The three output variables

This study identifies three features as indicators of difficulty in translation or interpretation – reordering, nesting changes and changes in semantic relations. Each of those features involves the linear or hierarchical relations between propositions. So we’ll start with a brief look at why the proposition was chosen as the unit of analysis for this study and at relevant research into propositions, before considering literature on each of the identified indicators of difficulty.

2.3.1 Choosing a unit of analysis

There are many linguistic approaches to segmenting discourse, with various units of analysis depending on their intended use. This study measures rates for indicators of difficulty in translating or interpreting complex sentences. So there were two main criteria in looking for an appropriate unit of analysis. One was whether the unit chosen could be the basis for a method of displaying and measuring such indicators in any language. The second was whether it would lend itself to cross‑linguistic comparison – specifically, whether it would reflect the similarity in function between different grammatical structures which languages can use to describe a given event or situation.

Syntactic analysis in the tradition of generative grammar considers relations between morphemes and other grammatical constituents. But those constituents are often too small to be suitable for analyzing complex sentence structure. Plus they don’t meet the criterion of cross‑linguistic comparability, because they highlight grammatical differences rather than functional similarities. Several methods of discourse segmentation, including the linguistic discourse model (Polanyi 1988) and centering theory (Grosz et al. 1995; Taboada and Hadic Zabala 2008), focus on the finite clause. But that doesn’t meet the criterion of cross‑linguistic comparability either, because an event or situation can be expressed in a finite clause in one language and in a non-finite clause or a clause-like nominal structure in another language. Tanguy et al. (2012) use the finite clause, together with the prosodic unit or punctuation, to define minimal discourse units. But prosody and punctuation were considered too language-specific and hard to compare for the purposes of this study.

Other methods analyze dialogue using main clauses with subordinate clauses, known as “T-units” (Hunt 1965), or main clauses with modifiers, known as “C-units” (Hughes et al. 1997), or utterances (Passonneau and Litman 1996). But dialogue isn’t typical of language that’s generally translated or interpreted. Also, dialogue doesn’t usually involve sentences with the degree of complexity associated with the type of difficulty under consideration. Some approaches to segmentation use dependency units, based on Tesnière’s dependency grammar (Kintsch 2013; Mel’čuk 1988). But those approaches focus mostly on constituents of clauses, rather than relations between clauses or clause‑like structures. The method proposed by Poesio (2000) treats both finite and non‑finite clauses as discourse units. But an event or situation described in a finite or non‑finite clause in one language can also be described in a nominal structure in another language. What this study needed was a method which treats clause‑like nominal structures in the same way as finite and non-finite clauses.

The unit of analysis finally chosen was the semantic proposition. A proposition refers here to the set of relations among entities in an event or situation established by a logical function – a predicate.

2.3.2 Propositions

Dryer (2007) identifies several different types of predicate, including transitive and intransitive verbs, adjectives, nominals and locative adpositions. Evidence for the propositional status of some nominal structures is given by Grimshaw (1992), who distinguishes “complex event nominals” with argument structure from “simple event nominals” and “result nominals” without argument structure. Alexiadou (2010) cites several researchers who analyze semantic event structure in gerunds and deverbal nouns.

Cosme (2008) observes: “What is expressed by a phrase in one language may be expressed by a clause in another language…. Translating from English into German frequently triggers … a phenomenon whereby a clause in the source language is turned into a phrase in the target language” (p. 91). Doherty (1999: 115) gives examples of this phenomenon in German-to-English translation.

Lyons (1995) sees propositions expressed in different syntactic forms as equivalent if they meet the same truth conditions. He also discusses “whether and how the traditional classification of subordinate clauses as nominal, adjectival, adverbial, etc., should be reflected in the formalization of the propositional content of complex sentences” (p. 159). This is the basis for classifying subordinate propositions as complement (nominal), modifier (adjectival) or adjunct (adverbial) in this study.

One dimension in which the arrangement of propositions can change is horizontal: with changes in their linear order or changes in structures where one proposition is nested inside another. The other dimension in which the arrangement of propositions can change is vertical: with hierarchical changes in the way one proposition is subordinated to another. Let’s now look at research into these three types of change and the cognitive difficulty associated with each one.

2.3.3 Reordering and difficulty

“Reordering” in this study means reordering of propositions: a translator or interpreter’s need or choice to move a proposition from where it was in the original version of a sentence to an earlier or later place in translation or interpretation, in relation to the other propositions. 

