Andrew Chesterman


2000l:      Empirical research methods in Translation Studies. Erikoiskielet ja käännösteoria (VAKKI-symposiumi XX) 27, 9-22.




Tässä artikkelissa esitellään empiirisen tutkimuksen pääpiirteitä, sovellettuina käännöstutkimukseen. Käännöstiede on “välitiede” (interdiscipline), jonka juuret ovat osittain hermeneutiikassa, osittain käytännön elämässä ja osittain ihmistieteissä. Se tutkii käännösten suhteita muihin teksteihin ja kääntämisen ajalliseen ja paikalliseen ympäristöön. Käännöstutkimus perustuu erilaisiin hypoteeseihin: niiden keksimiseen, perustelemiseen ja testaamiseen. Ns. tulkintahypoteesien avulla pyritään luomaan sopivia käsitteellisiä työkaluja käännösten kuvaamiseen, ymmärtämiseen ja selittämiseen. Kuvailevat, selittävät ja ennakoivat hypoteesit puolestaan ovat väitteitä, jotka ovat empiirisesti testattavissa.


Keywords: methodology, hypothesis, variable, model, categorization



1 The aims of empirical research


A classical statement of the aims of empirical research is given by Hempel (1952: 1, cited in Toury 1995: 9):


Empirical science has two major objectives: to describe particular phenomena in the world of our experience and to establish general principles by means of which they can be explained and predicted. The explanatory and predictive principles of a scientific discipline are stated in its hypothetical generalizations and its theories; they characterize general patterns or regularities to which the individual phenomena conform and by virtue of which their occurrence can be systematically anticipated.


Hempel’s position has often been taken to represent some kind of positivist extreme, in opposition to hermeneutics. Hempel’s view may be all very well for the hard sciences, it is claimed, but it is not applicable to the human sciences, nor to Translation Studies. However, I think Hempel’s points can be interpreted in such a way as to be highly relevant for Translation Studies, whatever kind of discipline we take it to be: (a) a hermeneutic study, like literary analysis or philosophy or aesthetics; (b) an applied discipline, like engineering or architecture; or (c) a human science, like sociology or psychology.  I take Translation Studies to be an interdiscipline, with elements of all these three types.


Consider the key terms in Hempel’s statement. First, particular and general. Any science seeks to describe particular instances of phenomena, but not only this: the aim is also to generalize, to abstract away from the particular in order to understand the larger picture. Some scholars in Translation Studies are interested in looking at what makes particular translations unique; others look for generalizations, patterns and regularities, even universal features shared by all translations.


Second: describing and explaining. Any science aims to describe, yes; but explaining is a more complex issue. Traditionally, a distinction has been made between explaining and understanding. On this view, explanation is the ultimate goal of a hard science, whereas understanding is the goal of the soft sciences. The reason for this distinction has been the link between explanation and causation: if you can explain something (in the hard sciences), this means that you have discovered its cause, in the strict, deterministic sense. In the soft sciences, this strict sense of causation is scarcely applicable. However, the distinction between explaining and understanding need not be interpreted so strictly. There are many ways of explaining, and many degrees of explanation; similarly, there are many kinds of causes, some more deterministic than others (even a vague influence can be regarded as a weak kind of cause). Understanding is not an absolute concept, either. Furthermore, the two concepts often overlap: if we think we can explain something, we tend to think that we can understand it. Understanding can also be interpreted more generally as the goal of an explanation — in fact, as the goal of any branch of knowledge. (For more on the relation between explanation and understanding, see von Wright 1971.)


If we can explain something, and hence claim to understand it (in some way), this does not necessarily mean that we can predict it. Sometimes, but not always. We can explain why volcanoes occur, but we cannot predict exactly when the next one will be, at a given place. In the human sciences, predictions are usually only probabilistic, at best. We often have to be content with vaguely anticipating, in order to lessen our eventual surprise.


Finally, Hempel highlights the concept of a hypothesis. A hypothesis is an attempt at a generalization, an attempt to capture an observed pattern or regularity. Some scholars in the human sciences use the term “laws” to describe very general hypotheses that have turned out to be well corroborated; others prefer not to, thinking that “laws” sound too deterministic and are thus more appropriate to the hard sciences. Hypotheses in general are the building blocks of any science: it is only by proposing, refining and testing hypotheses that we can make any progress towards greater understanding, building on the ideas and contributions of other scholars.


