Even though adequate analyses of the notion and the thesis of incommensurability are conspicuously absent from the literature, a somewhat more specific sense is nevertheless often associated with the notion—namely, incommensurability as untranslatability due to radical variance of meaning1 (sense and/or reference) of terms occurring in competing scientific languages.
It is widely accepted that Kuhn’s and Feyerabend’s thesis of incommensurability hangs on their contextual theories of meaning, which were advanced as a result of their rejection of the standard empirical account of the meaning of scientific theory. In brief, the traditional empirical theory of meaning has two main components: the reductionist account of meaning (reduction of meanings to sensory experience) and the ‘double-language’ model of scientific theory (the distinction between theoretical language and observation language). Feyerabend finds that such an ‘orthodox empiricist’ theory of meaning is untenable, for it depends on an a priori assumption of a special relationship between the human observer and an observer-independent physical reality. This a priori assumption is at the heart of the theoretical-observational distinction, which holds that sensations or sensory stimulations of the human observer have significance for meaning. Feyerabend contends that a truly empirical philosophical position must see the interaction between the observer and the ‘world’ as one of the processes that needs to be examined and cannot be taken as a priori. In particular, we need to examine it in the light of other hypotheses concerning the relation between observers and the ‘world’. Feyerabend proposes just such a hypothesis: All observation sentences are theory-laden in the sense that experience and empirical meaning, if any, must occur within a comprehensive ‘world-view’ or a theory. Based on this hypothesis, Feyerabend advances his version of the contextual theory of meaning, according to which, it is a theory that gives meaning to observation sentences and not the other way around.2 The following quotation from Feyerabend summarizes his position best:
After all, the meaning of every term we use depends upon the theoretical context in which it occurs. Words do not mean something in isolation; they obtain their meanings by being part of a theoretical system. Hence if we consider two contexts with basic principles that either contradict each other or lead to inconsistent consequences in certain domains, it is to be expected that some terms of the first context will not occur in the second with exactly the same meaning. Moreover, if our methodology demands the use of mutually inconsistent, partly overlapping, and empirically adequate theories, then it thereby also demands the use of conceptual systems that are mutually irreducible (their primitives cannot be connected by bridge laws that are meaningful and factually correct), and it demands that the meanings of all terms be left elastic and that no binding commitment be made to a certain set of concepts. (1965c, p. 180)
The early Kuhn concurs: ‘Successive theories are incommensurable (which is not the same as incompatible) in the sense that the referents of some of the terms which occur in both are a function of the theory within which those terms appear’ (1979, p. 416).
According to the above readings of Kuhn’s and Feyerabend’s contextual theories of meaning, the meanings of some apparently similar terms employed by two competing scientific languages would be different. Moreover, it is hardly possible to identify and is impossible to formulate the meanings of the terms occurring in a competing language within the home language. This is because the meaning of a theoretical term is supposed to be determined by the theory in which it occurs. If a theoretical term t (say ‘phlogiston’) is taken out of its appropriate theoretical background LP (say, the phlogiston theory), it would lose its original semantic value in that theory. So, in this view of meaning, when t is examined from the point of view of a competing theory LC (say, modern theory of chemistry), even if the original semantic value of t in LP might be identifiable in terms of some expressions within LC, the value cannot be expressed or reformulated in LC without loss. In this case, we would say that the full semantic value of t in LP is unrecoverable within LC. In particular, the meanings of the shared terms in two competing languages (such as ‘mass’ in both Newtonian and relativity theory) would be different. If this is the case, we say that there is a lack of a certain desirable semantic relation—that is, meaning continuity—between LP and LC. Consequently, the proponents of two competing scientific languages would inevitably talk past each other when they attempt to resolve their disagreement. It is not even clear whether there is any genuine logical disagreement at all between them. In this case, the two languages are incommensurable.
Presumably, a lack of meaning continuity between two competing scientific languages has one closely related chief consequence for inter-language translation between them. It seems self-evident that mutual translation between two languages is possible only if at least two conditions can be met: (a) the meanings of shared terms (such as ‘mass’ in both Newtonian physics and relativity theory) can be preserved; (b) the meanings of different terms in the competing language (such as ‘phlogiston’ in phlogiston theory) can be identified and reformulated in the home language (for example, the modern theory of chemistry). The lack of meaning continuity between the terms occurring in two languages would inevitably render the two languages untranslatable. As such, it is believed that untranslatability turns out to be the very nature of incommensurability {incommensurability as untranslatability). As a matter of fact, as early as at the end of 1960s and during 1970s, Kuhn3 realized that ‘untranslatable’ (Quine’s term) is a better word than ‘incommensurable’ to describe a communication breakdown between two alien contexts: ‘In applying the term “incommensurability” to theories, I had intended only to insist that there was no common language within which both could be fully expressed’ (Kuhn, 1976, p. 191). Thus, instead of saying that Aristotle’s physics and Newton’s physics are incommensurable, one should say that many substantial Aristotelian sentences could not be translated into Newtonian sentences.