Several studies have shown reverse recall of verbal information to be more difficult than forward recall. Donolato et al.’s (2017) review of literature on the subject concludes: “In verbal span tasks, performance is worse when recalling things in backward sequence rather than the original forward sequence.” Similarly, experiments by Anders and Lillyquist (2013) and by Thomas et al. (2003) find reverse recall of information to be much slower than forward recall. Not to mention the difficulty of recalling reordered bits of split propositions.

The more the order of propositions changes from source to target language in translation or interpretation, the more the task of recalling them approaches totally reverse recall and, according to the above findings, the more difficult that task becomes. Chesterman (2011) examines the testability and implications of the “literal translation hypothesis,” according to which, “during the translation process, translators tend to proceed from more literal versions to less literal ones.” Schaeffer and Carl (2014) propose parallel word order as a criterion for “literal translation,” which they find to be associated with lower cognitive load than “non-literal” solutions in English-to Spanish translation. Birch et al. (2008) find reordering to be a strong predictor of translation “difficulty” as reflected in the performance of statistical translation engines.

Section 2.2 above cites a number of empirical studies on production difficulty in translation and interpretation associated with structural difference between languages. Several of those studies identify reordering as a specific source of difficulty. Those include the studies reported by Bangalore et al. (2015; 2016), which find differently ordered syntax to be associated with an increase in cognitive load, and by Vanroy (2021), which associates linear and hierarchical differences in corresponding word groups with difficulty in translation. Vanroy also proposes a tool for predicting the translation difficulty of a sentence based on various features of the language pair of translation, including the need for reordering.

Vanroy et al.’s (2019) empirical study of Dutch-English translation by professionals and students was cited above, in section 2.1.3. The illustration of reordering in that study involves the predicate, arguments and adjuncts of a simple sentence. By correlating such product data with physiological data recorded during the translation process, the authors find that “moving phrases around and re-ordering the translation requires more cognitive effort” (p. 936), and that “the more syntactic re-ordering has to take place, the more difficult a translation is to create” (p. 929).

The present study considers reordering on a larger scale – changing the order of propositions in a complex sentence. Reordering on that scale can be characteristic of translation and interpretation between languages from different families. If local reordering of elements within a proposition is associated with cognitive difficulty, as suggested by the studies cited in this section, it seems reasonable to assume that larger-scale reordering of propositions in a complex sentence is associated with a correspondingly greater degree of difficulty.

2.3.4 Nesting changes and difficulty

A “nesting” in this study means a structure where one proposition is syntactically surrounded by the predicate and arguments of another proposition. 

Hawkins (2014) posits a number of universal efficiency principles of the human language faculty, including a preference for minimal phrasal combination domains (PCDs). A PCD is defined as the smallest linear string required by the human syntax processor to link a phrase head to a directly subordinate constituent. A universal tendency towards minimal PCDs is reflected in studies of grammaticalized phrase order preferences across languages, including studies by Greenberg (1963) and Dryer (1992).

This tendency has been confirmed in several empirical corpus-based studies of subjective phrase order preferences across languages. Those studies have involved right-branching languages like English (Hawkins 2000), left-branching languages like Japanese and Korean (Hawkins 1994; Yamashita 2002; Yamashita and Chang 2001, 2006; Choi 2007) and languages with mixed branching direction like Cantonese (Matthews and Yeung 2001). In all languages and for all phrase types tested, the evidence indicates that “processing becomes harder, the more items are held and operated on simultaneously when reaching any one parsing decision,” and that “processing complexity and difficulty increase as the size and complexity of the different processing domains increase” (Hawkins 2014: 47). Chafe’s (1994: 69) findings on intonation units in spoken language also point to “a cognitive constraint on how much information can be fully active in the mind at one time.”

Such views are in keeping with Gibson’s (2000) dependency locality theory. That theory analyzes two kinds of human computational resource involved in the process of parsing a sentence: keeping track of a forming sentence structure (including unattached items) and integrating new items into that structure. According to the theory, the processing cost of linking two syntactic elements is a function of the linear distance between them. The theory is supported by cross-linguistic evidence involving English (Gibson and Ko 1999), Dutch and German (Bach, Brown and Marslen-Wilson 1986), and Japanese (Babyonyshev and Gibson 1999).

