Appendix C

Examples of Common Causes of Mistranslation

This appendix will review the most common causes of mistranslation, with special attention to issues unique in translation of survey instruments, providing examples drawn directly from the ESS Translation Guidelines for Round 7 (European Social Survey, 2014).

Interference: False friends (lexis)

Translators can be misled by so-called ‘false friends’. These do, of course, differ from one language to another: simply looking at the surface structure of language, translators may, for instance, decide that “intimate” will be translated as “intim” in German or as “intiem” in Dutch. While this may sometimes be true, in other cases, this may not work as both German and Dutch have a number of additional translations for the English word “intimate”, depending on the exact intended meaning. Words that sound similar across languages may (a) cover the same scope of meaning, there may (b) be overlap in meaning or these words may (c) have different meanings. Therefore, translators should be aware that a similar sounding word may not be what is appropriate in a given context (although in some cases it certainly can be appropriate).

Example 1 (item E27, ESS Round2):

How often, if ever, have you…...misused or altered a card or document to pretend you were eligible for something you were not?

In Example 1, some countries produced a similar sounding translation for “card” (“Karte”, “caart”). Independent assessors of these translations were unsure about the meaning of “card” in this context in the source text (In this context, ‘card’ refers, for example, to ‘Identity Card’), and were even more uncertain about the translated versions (“Karte”, “caart”), which did not make sense in the context.

Interference: Grammar and syntax

Being concerned about a comparable translation, translators may sometimes stick too closely to source text structures, thereby neglecting the usual target language requirements and the usual way of forming sentences in the target language. Look out for fluency and clarity in the target language while at the same time taking into account comparability requirements, i.e. faithfulness. A noun is not always rendered by a noun in the target language, a singular noun not always by a singular noun and an adverb not always by an adverb. Syntactical structures may equally change. For example, “information” is a typical English singular word that often gets translated by a plural noun in other languages.  

One-to-one equivalencies and their fallacies

Translation documentation from previous rounds has shown that translators occasionally use the words that typically or automatically come to their mind as one-to-one equivalencies. It is an erroneous belief, however, to think that word ‘x’ in the source language always leads to word ‘y’ in the target language (Hönig, 1997). “Government” can have different translations, “work” can have different translations, “job” can have different translations and “reasonable” can have different translations, depending in each case on which of the meaning dimensions of the source language – in this example, English language – words get activated in the given context.

Inexperienced translators are especially prone to using one-to-one equivalencies without further questioning the deeper meaning of the source text (Krings, 1986). For this reason, it is of utmost importance to assemble in a team people with excellent translation and language skills.

Careless reading

There have been cases in the past where careless reading has led to mistranslations. Rather than translating “wealthy” one country translated “healthy” and then others copied this through the shared languages consultations. Rather than translating “wanting a job” countries have translated “waiting for a job”. Parallel translation and review (and adjudication) are meant in particular to pick up issues such as these. These oversights can easily happen but one can expect that they are spotted in a carefully implemented team approach.

Imbalance between cognitive processes

To put it in psycholinguistic terms, understanding involves bottom-up and top-down processes. Bottom-up processes take the textual material as a basis, top-down processes activate world knowledge, experiences, etc. (Kussmaul, 1995; Kussmaul, 2007). Make sure that those processes are kept in balance. Too heavy use of top-down processes may lead to translations that divert too much from the actual source text and which, consequently, may compromise data comparability.

Example 2 (ESS 7 core item A3):

“Using this card, generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?”

is translated as:

“Using this card, generally speaking, would you say that most people can be trusted, or that you can’t be mistrustful with strangers?”

Following experience or stereotypes, the translator might have thought of strangers in connection with “can’t be too careful” and thus rendered the abstract term “people” by “strangers” (= top-down processes). He or she did not adequately take into account that “people” in the English source is not specified and so covers both people you know well and strangers, so the textual material itself (= bottom-up processes) was probably not adequately taken into account.