As far as the sources of incommensurability are concerned, we can identify two alternatives within the framework of the translation-failure approach with respect to which semantic value, meaning (sense) or reference, plays the essential role in the explication of incommensurability. The meaning alternative regards incommensurability as the result of a radical change in the meanings (senses) of the non-logical terms occurring in two competing languages in question (incommensurability due to radical meaning variance). The reference alternative attributes incommensurability to radical change in the references of the terms employed in two competing languages (incommensurability due to radical reference variance).
To sum up, according to the received translation-failure interpretation, to say that two scientific theories T1 and T2 are incommensurable is to say that the languages employed by T1 and T2 do not exhibit meaning continuity due to the different semantic values attached to their key terms. Consequently, the two languages are mutually untranslatable.
The above standard reading of Kuhn’s and Feyerabend’s thesis of incommensurability has been the target of an influential line of criticism. It is not surprising that almost all the criticisms have focused on Kuhn’s and Feyerabend’s alleged contextual theories of meaning. Critics argue that Kuhn’s and Feyerabend’s thesis of incommensurability based on their contextual theories of meaning faces many difficulties.
First, critics argue that Kuhn’s and Feyerabend’s treatment of meaning is too general. Kuhn does not specify precisely how he thinks the referents of scientific terms are fixed or established by the theoretical context, and how referents of these terms can change with shift in the theoretical context in which these terms occur. For example, the content of ‘context’ in a contextual theory has to be specified. Otherwise, it will become absurd: If any change in a theory radically alters the meanings of its terms, then any two theories with any slight difference would be incommensurable. However, there is no appropriate way to specify ‘context’ no matter whether we identify it as the whole theory or as a part of it. Kuhn’s and Feyerabend’s positions are too ambiguous so no unequivocal analysis can be given.4 This is the reason why many complain that Kuhn and Feyerabend owe us a clearly formulated theory of meaning that is specific enough to permit extensive scrutiny.
Second, critics claim that there are no good reasons, nor does Kuhn or Feyerabend provide any, for maintaining that all concepts necessarily have their meanings wholly determined by their contexts of distinctive theoretical concepts. A contextual theory, like any other proper theory of meaning, should concern two aspects of ‘the problem of meaning’, namely, the determination of meaning on the one hand, and the change or likeness of meaning on the other. Let us assume, for the sake of argument, that the meanings of the terms of a theory T were fully determined by T. The fact that the meanings of the terms in T were fully determined by T does not compel us to conclude that the meanings of these terms would change radically if T were modified radically. In other words, from the fact that the meanings of the terms of a theory T were established by T it does not follow that there would be a one-to-one relationship between the meanings of the terms in T and theory T. It is very likely that some shared terms in two different theories T1 and T2 have the same meanings. Advocates of the contextual theory do not provide us with any satisfactory criteria of meaning change. Therefore, the thesis of incommensurability does not necessarily follow from the contextual theory of meaning.5
Third, critics question in what sense can two theories (such as Newtonian physics vs. relativity theory) be rivals or compete with each other if they are so semantically disparate that there is no meaning continuity between them. For example, if the meaning of a shared term ‘mass’ is different in relativity theory and Newtonian theory, how can we say of a statement in relativity theory that
(2) The mass of a particle increases with the velocity of the particle
is contradictory with a corresponding statement in Newtonian physics that
(3) The mass of a particle does not change with the change of the speed of the particle.
Last but also most significantly, according to some critics, not only is the contextual theory of meaning refuted by actual scientific practice, but also it theoretically stems from a confused concept of reference, namely, the traditional descriptive theory of reference.6 According to this theory, each term is logically associated with a more or less certain number of descriptions. The referent of a term is the object that fits most of these descriptions. These associated descriptions are the meaning (Frege’s sense) of the term. So, the reference of a term is determined by its meaning (sense). In this way, terms have meanings (senses) essentially and references only contingently. Consequently, whenever meaning (sense) changes reference changes accordingly.