Karlsson’s (2007: 2) study on center-embedding (nested clauses) in various languages finds that “multiple clausal center-embedding is not a central design feature of language in use” and that “the maximal degree of center-embedding in written language is three.” Quirk et al. (1989: 1040) consider even doubly nested clauses to be ungrammatical and “completely baffling.” Karlsson’s study involves European languages, but he suggests that “it nevertheless seems reasonable to assume a more general validity.” In his view, constraints on nested clauses “have their ultimate basis in the material language-processing resources and limitations of the human organism … especially short-term memory limitations.” The languages considered in Karlsson’s study have largely right-branching subordinate clauses, so most of the nested clauses it finds are of the same type: a relative clause modifying the subject of its parent clause.

Hawkins (2014: 197), citing evidence from Tomlin (1986), finds cross-linguistically that “subjects precede verbs and their direct objects within VP regardless of VO or OV ordering.” This is supported by Primus’s (1999) generalization that higher syntactic nodes tend to precede lower ones, which may be the main reason why complex sentences with overt subjects in left-branching languages tend to have more nested structures than their equivalents in right-branching languages. This suggests that formal Japanese, Turkish or (to some extent) Mandarin can be structurally more difficult to process, even for a native speaker, than comparable discourse in a European language. Fodor et al. (1974) suggest that left-branching structures are harder to process than right-branching ones, as they require more working memory to retain elements and attach them syntactically. Uddén et al. (2022: 594) find evidence of greater cognitive burden in processing left-branching than right-branching complex sentence structures, as left-branching structures involve “many simultaneous non-attached constituents, a cost which increases with each non-attached constituent presented.”

2.3.5 Changes in semantic relations

“Semantic relations” refers here to the place or type of attachment between propositions in a complex sentence. A key criterion in determining the type of attachment between propositions is the distinction between functional coordination and subordination. So this part starts by looking at that often fuzzy distinction, before considering changes in semantic relations as an indicator of difficulty in translation or interpretation.

2.3.5.1 Coordination and subordination

Cristofaro’s (2003) analysis of coordination and subordination focuses on underlying meaning rather than grammatical form. Instead of the traditional unit of analysis for complex sentence structure, the clause, Cristofaro uses the state of affairs (the description of a situation or event). Her approach is based on Langacker’s (1991) definition, which sees two event descriptions as being in a relation of subordination if one imposes its “profile,” or functional type, on the other. Lambrecht (1994) gives a functional definition of subordination based on a distinction between presupposed and new or asserted information.

Those approaches represent a functional version of Mann and Thompson’s (1988) rhetorical structure theory. That theory states that, in an asymmetric relation between two clauses, one clause (the “nucleus”) is more salient than the other, subordinate clause (the “satellite”). The concept of subordination as a relation between foreground and background information is discussed in Reinhart (1984), Thompson (1987) and Tomlin (1985). Klein and Von Stutterheim (1991) see salient information as information contained in the “main structure” of a sentence, as opposed to non-salient information in the “side structure.” Stede (2008) refines this view, stating that either the main or the subordinate clause can be more salient, with salience shaped by a number of factors. Blüdorn (2008) agrees that semantically “nuclear” information can be contained in a subordinate clause, with “satellite” information in the main clause. Bril (2010) highlights mismatches between syntax, function and semantics in relations between linked clauses. Similarly, Visapää (2014: 169) sees Finnish relative clauses, though “grammatically subordinate,” as not always “functionally subordinate.”

Langacker (1987) sees nouns as denoting “things” and verbs as denoting “processes” involving things. He sees complement clauses and nominalized forms, such as infinitives, as sharing features of both nouns and verbs. Croft (1991) also makes a semantic distinction between prototypical nouns and verbs, based on features such as predicate-argument structure. Cristofaro (2003) considers that functionally subordinate states of affairs are perceived differently from functionally independent ones. She sees the former as lacking an “autonomous profile,” so they’re “not scanned sequentially, but construed as a unitary whole, just like things” (p. 262). That’s why, in her view, subordinate states of affairs are often described using predicates with nominal features, though preserving some or all of their argument structure.