Omission of words or phrases

Translations that are fine from the translation point of view may not be strong enough from a measurement perspective. It is crucial for translators to refrain from omitting (or changing) any words or phrases that provide temporal, spatial or any other type of framework within which the respondent is requested to position their answer (e.g. last week; in general; on average; all things considered; mainly; very, as in ‘very old’ or ‘very weak sense’; about as in ‘about how many’, etc.). Omitting words or phrases of this kind would mean that the mental calculations from respondents in your country are not comparable to those elsewhere which in turn might compromise data quality

Example 3 (based on ESS 7 core item A1):

"On an average weekday, how much time, in total, do you spend watching television? "

In Example 3, a translator might be inclined to reduce the numerous adverbial references, assuming that any one of "average" or "in total" could usefully be omitted to make the sentence clearer. However, for example, if “average” was omitted, an important part of measurement would be lost; respondents might think of their most recent experience rather than taking into account their usual TV watching habits. Rightly or wrongly, the question designer presumably felt it important to include each of these phrases to "guide" the respondent in what to consider.

Example 4 (ESS 7 core item F31):

“What does/did the firm/organisation you work/worked for mainly make or do?”

In Example 4, if a translator in one country omitted “mainly”, that would mean that a respondent’s answer in target culture X would not be as focused on the primary tasks or functions of the firm as in countries where this was included. The respondent may say: “Well, there are many things to say. Which one should I list?” Or they might end up mentioning only one of the rarer functions and miss the main ones entirely. 

Errors can also occur if translators inadvertently change the form of conjunction. Conjunctions join together elements of thought, such as words, phrases or sentences. It is important that coordinating junctions such as “and” or “or” or “because of” are adequately rendered in the target language.

Example 5 (ESS 7 core item F36):

“Have you ever been unemployed and seeking work for a period of more than three months?” 

The conjunction “and” suggests that “seeking work” is to be undertaken while being unemployed. Translating the question along the lines of “being unemployed OR seeking work” does not tap the same concept as in the source text. It is crucial to maintain the original idea in translation.

It is also important not to omit interviewer / respondent instructions or any definitions provided to the respondent. For example, an interviewer instruction such as “CODE ALL THAT APPLY” indicates that several answers are possible. Without such an instruction, interviewers in some countries may believe that only one answer is possible and prevent the respondent from volunteering several answers. This would then compromise comparability between countries with different rules being applied. Being unsure of the meaning of certain words or phrases should never result in omitting them, i.e. in not translating them. As things currently stand, there is little basic research into how respondents specifically process questions with such multiple "signposts" .

Sequence

Translators should preserve the order of enumeration elements, listing multiple components in the target item in the same order as in the source item.

Example 6 (ESS 7 core item B12):

“During the last 12 months, have you done any of the following? Have you worked in a political party or action group?”

The translation in Example 6 should thus read “worked in a political party or action group” and not “worked in an action group or political party.” Intentional deviations should be documented.

Pronominal Systems/Frame of Reference

In contrast to English, many languages have complex pronoun systems that indicate number, gender, age, kinship or in-group/out-group relationships, and social status. A system is often abbreviated to a tu/vous distinction in French, distinguishing between “you, familiar” and “you, nonfamiliar.” Language-specific differences apart, adult users of languages with a tu/vous distinction address young children with the familiar tu form, and address all others with the more formal vous form. When one and the same questionnaire is to be used for different age groups, this can become a translation or version administrative logistics problem.

It is essential to consider who a survey question is asking about in instances of otherwise ambiguous pronouns. Is it the respondent themselves, the respondent’s partner, people in general, people like me, etc.? If the reference person differs between the source text and translation, this may lead to artifacts in the data that make comparison impossible.