Any kind of research seeks to answer questions and/or solve problems, or discover new questions and problems. The general aims of empirical research are: to provide new data (such as a new corpus, or a detailed case study) on which hypotheses can be tested; to test an existing hypothesis (and perhaps refine it); to propose a new hypothesis (and justify it); to propose a new way of testing a hypothesis (or of generating one); to suggest new connections between hypotheses, i.e. a new theory.



2 Relations between variables


Like any empirical discipline, Translation Studies examines relations between variables. That is, when we study translation we are actually studying relations between two things (or more than two): relations between translations themselves and something else. A major problem is that there are so many variables to be considered, of so many different kinds.


On one hand, we have variables having to do with translations themselves, simply as texts that are assumed to be translations. These refer to aspects of the existence and form of a translation (or set of translations), its linguistic profile, so I will call these  profile variables. These can be any stylistic or syntactic feature, from use of slang to lexical density or the distribution of particular structures.


Interacting with these profile variables we have others having to do with aspects of a translation’s context, including its production process (situational, socio-cultural and cognitive) and its consequences. The context covers both the translation event and the translation act (Toury 1995: 249). I will call these variables context variables. Examples include: source text variables (text type, style...); strategy or procedure variables (shifts from source to target text, explicitation, domesticating devices etc.); target language variables (degree of naturalness as compared with parallel texts...); translator variables (experience, sex, native language...); cognitive variables (translator’s self-image, attitude...); reader/receiver variables (reception, evaluation...); situational variables (tools, skopos, client, time...); historical variables (social and political); cultural variables (norms, cultural identities...); ideological variables (power structures, ethics...); computer program variables.


Relations between profile variables and context variables may be of different kinds. In the first place, a relation might be simply chance. Or there might be a sequence relation: one variable seems to change after another one changes, in a temporal succession. Or there might be a correlation, such that both variables seem to interact in some consistent way. Or there might be a causal relation (see e.g. Chesterman 1998).


Empirical methodology further makes a distinction between dependent and independent variables. The independent variable is the one you might try to control or manipulate in an experiment, the one you might take as your starting point in a comparative analysis. The dependent variable is then the one that (you think) might depend on the independent one: change the independent one, and check if the dependent one also changes.


If your context variable is the independent one and your profile variable the dependent one, you are interested in the effect of context on profile. This is the situation with research questions like:

What are the most frequent strategies for translating culture-bound terms (a source-text variable)? Or: How are retranslations typically different from first translations? (The existence of a previous translation helps to create a different socio-cultural context: how does this affect the profile? On the retranslation hypothesis, see Gambier 1994 and the papers in Palimpsestes 4, 1990.)


If your profile variable is the independent one and your context variable the dependent one, you are interested in the effect of profile on context, and questions like the following can be studied: what is the relation between the number of relative clauses in the translation profile and the degree of fit with target language norms? Or: how does the use of taboo words in a translation affect a translation’s reception (in a given culture at a given time)?



3 Hypotheses


Most empirical research either starts or ends with a hypothesis of some kind. Hypotheses suggest ways of generalising beyond the particular, ways of understanding better, ways of relating a particular research project to other work in the same area. Four basic kinds of hypothesis are commonly distinguished.


An interpretive hypothesis claims that something can be usefully defined as, or seen as, or interpreted as, something else; i.e. that a given concept is useful for describing or understanding something; or that X means Y. Interpretive hypotheses are fundamental to any hermeneutic endeavour; they are also fundamental to empirical research. A classic example: in studies of Shakespeare’s Macbeth, it is often claimed that the three witches represent the unconscious. In other words, the claim is that we can make good sense of the witch scenes if we interpret them as representing or “meaning” Macbeth’s unconscious. This is an interpretive hypothesis.  The key word of an interpretive hypothesis is the word as: we hypothesize that X can be interpreted as Y. Some philosophers even talk about the “hermeneutic as” as the ground of all understanding (e.g. Gadamer: see Koski 1995: 72). All attempts to understand anything unknown begin with an attempt to understand what this thing is like, what we can see it as. Hence the usefulness of metaphors in science — even in empirical science. Interpretive hypotheses are the basis of all conceptual analysis, all attempts to set up definitions and classifications of all kinds. Underlying them all is the claim that we shall understand this thing better if we see it this way, for instance if we interpret it as being divided into these three types... etc. (For more on interpretive hypotheses, see Niiniluoto 1983: 166f.)