Several authors study main clauses with little or no propositional content, which are formulaic expressions of the speaker or writer’s attitude to an assertion in a subordinate clause. Such pragmatic markers have been analyzed by Brinton (2008) and others. Noonan (1985) calls the verbs in such clauses “propositional attitude predicates,” which he describes as conveying a subjective attitude towards an event. Thompson (2002) takes a similar view, arguing that most English complement clauses aren’t functionally subordinate, and that their syntactic parents should be seen as “epistemic/evidential/evaluative” fragments. Langacker (2014: 19) sees such clauses as syntactically independent but subordinate in prominence. He says: “In presenting such assessments, we are generally not describing them for their own sake, but rather as a way of indicating the epistemic status of the event the proposition pertains to” (p. 64).

Several researchers describe cases of mismatch between syntactic and functional subordination. Hooper and Thompson (1975) and others investigate so-called “main clause phenomena” in subordinate clauses which express assertions. But they suggest that such cases are the exception, rather than the rule. Others describe areas where it’s hard to give a black-or-white answer as to whether one clause subordinates or is subordinate to or coordinate with another clause in functional terms.

Dixon (2009) sees most pairs of asymmetrically linked clauses as having one semantic “focal clause” and one “supporting clause.” He finds that there’s often a mismatch between syntactic and functional subordination. And he sets criteria for deciding which of two linked clauses is functionally subordinate to the other: a conditional clause is a “supporting clause,” the second of two clauses describing successive events is a “focal clause,” etc. Other approaches are similarly categorical, though they don’t always agree. For example, Dixon (2009) sees “but” as introducing a “focal clause”, while Karlsson (2007) assigns the same link a coordinating role.

Verhagen (2001: 349) sees the main clause in many sentences as conceptually dependent on the subordinate clause, which often “provides the most important information.” He views the main clause in such sentences as “evoking … a mental space for the contents of another clause,” describing this as “a very general function of matrix clauses of complements” (p. 346).

Culicover and Jackendoff (1997) discuss cases of “semantic subordination despite syntactic coordination,” focusing on left-subordinating uses of “and” in English. Yuasa and Sadock (2002) expand on such cases of syntactic coordinators with subordinating function, which they refer to as “pseudo-subordination.” Ross (2016) provides a cross-linguistic analysis of this phenomenon, which he calls “pseudocoordination.” Evans (2007), Schröder (2016) and others analyze the reverse phenomenon – a clause with the syntactic structure of a subordinate clause functioning as the main clause of a sentence – which they refer to as “insubordination.”

Andersson (1975: 45) distinguishes between syntactic and semantic subordination, characterizing clauses which don’t “make a statement, ask a question or give a command” as semantically subordinate. Verstraete (2007) defines a subordinate clause as one which doesn’t represent a full speech act. The same author (2005) discusses clause links which can function either as coordinating or subordinating, using the criterion of assertive force to distinguish between the two types.

Some languages have clause-chaining constructions, where one or more clauses are syntactically subordinate to a main verb, but each clause has similar semantic weight. This phenomenon, which is syntactically like subordination and functionally like coordination, was initially seen by Foley and Van Valin (1984) as a third type of clause link, which they called “cosubordination.” Foley (2010) later revised that view, seeing such constructions as a type of coordination. Bickel’s (2010) cross-linguistic research supports this latter view.

Cosme (2008: 90) sees coordination and subordination as “the two prototypical poles of a clause linking continuum allowing for a number of in-between constructions.” Bril (2010: 5) cites asymmetric features in coordinating links as evidence of hierarchical structure in relations of clause coordination. Paul (2016: 185) sees links between clauses in Mandarin as “a challenge for the traditional analysis of complex sentences into a ‘subordinate’ and a ‘main’ clause.” Cristofaro (2003: 22) says that “clause linkage types should not be described in terms of the binary opposition between coordination and subordination,” but instead form a continuum between the two. Herlin et al. (2014: 2) agree that relations of coordination and subordination represent “a continuum rather than a dichotomy (symmetry/ asymmetry).” Langacker (2014: 68) shares this view, concluding: “There is no point asking whether a certain clause is or is not subordinate, for usually there is no simple answer.”