Example 7 (ESS 7 core item C6):

 “How safe do you – or would you – feel walking alone in this area* after dark? Do – or would – you feel…”

*respondent’s local area or neighborhood 

Example 7 addresses the respondent personally (“you”). The item is thus about the respondent’s own feelings and not about others’ feelings. If one country translated this item in a very general way, that is, ‘How safe is this neighborhood after dark, walking alone?’, the data may not be comparable if general vs. individual perceptions differ. 

Some languages need to be more explicit than other languages such as English: in many languages ‘you’ can be translated in three senses: (1) the respondent personally (singular); (2) the respondent and any other people (plural); (3) ‘you’ in the sense of general statements, without referring to specific individuals.

Connotations

Be aware that words carry connotations, i.e. associations implied by a word in addition to its referential meaning. These connotations may then lead to unintended reactions on the part of the respondents, resulting in biased data. This may apply, for instance, to translations of “race.”

Another example comes from the European Value Survey (EVS): the Spanish scores for an item which measured loyalty deviated from the overall pattern of results for Spain. Upon examination it appeared that, unlike in other languages, the Spanish word for loyalty that was used in the translation had the connotation of “sexual faithfulness” (Van de Vijver & Poortinga, 2005). Take care that the translations used do not convey any ambiguous / unintended connotations that would distort the results.

Unintended Ambiguity

Be careful to not introduce unintended ambiguity during the translation process. If, for example, the source text asks how often the respondent ‘attends sporting events as a spectator’ and the translation provides a formulation that can equally well be understood as directly participating in sport activities themselves, then this translation option should be discarded. Clarity on the concept required from the item will be useful in making final decisions. Ambiguity can also result from syntactical ambiguity. Syntactical ambiguity can arise when respondents do not know which part of the question goes with which part. These links should always be made explicit to the respondent. For example, should the item “I really dislike answering machines” be understood as “I dislike answering” or as “I dislike the machines”? (Harkness, Pennell, & Schoua-Glusberg, 2004, p. 456).

Gender

Gender is an aspect that differs between many languages and therefore often causes problems in translation, also in questionnaire translation. 

Gender issues can have different forms.

a)    A language may require masculine and feminine versions of certain adjectives, nouns, etc. (Harkness, 2003Harkness et al., 2004) where the English language is gender-neutral.

Example 8 (ESS 7 core item B20):

“All things considered, how satisfied are you with your life as a whole nowadays?”

In Example 8, some languages may require both masculine and feminine versions for “satisfied”, e.g. in French satisfait and satisfaite. It would be good to clarify in advance how this gender issue should be dealt with in each country so that translators can accommodate the specified requirements when doing the first draft translations.

b)    Gender can also become an issue in other cases, as Example 9 demonstrates:

Example 9 (item D32, ESS Round 4):

“Using this card, please tell me whether you think doctors and nurses in [country] give special advantages to certain people or deal with everyone equally?”

In Example 9, “doctors” covers all doctors regardless of their sex and “nurses” covers all nurses who care for the sick or the infirm, regardless of their sex. In some languages and translations, the masculine form of “doctors” and “nurses” can be used to refer to both men and women because it can be used in a generic way. In other languages, one may need to find paraphrases in order to avoid making this item a gender-specific item: for example, “nursing staff members” could be used as a translation for “nurses.” However, care should always be taken to cover the intended meaning as succinctly as possible so that questions do not become too long.

c)     Similar issues also need to be taken into account when asking questions about the respondent’s partner. For example, in British English, the word ‘partner’ could refer to a partner of the opposite or the same sex. However, in some languages both feminine and masculine partners may need to be explicitly referred to in order to allow for all possibilities, e.g. in German “Partner oder Partnerin”.

Example 10 (SHARE)

Generic English Questionnaire: “Now I would like to ask you about any partners you may have had who you have not lived with. Have you ever been in a long-term relationship that was important to you, where your partner lived at a different address from you for most of the time?”