Critics also contend that Kuhn and Feyerabend neglect to consider the possible separation between the change of meaning (sense) and that of reference. To defend this contention, the critics have been busy finding a way to detach the change of reference from meaning (sense) in order to keep either meaning (sense) or reference fully or partially unchanged during theory change. The believers of informal semantics7 take an externalistic view on meaning and separate meaning (sense) from reference. It is believed that full references of scientific terms remain stable while their meanings (senses) might change with the variance of the theory containing these terms. Kripke-Putnam’s causal theory of reference (Kripke, 1972; Putnam, 1973), I. Scheffler’s notion of co-reference (1967), M. Prezelecki’s relativized full-identity of reference (1979), and P. Kitcher’s full-identity of token-reference (1978) are some representatives of informal semantics. In contrast, the followers of formal semantics8 admit that the references of scientific terms change with the theory in which they occur. However, they try to locate some unchanged common core or commensurable part of the two references of the same term before and after the change of the theory in which the term occurs. For example, M. Martin (1971) argues that although the references of the same term in two competing theories are not identical, they somehow overlap. H. Field (1973) agrees, contending that we can identify the shared partial reference between the same terms occurring in two competing theories.
In comparison with the above reference alternative, those following the radical meaning alternative argue that what changes with the change of a theory is the references of terms while the meanings (senses) of the terms remain stable. Others, such as D. Shapere (1984) and N. Nersessian (1984, 1989), who follow the modest meaning alternative, grant meaning variance by accepting the basic principle of the contextual theory of meaning, but argue that there is a continuity of meaning (sense) during meaning variance on the basis of the chain-of-reason.
In sum, critics conclude that, contrary to Kuhn and Feyerabend, there is no such radical change in the meanings of scientific terms corresponding to any substantial change in the theory in which the terms occur. No matter how much the meanings of the terms change with the variance of the theory containing them, we can always locate some sort of meaning continuity between the terms in two theories. This continuity manifests itself either as a common core of semantic values shared by the same terms or as the recoverable semantic values between the different terms in two competing theories. It is the very existence of such a meaning continuity between the terms of two competing theories that makes mutual translation possible. The thesis of incommensurability as untranslatability fails.
* * * * *
My objections to the above influential line of criticism of the thesis of incommensurability fall into two parts. In the first part, I will assume, for the sake of argument, that the basic premise of the received translation-failure interpretation, namely, incommensurability should be identified as untranslatability. Even so, the influential line of criticism fails for two major reasons. First, Kuhn’s and Feyerabend’s expositions of the contextual theory of meaning are not as vague as the critics think. On the contrary, Kuhn and Feyerabend do offer us two quite comprehensive theories of meaning, which deal with all three aspects that any adequate theory of meaning is supposed to address: determination, change, and formulation of meaning. Second, the critics’ attempts to locate meaning continuity is either unsuccessful or fails to render mutual translation possible between two alleged incommensurable languages.
In the second part of my response, the basic premise of the influential line of criticism, namely, the identification of incommensurability with untranslatability, is questioned and rejected. In fact, Feyerabend’s and Kuhn’s explication of the thesis of incommensurability is different from the standard interpretation. As will become clear later in chapters 7 and 10, Feyerabend’s and the later Kuhn’s expositions of incommensurability have moved away from the translation-failure interpretation.
In some general philosophical positions concerning the theory of meaning, the problem of meaning is usually taken to have two thrusts. One is the significance or the having of meaning; the other is the likeness of meaning or synonymy. I would like to add another aspect to any adequate theory of meaning: the formulation of meaning—i.e., how to formulate the meaning of a term employed by one scientific language in a competing language. This is because whether the meaning of a term in one scientific language can be identified by the users of another scientific language is a different issue from whether the meaning of that term can be formulated or expressed fully within the second language. For example, the term ‘dephlogisticated air’ in phlogiston theory, if treated as a token expression, sometime refers to oxygen itself, sometimes refers to an oxygen-enriched atmosphere. However, although the language of modern chemistry can be used to identify the reference of ‘dephlogisticated air’ (as a token) employed by phlogiston theory, it is a totally different issue whether the meaning of the same term ‘dephlogisticated air’ can be expressed without loss within the language of modern chemistry.9 Therefore, a full-fledged theory of meaning needs to deal with three separate aspects of meaning: (a) the determination of meaning; (b) the change of meaning; (c) the formulation of meaning.
Examining Feyerabend’s contextual theory of meaning (sense) from these three aspects, we find that Feyerabend’s view, while it is certainly contextualistic, is not simply that the meanings of scientific terms are determined by the theoretical context in which they occur and vary with change of the context as usually construed.