In translation from one language to another, a syntactic subordinator can be changed into a coordinator and vice versa. Fabricius‑Hansen (1999) provides examples to support her claim that some languages, like German, favor syntactic subordination, while others, like English and Norwegian, favor coordination. Cosme (2008) concludes from a corpus analysis that French prefers syntactic subordination more than English, and English in turn more than Dutch, which prefers coordination. She concludes that such tendencies have a “clear impact on the translation process.” Solfjeld (2008: 116) examines the effects of raising subordinate clauses to coordinate status in German-to-Norwegian translation, concluding that this can lead to a loss of syntactic clues and misunderstanding of background information as salient. The tension between coordination and subordination in translation is illustrated by Ramm (2008), who examines the syntactic upgrading of German non‑restrictive relative clauses to independent sentences in Norwegian. She says that this “almost always changes the referential structure in the translation in some way” (p. 148), including problems in interpreting the following clause because relations of salience are lost.

A special type of subordinate clause with semantically independent function is a continuative relative clause (a non‑restrictive relative clause that continues a narrative). Ramm (2008) characterizes such clauses as functionally coordinate rather than subordinate. So do Aho, Sethi and Ullman (1986: 53), who say that a series of such clauses is “always convertible to iteration.” Langacker (2014: 19-23) notes that these clauses show “prosodic seriality,” suggesting a “flat structure with no internal grouping.” But the view of such clauses as functionally coordinate is debated. Brandt (1990: 128) says: “The subordinate clause form of a continuative RC is used to indicate that an information unit is less important than its matrix clause.” Becker (1978) reports that she’s never met a German speaker with an intuitive understanding of the difference between restrictive and non‑restrictive relative clauses. Visapää (2014: 168-169) argues that “the subordinate status of non-restrictive relatives should also be seen as a continuum” and that “non‑restrictiveness never equals non-subordination.”

Langacker (2014: 69) observes that “grammar proceeds through time…. The static formulas and diagrams employed in describing it should not obscure the fact that grammar is something that happens” (emphasis in original). His dynamic view of cognitive grammar posits a “primary window” of attention, which opens successively onto each clause in a chain. He sees chains of clauses in relations of syntactic subordination as cognitively requiring no “grammatical constituents larger than clauses” (p. 14), with the assembly of such clauses requiring “only local connections, with no more than two clausal processes appearing in any single window” (pp. 26-27). In his view, “conceptual access proceeds on a clause-by-clause basis, with each clause being interpreted locally in relation to its head…. The clauses are successively accessed in basic level windows, each locally connected to the one that comes before” (p. 57).

Such flexibility between syntactic and functional status is harder to achieve in some languages than in others. That’s because subordinate clauses in some languages tend to be grammatically deranked – formed in a way that indicates subordinate, background function. Stassen (1985) identifies two ways in which a language can describe a functionally subordinate event or situation: grammatical deranking and “balancing” – using a verb form which could appear in an independent clause. Cristofaro’s (2003) functional analysis of subordination relies largely on the same distinction. She concludes that “the cross-linguistic coding of dependent [states of affairs], as manifested in the occurrence of non-independent clause-like patterns, reflects the cognitive status” of those descriptions as subordinate (p. 298).

Bril (2010) discusses the phenomenon of grammatically deranked subordinate clauses in various languages, including the use of participles and other non-finite verb forms, as well as nominalized verb forms with adpositions or case markers. Many languages tend to describe functionally dependent events or situations using structures with features more characteristic of nouns or adjectives than of verbs. Langacker (1987) explains this by pointing to a basic cognitive distinction between “things” and “processes,” with dependent events or situations perceived to some extent as things. Croft (1991) sees functionally dependent states of affairs as displaying non-prototypical correlations between the semantic classes and the pragmatic function of lexical roots.

2.3.5.2 Assessing semantic equivalence

Of the three target sentence features proposed in this study as indicators of difficulty in translation or interpretation, changes in semantic relations is the only one that potentially affects the quality of the product. A high count for reordering or for nesting changes may reflect a greater degree of cognitive effort than a lower count. But it doesn’t say anything about whether that effort was well spent. In contrast, changes in semantic relations between an original sentence and its translation or interpretation suggest that the product is in some way not equivalent to the original. So it’s useful to clarify what type of equivalence we’re talking about.

If I want to assess the equivalence of two things – two houses, two cars or two muffins – I need to choose one or more essential common features to compare – like their orientation, their gas mileage or their sponginess. Such a yardstick used to assess the equivalence of two things is referred to by some scholars, and in some translation research, as a tertium comparationis.

Various types of essential common feature for assessing translation equivalence have been proposed. Those features can be structural, semantic or functional. Jaszczolt (2003: 442) illustrates the difference between yardsticks for comparison with the following example of a possible English and a possible Polish response to a compliment on one’s appearance:

                     (1) Thank you.