Translation Studies abounds with interpretive hypotheses. Here are some: translation can be defined as... / should be seen as...; there are two / five types of equivalence:... (i.e. equivalence can be seen as...); norms of translation fall into three classes; etc. We need interpretive hypotheses, but they are not enough for an empirical science. (See Gile 1998, and the distinction he discusses between theoretical and empirical research on interpreting.) Interpretive hypotheses nevertheless underlie all other hypotheses, insofar as they offer concepts in terms of which other hypotheses can be formulated.


A descriptive hypothesis claims that all instances (of a given type) of phenomenon X have observable feature Y. A descriptive hypothesis makes an empirical claim about the generality of a condition. If I claim that all dogs have tails — that the condition of having tails is valid for all dogs — I am making a descriptive hypothesis, which I can of course test empirically. In Translation Studies, we find descriptive hypotheses in research on translation universals or laws. At a lower level of generality, we also find them in research on particular translation types or text types. Because our field is a human one, descriptive hypotheses are usually formulated here as tendencies rather than universal statements. Thus: translations tend to be more explicit than their source texts; translations tend to reduce repetition; translations tend to be longer than their originals; translations tend to have simpler style / syntax / lexis than parallel texts; translations tend to be more conservative / conventional than parallel texts.


Descriptive hypotheses aim to describe, not to explain; they are ways of answering the question “what?” The remaining two kinds of hypothesis both have to do with the question “why?” They thus concern causal relations. Explanatory hypotheses start with the thing to be explained (the explanandum) and propose an explanation or a cause. Predictive hypotheses start with conditions that are thought to be causal, and predict the resulting phenomena. Predictive hypotheses can thus be used to test explanatory ones.  An explanatory hypothesis claims that the cause of / reason for explanandum E is X; or that E is (probably) caused by / influenced by conditions ABC. A predictive hypothesis claims that factor X will cause event or state Y; or that in conditions ABC, event or state Y will occur / will tend to occur.


In Translation Studies, what you are trying to explain or predict might be some feature of a translation profile (e.g. an error, or a surprising abundance of relative clauses), or some feature of a translation effect (e.g. rejection by the client, assessment by a critic, reaction by the reader).



4 Hypothesis testing


Good hypotheses must be both justified and tested, even though they might start life as an intuition or even in a dream. You might justify a hypothesis by argument; by relating it to other, more established hypotheses; by preliminary evidence; or by a case study.


There are various degrees of “testability”. The strongest requirement for an empirical hypothesis is that it should be falsifiable: it should be possible to prove the hypothesis wrong. If a hypothesis is not, strictly speaking, falsifiable, a weaker requirement is that it should nevertheless be testable. A claim that cannot be tested at all is not worth making, from an empirical point of view: it would be mere speculation. It may nevertheless be the case that if a hypothesis cannot be tested directly, it still has testable consequences.


Hypotheses can be tested on various criteria: for added value in general (increased understanding); for plausibility in comparison with competing hypotheses; for internal logic, elegance, economy; and against empirical data. If a hypothesis turns out to be supported by empirical evidence, this of course might simply be due to chance, so a replication of the test might be needed. Replications of hypothesis-testing procedures are a standard part of empirical science, but so far they have been rare in Translation Studies. For a test to be replicated, the methodology must be described explicitly, in enough detail. Further and different tests might also be warranted, to check the validity of the hypothesis. A well-corroborated hypothesis can also lead to further generalizations, so that understanding grows.


If a hypothesis is not supported, this is usually an interesting result, especially if the hypothesis seemed to be well justified in the first place. Such a result raises new questions. Was the empirical test perhaps inappropriate, was the material badly chosen, not typical, not valid? Were the calculations wrong, not reliable? If you come to suspect the test itself rather than the hypothesis, the next stage is to test again, or to replicate the test on other material. Or maybe the hypothesis itself needs to be refined, or even rejected? A research project might even start off with two, opposing hypotheses, and see which gains more support.