In the verification step of the translation process, a professional verifier commented: “National version excludes (from the point of view of grammar) possibility of man having a male partner or a woman having a female partner.”

Depending on the target language, some countries will need to decide whether to mention both masculine and feminine forms in order to be politically correct or to only use one of these forms. In this regard, the national teams should follow the line that is best accepted in the respective country. However, the aim is not to exclude one of the genders while at the same time avoiding making a question too complicated or too difficult to ask by continuously repeating both genders.

Response Scales

Translation of scales is among the greatest challenges in questionnaire translation, as response scales represent the data that is analyzed (Behr, 2009Harkness, 2003Harkness et al., 2004Harkness et al., 2010b). Several dimensions of response categories are addressed below.

Intervals: 

        Make sure that the intervals in the target text are comparable to the source text. If the source has no overlap or gaps, then the translated question should not have them either.

Example 11 (ESS 7 core item A1):

  • No time at all
  • Less than ½ hour
  • ½ hour to 1 hour
  • More than 1 hour, up to1 ½ hours
  • More than 1 ½ hours, up to 2 hours
  • More than 2 hours, up to 2 ½ hours
  • More than 2 ½ hours, up to 3 hours
  • More than 3 hours
  • (Don’t know)

For example, if, in the translation, the third category (“1/2 hour to 1 hour”) and the fourth category (“More than 1 hour, up to 1 1/2 hours”) both include “1 hour,” unambiguous assignment to a response category is not assured any more. If, in the translation, neither of those categories include “1 hour”, then the respondent would be at a loss as to which category to assign his or her answer of 1 hour.  

        Labels of categories

Try to produce labels which are as equivalent as possible to the source text and which work at the same time in the target language context.

a)     In this case try to mirror the intensity of scale points as expressed in the source language. For example, the translation of “quite interested” (cf. Example 12) should have a lower intensity than that of “very interested”, whilst “hardly interested” should be less in intensity than “quite interested” and so on. Make sure that the qualifiers (very, quite, etc.) chosen for the labels adequately convey the graduation required.

Example 12 (core item B1 in ESS 7):

“How interested would you say you are in politics – are you
very interested,
quite interested,
hardly interested,
or, not at all interested?”

b)     In Example 13 below, target country translators should produce labels that convey the intensity of “extremely”. “Extremely” is a fixed reference point, i.e., an extreme end point on the scale where nothing can go beyond it. The same extremity should apply to corresponding labels in the translations. A literal translation of “extremely” is not required, but rather the same “extremeness” – this might be represented in target languages also by ‘completely’, ‘fully’, ‘absolutely’, ‘totally’, etc. It is important to take into account that “extremely” should not be translated using a word equivalent to “very” because they do not have the same graduation i.e. ‘very’ has less intensity.

Example 13 (“extremely” scale):

Extremely bad

Extremely good

We would secondly also expect countries to produce a linguistically symmetrical scale in cases where the source language scale is linguistically symmetrical. By a linguistically symmetrical scale we mean: “extremely” on both ends of the scale.

However, experience – and literature – dictates caution: In some languages there may not be a close equivalent to “extremely” that collocates, that is, typically occurs in conjunction with the corresponding adjective ‘good’, ‘satisfied’, ‘happy’, etc. In addition, while “extremely” works with both positive and negative adjectives in the English language, in other languages there may not be an adverb available that can work at both ends of a scale. In these cases it may not be possible to employ linguistic symmetry.

However, what should normally be avoided is swapping between bipolar and unipolar scales (e.g. bad <-> good becomes not good <-> good). This decision should only be taken as a last resort and must be documented accordingly.

To get a better impression of the linguistic forces at work when translating response scales and to see where research is urgently needed and to support interpretation of results, if needed, countries should document their scale translation in case of an unavoidable deviation making use of an English rendering or explanation so that everybody in the project can understand the nature of the deviation.