                     (2) To tylko stara sukienka.
                           It’s only an old dress.

Clearly, there’s no structural or even semantic equivalence between (1) and (2). But each response may be similarly appropriate in its sociolinguistic context. So their equivalence can be assessed in terms of the pragmatic function they fulfill.

This study assesses the equivalence of different language versions of a complex sentence in terms of hierarchical relations of coordination and subordination between propositions. That yardstick is a mix of the structural, semantic and functional approach to assessing translation equivalence. Within a single proposition, assessment of equivalence is purely pragmatic. Since responses (1) and (2) above each consist of a single proposition and can be seen as fulfilling the same pragmatic function, they would be assessed as equivalent by the yardstick of preserving semantic relations. That assessment would be different if one of the responses contained a second proposition, as in (3):

                     (3) To tylko stara sukienka, co znalazłam w szafie.
                           It’s only an old dress I found in my closet.

The semantic relations yardstick used in this study would assess response (3) as containing an extra proposition not corresponding to any proposition in (1). That difference would be recorded as a single change in semantic relations if (1) was an original English sentence and (3) was its Polish translation or interpretation. The semantic relations yardstick used here is pragmatic at the local scale, within a single proposition, where both structural and semantic changes are considered relatively inconsequential as indicators of difficulty. But the yardstick is semantic at a larger scale, in assessing the equivalence of hierarchical relations between propositions in a complex sentence. Changes in semantic relations at that scale are taken here as relevant indicators of difficulty. Even at that larger scale, the yardstick is still functional. It assesses the equivalence of functional relations of coordination and subordination between propositions, rather than the syntactic structures in which those relations are expressed.

This difference in criterion at different scales results from the choice to use the semantic proposition – the semantic equivalent of a syntactic clause – as the basic unit of analysis for investigating structural difficulty in translation and interpretation in this study. That choice was made to capture the semantic equivalence between propositions as expressed in different syntactic structures in different languages. It was the basis for the semantic parsing method developed for this study. The theoretical foundation and functioning of that method are explained briefly in section 3.1 and detailed in annex 1.

In using the semantic proposition as the basic unit of analysis, this study assesses equivalence between original and translated or interpreted versions of sentences on a larger scale than is used in many studies of related phenomena. Vanroy et al.’s (2019) study of difficulty in English-Dutch translation and Bangalore et al.’s (2015; 2016) study of syntactic variation in translation, both cited earlier, largely consider the arrangement of elements within a single clause. Perhaps the most comprehensive and authoritative repository of language typologies is the World Atlas of Language Structures (WALS). A few classifications in the WALS involve the typical branching direction of subordinate clauses, as used in this study. The large majority of typological features analyzed there involve morphemes, words and phrases on a more local scale.

2.3.5.3 Changes in semantic relations and difficulty

Associating difficulty with reordering and nesting changes is supported by empirical evidence from fields outside translation and interpretation, as explained above, in sections 2.3.3 and 2.3.4. The argument for associating difficulty with changes in semantic relations is more pragmatic and inherent to translation and interpretation.

Larson’s (1984) theoretical guide to translation technique uses the same unit of analysis as this study – the semantic proposition. Her method involves first rewriting the text to be translated as a set of propositions, then translating those propositions appropriately into the syntax of the other language. This is done to eliminate the “skewing between semantics and grammar” (p. 288) in the original text – the various types of syntactic-semantic mismatch described above. Larson strongly advises translators to try to preserve the semantic relations within and between propositions, regardless of syntactic form, as the key to preserving meaning in a successful translation. By that token, a change in semantic relations in translation or interpretation can be taken as a sign that the translator or interpreter has encountered some sort of difficulty that has prevented them from reproducing the original relations between propositions in the target language.

Few empirical studies I’m aware of consider difficulty in translation or interpretation as reflected in changes in hierarchical relations between clauses or propositions.

Vanroy (2021) reports on a series of experiments involving English-to-Dutch translation. That study finds a significant association between changes in hierarchical dependency relations among syntactic constituents and translation difficulty. From this finding, Vanroy concludes that a “change in the relationship between words requires more processing effort from the translator” (p. 22). The author discusses various ways of measuring the syntactic equivalence of a source and target text, which he sees as consisting in a combination of reordering and changes in hierarchical relations. He concludes that such changes make the translation process more difficult.