5 Categorization


Research involves two basic cognitive processes: looking for differences and looking for similarities. Looking for differences is a process of analysis. This means breaking a concept or a set of data down into smaller units; it needs concentration, convergent intelligence. Looking for similarities is a process of synthesis, of generalization. It means looking for regularities, patterns; it needs imagination, divergent intelligence. Both these processes come together in categorization. The formation of relevant categories is indeed one of the most crucial and difficult parts of a research project. Categories are yet another form of interpretive hypothesis: you propose a category if you think it is useful, if it allows you to say something interesting, to make a valid generalization, to formulate a precise hypothesis.


Classical (Aristotelian) categories are the black-and-white, watertight-box kind. You either pass an exam or fail it, for instance: here, there are two categories, and they are mutually exclusive and non-overlapping. Many of the categories we use in everyday life, however, are not of this kind, but “natural” or fuzzy ones. For instance, take the category-pair “young” and “old”: it is impossible to draw a precise dividing-line between them. Natural categories often have a prototype structure, with clear, most typical examples in the centre of the category and less typical examples on the periphery. So we have typical birds like robins and blackbirds in the centre of our “bird” category, and less typical ones like penguins and ostriches on the periphery. Fuzzy categories easily overlap with neighbouring ones.


A related set of categories constitutes a classification. Here again, there are various options. A classification might be a simple binary one (colour film vs. black-and-white film, for instance). Or it might be a combination of two binary ones, as in a four-cell diagram. Another kind of classification is a continuum, along a single dimension between two poles, such as free vs. literal translation. Such a continuum might be punctuated by various intermediate stages. Categories on a continuum tend to be fuzzy ones.  A more complex classification might use more than one such continuum and thus be multidimensional.


The formulation of categories in a particular research project is determined partly by the nature of the material being studied and partly by the choice of theoretical model and its basic concepts. Because categories and classifications are interpretive hypotheses — other ways of categorizing and classifying a set of data are always possible — they too need to be justified and tested. Do they allow interesting generalisations? How do they relate to categories and classifications proposed by other scholars? Are they comparable? Are they explicit enough to be used in replicating studies? Do they represent the data adequately?



6 Conclusion


In conclusion I offer a checklist of methodological points worth bearing in mind in empirical research.


       1. Research question / aim: Clearly stated? Why is this a good question / an important or interesting aim?

       2. Other relevant research: How well can you relate what you are doing to what others have done?

       3. Hypothesis: Are you starting or concluding with a specific hypothesis? What kind of hypothesis is it? Why is it interesting / important? Is it well justified?

       4. Material: What is your empirical material? Why did you choose it? How did you collect it? Is it representative?

       5. Relation between variables: What kind of relation are you looking for / do you think you have found? Between what variables, exactly?

       6. Theoretical model: Why did you choose a particular theoretical model or approach / a particular variant of that model? What about other possibilities? Why did you reject those? Have you adapted the model at all? Why?

       7. Central concepts and categories: Adequately defined? Justified against alternative concepts, categories and definitions? What kind of categories? What kind of classification?

       8. Counter-evidence: Considered? Borderline cases dealt with adequately? Counter-arguments? Alternative explanations?

       9. Reliability: Is the analysis reliable? Explicit enough to be replicable? Calculations accurate? Classifications consistent? Statistics appropriate?

       10. Validity: Are the conclusions valid? Hypotheses supported or not? Adequate evidence? Logical argument?

       11. Follow-up: Now what?

       12. Implications: So what?





Chesterman, Andrew (1998). Causes, translations, effects. Target 10, 2, 201-230.

Gambier, Yves (1994). La retraduction, retour et détour. Meta 39, 3, 413-417.

Gile, Daniel (1998). Observational studies and experimental studies in the investigation of conference interpreting. Target 10, 1, 69-93.

Hempel, Carl G. (1952). Fundamentals of concept formation in empirical science. Chicago: University of Chicago Press.

Koski, Jussi T. (1995). Horisonttiensulautumisia. Keskustelua Hans-Georg Gadamerin kanssa hermeneutiikasta, kasvamisesta, tietämisestä ja kasvatustieteestä. Helsinki: Helsinki University Press.

Niiniluoto, Ilkka (1983). Tieteellinen päättely ja selittäminen. Helsinki: Otava.

Nord, Christiane (1997). Translating as a purposful activity. Manchester: St. Jerome Publishing.

Toury, Gideon (1995). Descriptive translation studies and beyond. Amsterdam: Benjamins.

Wright, Georg Henrik von (1971). Explanation and understanding. London: Routledge and Kegan Paul.





Back to [Publications list]                     Back to [Homepage]