Experience tells us also that where an English source language can use the same scale unchanged for a number of items (e.g., “extremely bad” – “extremely good”), this may not be the case in other languages; other languages may, for instance, need to adapt the adjective in gender and number to the corresponding noun, or different translations of the words ‘good’ and ‘bad’ may be required in different contexts. Also in this case, countries should document any deviations such as additional show cards added for such reasons.

c)     Experience has also told us that for some countries the translation of “not at all often” is problematic. Some countries may solve this problem by using an adverb in the form of “never” in a given context. In this case, countries should document any deviation such as this one.

Length of labels

Try to keep the length of labels as equivalent to the source as possible. This means: If the English label only contains individual words / phrases (extremely good, not at all, to some extent, etc.), do not produce entire sentences such as ‘I am not at all happy with the government’s work’. Contrary to that, if the English source questionnaire contains entire sentences as response category (e.g., “I plan for my future as much as possible” or “I never plan my future”), the translation should contain entire sentences as well rather than simply saying “as much as possible” or “not at all.”

In case this is, for linguistic reasons, not possible in the target language, documentation is essential.

Question beginnings

This paragraph refers to introductory phrases such as “To what extent”, “How difficult or easy …” or “To what extent do you agree or disagree …”.

“To what extent do you agree or disagree …” or “How difficult or easy …” is a deliberate wording technique in order to introduce the range of answer categories. Simply asking “Do you find it difficult or easy to …” or “Do you agree or disagree …” would not match the answer categories if those range from “Very difficult” over “difficult” and “easy” to “very easy”. So please try to match this open phrasing, if possible, in your language. However, in some languages this becomes very long and clumsy and may mean a too high burden for the respondents. In these cases, the reason for deviating from the English structure should be documented and a ‘lighter’ translation used.

In addition, try to the extent possible to mirror the deliberate balancing in your language (“agree”/“disagree”; “difficult”/“easy”). This balancing suggests to the respondent that all answers are equally valid.

If the question begins with an interrogative word (i.e., what, why, where, which, who, or how if English is the source language), try to reflect the meaning in the translation.

Example 14 (item B2 in ESS 4):

“How often does politics seem so complicated that you can’t really understand what is going on? Please use this card.

Never
Seldom
Occasionally
Regularly
Frequently
(Don’t know)”

Regarding example 14: A translation along the lines of “It is sometimes said that politics is so complicated that one doesn’t really understand what is going on” with the response categories translated as “I never have this impression”, “I seldom have this impression”, … would deviate without reason from the formal characteristics (WH-question) of the source text and should not be implemented.

Document any cases where this is, for linguistic reasons, not possible in a particular language.

Omission and addition of answer categories

Do not add or omit answer categories. This also applies to different types of item non-response categories: E.g., when the English source text only uses the “Don’t know” category, do not add “refuse” or “no answer” categories to your questionnaire. In fact, in the past, different approaches from countries on the number of item-non-response categories added have made research into item nonresponse quite difficult. Also please do not add answer categories. For example, you may feel that adding ‘farmer’ to an occupational answer list is necessary. But if this is only added in one country but not elsewhere this would be problematic.

Consistency between question and response scale

Question and corresponding answer categories should match linguistically.

Example 15 (item B39, ESS Round 4):

“And, using this card, would you say that [country]’s cultural life is generally undermined or enriched by people coming to live here from other countries? Response categories: Cultural life undermined vs. cultural life enriched”

In example 15, the translation chosen for “cultural life is […] undermined or enriched” in the question itself should also be used in the response categories. Be careful not to use different translations for “cultural life”, “enriched” or “undermined” in the question stem and response categories.  

Scale layout

Do not change the layout of the scale, e.g. a horizontal scale should never be changed into a vertical scale. Equally, do not reverse the order of the response categories, e.g. “extremely happy” – “extremely unhappy” should not become “extremely unhappy” – “extremely happy”. If form changes like these are made they would always be seen as a deviation.