I’m not aware of any empirical studies to date that have recorded changes in semantic relations between clauses or propositions in translation or interpretation between languages from different families.

2.3.6 Is different always difficult?

For translation and interpretation, similarity between source and target language can refer to similarity in words or in structure. Words which are similar in form and have a shared etymology are cognates. Similarity in structure implies parallel ordering of syntactic constituents. At first sight, such similarity in lexical or syntactic form may seem to make the task of translation or interpretation easier. But is that always the case?

Let’s start with cognates. De Baets and De Sutter (2022: p. 9) cite several empirical studies which show that cognates, when available between two languages, are easier than unrelated words for bilinguals to produce and process. The authors cite a study by Egan (2012) involving Norwegian translations of the English words “begin” and “start.” That study suggests that translators are drawn towards using cognates when they can (p. 5).

On the other hand, the same authors cite studies by Shlesinger and Malkiel (2005) and by Malkiel (2009) which find that translators tend to “prefer a non-cognate translation over a cognate translation, even when both are equally good equivalents for the source word” (p. 10). De Baets and De Sutter suggest that this may be partly because student translators are taught to mistrust cognates as potential “false friends.” Granger and Swallow (1998) give a detailed instructional overview of the translation pitfalls posed by many cognates in English and French. Translators and interpreters, it seems, may consciously choose to or need to avoid cognates, even if doing so makes their task more difficult. This could suggest seeing the existence of cognates between source and target language as a source of difficulty.

How about similarity in the ordering of syntactic constituents? As described above, in sections 2.2.1.3 and 2.2.1.4, there’s a growing body of evidence associating syntactic difference in a language pair with difficulty in translation and interpretation. At the same time, it’s plausible that translators and interpreters may choose to or need to avoid parallel order in local syntactic structures, for reasons similar to their reasons for avoiding cognates, even if doing so makes their task more difficult.

Sometimes going for a cognate or a parallel syntactic structure, if one is available, can be the best option in translation or interpretation. Especially under the time constraint of simultaneous interpretation, that option may be easier than having to choose a formally different word or structure. But sometimes going for the easier solution can lead to a less appropriate result in the target language. In that case, it’s worth the extra effort to find a word or structure that’s different in form and more appropriate.

But that doesn’t mean that it’s harder to work in a language pair with cognates or parallel syntactic structures than in another type of language pair. Translators and interpreters working in certain language pairs may often have to make choices between easier, potentially less appropriate lexical and structural solutions and harder, potentially more appropriate ones. But that doesn’t mean that working in a language pair without such similarities is necessarily easier. It just means that a pair like that doesn’t offer the potentially misleading easier options.

Whether or not having to resist the temptation of cognates or parallel local structures increases difficulty, this study doesn’t consider choices of words or local syntactic structures. Rather, it analyzes larger-scale structural changes – changes in the linear and hierarchical arrangement of propositions in a complex sentence.

The first two types of such structural change – changes in linear order and in nested structures – are easier to make in written translation than in simultaneous interpretation, because there’s less of a time constraint. Despite the extra effort involved, a translator may choose to change the linear arrangement of propositions in order to preserve the semantic relations between them. But in simultaneous interpretation, the working memory constraint exerts strong pressure on interpreters to produce parallel complex sentence structure to the original speech. Under that pressure, an interpreter is less likely than a translator to make the extra effort invoved in changing the linear arrangement of propositions in order to preserve the hierarchical relations between them. So the meaning of the original message is more liable to be altered in interpretation than in translation.

Again, few empirical studies to date have investigated the effect of structural difference in a language pair on difficulty in translation and interpretation. And I’m not aware of any other studies on that difficulty as affected by major typological differences between languages in complex sentence structure.

This literature review has considered research on: (1) sources of difficulty in translation and interpretation and ways of measuring that difficulty; (2) the three input variables analyzed in this study – structural difference between languages, mode of language transfer and sentence complexity; and (3) the three output variables – the three features identified as indicators of difficulty in translation or interpretation.

The study records values for those variables for each translated or interpreted version of each sentence in the corpus, based on the linear and hierarchical arrangement of propositions. The next chapter explains how that’s done. It starts with a description of the parsing model used to segment sentences into propositions and to indicate the semantic relations between them.