lead to awkward phrasing in subtitling between languages with very different structure. Subtitlers are generally advised to keep translated segments short and easy to read. Except for such general advice, I haven’t found any studies which mention the specific difficulty of translating long, complex sentences in subtitles between languages with different structure. The instructions for subtitle translators on the TED website make no reference to that challenge.

2.2.2.3 Simultaneous interpretation

As a genre of interpretation, this study has chosen to focus on simultaneous interpretation, as opposed to other forms of spoken interpretation such as consecutive, liaison, community or telephone interpretation. The main reason for that choice is that the working memory constraints which can have a major effect on the linear order and hierarchical structure of complex sentences, particularly in language pairs where subordinate clauses branch in opposite directions, are most prevalent in simultaneous interpretation.

Various types of difficulty associated with simultaneous interpretation were discussed above, in section 2.1.2. Difficulty in interpretating complex sentences between languages with very different structure was discussed separately, in section 2.2.1.4. Some of the strategies used to cope with that difficulty are illustrated later, in section 5.2.

2.2.2.4 Hypotheses for findings by mode

What does all this mean for the expected findings of this study on difficulty associated with the three modes of language transfer considered – legal translation, subtitle translation and simultaneous interpretation?

Legal translation is characterized by more long, complex sentences than the other two modes. And sentence complexity is associated with difficulty in translation and interpretation, as we’ll see in the next section. There’s also much less freedom for legal translators to shorten or simplify such sentences in the target language than there is for subtitle translators or interpreters. On the other hand, legal translators have more freedom to get the wording of the target sentence right, as standard written translation isn’t subject to the timing constraint of subtitle translation or the working memory constraint of simultaneous interpretation. Legal translators may also expect their readers to be used to dealing with long, complex sentences.

For all these reasons, greater indications of difficulty as reflected in restructuring and nesting changes are expected in legal translation than in the other two modes.

It’s harder to venture a hypothesis about indications of difficulty as reflected in changes in semantic relations in the three different modes of transfer. One the one hand, the long, complex sentences characteristic of legal translation could create lots of room for distortion of hierarchical relations in that mode. On the other hand, translators of legal texts have fewer timing or working memory constraints than subtitle translators or interpreters. They’re generally used to dealing with complex legal language. And they’re likely to be under instructions to render that language faithfully in meaning and in structure. In contrast, interpreters are under greater cognitive strain than translators, which may make them more prone to distorting hierarchical relations between propositions.

Depending on the balance of these factors, counts for difficulty as reflected in changes in semantic relations may turn out to be higher either in legal translation or in simultaneous interpretation than in the other two modes. The lowest counts for changes in semantic relations are expected in subtitle translation, as that mode is characterized by shorter sentences than legal translation and isn’t subject to the working memory constraint of simultaneous interpretation.

2.2.3 Sentence complexity and difficulty

In this study, the complexity of a sentence is measured by counting the number of subordinate propositions in it. Of the three input variables considered, this is the one most obviously associated with difficulty as reflected in our three indicators – reordering, nesting changes and changes in semantic relations. Those indicators each involve changes in the linear or hierarchical arrangement of propositions in the target language version of a sentence compared to its source language version. Naturally, the more subordinate propositions there are in a sentence, the more likely such changes are to occur. Linking sentence complexity as measured in this way to the three indicators of difficulty would hardly be revealing.

However, the statistical analysis in this study doesn’t consider the effect of any single input variable acting alone on the output variables. Instead, the study considers the effects of the three input variables – structural difference in language pair, mode of transfer and sentence complexity – in interaction. That’s because statistically significant interactions were found among all three input variables. When such interactions are found, reliable statistical analysis always considers the effects of those interactions. This is detailed in section 4.2.3.

What we want to find out about sentence complexity isn’t its effect on the three indicators of difficulty – which, again, is trivial. Rather, we want to see how sentence complexity interacts with the two other input variables – structural difference in language pair and mode of transfer – to produce a combined effect on those indicators. If structural difference and mode of transfer are found to be associated with the indicators of difficulty, each of those associations is expected to increase significantly with greater sentence complexity.

Several studies linking sentence complexity to translation and interpretation difficulty were cited above, in section 2.1. The theoretical studies by Nord (2005), Jensen (2009) and Sun (2015), as well as the empirical studies by Campbell (1999), Hale and Campbell (2002) and Sun and Shreve (2014), cited in section 2.1.1, all involve assessment of source text readability, including sentence complexity, as a potential source of difficulty in translation. The empirical studies by Darò et al. (reported in Christoffels and Groot 2005), Meuleman and Van Besien (2009) and Cai et al. (2018), cited in section 2.1.2, associate the processing of syntactically complex structures with indications of greater cognitive effort in simultaneous interpretation.

This section has reviewed research and discussions on translation and interpretation difficulty associated with the three input variables in this study: structural difference between languages, mode of language transfer and sentence complexity. The next section looks at research on our three output variables: the three features identified as indicators of difficulty in translation or interpretation.