The first issue is the determination of meaning. According to the standard reading, Feyerabend’s account of meaning may initially suggest that the meanings (senses) of terms are determined by the entire theoretical context in which they occur. A theory as a whole is somehow constitutive of the meanings of the terms occurring in the theory. However, according to Feyerabend, it is not a theory as a whole, but only some specific part of it, that determines the meanings of the terms within it.
Now it seems reasonable to assume that the customary concept of meaning is closely connected, not with definitions which after all work only when a large part of a conceptual system is already available, but with the idea of fundamental rule, or a fundamental law. Changes of fundamental laws are regarded as affecting meanings. ... There exists therefore a rather close connection between meanings and certain parts of theories. (Feyerabend, 1965a, p. 14, fn. 17; italics as original)
Very briefly put, by ‘fundamental rules or laws’—which Feyerabend later also calls ‘universal principles’—Feyerabend means some fundamental factual presuppositions underlying a scientific language. A fundamental rule of a scientific theory determines the meanings of the terms occurring in the theory in a way that suspending the rule will suspend the meanings of the terms. For instance, the notion of impetus depends upon the Aristotelian principle that all motion is the result of the continuous action of some force, which constitutes the fundamental rule of the impetus theory (Feyerabend, 1962, pp. 52-5).
Second, the issue of meaning determination is not the issue of meaning variance. From the mere fact that meaning depends on theoretical context it does not at all follow that meaning therefore bears a one-to-one relationship to theoretical context. It is possible for two terms in different theories to have the same meaning even if the meaning of each of them is fully determined by the theory in which they occur. As a matter of fact, Feyerabend himself explicitly denies a charge of extreme meaning variance based on a reading of his account. According to this reading, Feyerabend commits himself to an extreme view that, with any change of theory, no matter how slight, the meanings of the terms in this theory must change.10 However, for Feyerabend, not every change of theory involves change of meaning, but only some more restricted forms of change do. In the following passages, Feyerabend distinguishes changes not affecting meaning from those doing so:
The change of rules accompanying the transition T —> T’ is a fundamental change, and that the meanings of all descriptive terms of the two theories, primitive as well as defined terms, will be different: T and T’ are incommensurable theories. (1965a, p. 115)
A diagnosis of a stability of meaning involves two elements. First, reference is made to rules according to which objects or events are collected into classes. We may say that such rules determine concepts or kinds of objects. Secondly, it is found that the changes brought about by a new point of view occur within the extension of these classes, therefore, leave the concepts unchanged. Conversely, we shall diagnose a change of meaning either if a new theory entails that all concepts of the preceding theory have extension zero or if it introduces rules which cannot be interpreted as attributing specific properties to objects within already existing classes, but which change the system of classes itself. (1965b, p. 268)
Actually, Feyerabend here specifies two criteria of meaning change with respect to theory change: The meanings of terms in a theory T change if T undergoes substantial change to the extent either that the scheme of categorization of T is affected, or that the fundamental rules are changed.
Third and much more importantly, there is another sense in which Feyerabend’s account of meaning is not merely contextualistic. Not all meaning variances lead to incommensurability in Feyerabend’s sense. Feyerabend repeatedly stresses that incommensurability is not simply a matter of difference of meaning between theories:
Mere difference of concepts does not suffice to make theories incommensurable in my sense. The situation must be rigged in such a way that the conditions of concept formulation in one theory forbid the formulation of the basic concepts of the other. (1978, p. 68, fn. 118; my italics)
Ten years later, Feyerabend, while denying the semantic interpretation given by H. Putnam (1981) of his thesis of incommensurability, emphasizes two differences between Putnam’s interpretation and his own:
First, incommensurability as understood by me is a rare event. It occurs only when the conditions of meaningfulness for the descriptive terms of one language (theory, point of view) do not permit the use of the descriptive terms of another language (theory, point of view); mere difference of meanings does not yet lead to incommensurability in my sense. Secondly, incommensurable languages (theories, points of view) are not completely disconnected—there exists a subtle and interesting relation between their conditions of meaningfulness. (1987, p. 81; my italics)
According to Feyerabend, neither the failure of meaning identification nor even meaning variance constitutes a sufficient condition for incommensurability. Instead, two theories are incommensurable if some central concepts in one theory cannot be formulated in the language of the other because the conditions of meaningfulness (the conditions of concept formulation) for these concepts are incompatible with each other. Furthermore, as we have mentioned above, Feyerabend identifies two basic conditions of meaningfulness: one relates to the universal principles and the other to the schemes of categorization underlying the language in which the concepts are to be formulated. According to this reading, Feyerabend, in saying that the conditions of concept formulation of one theory forbid the formulation of the concepts of another theory, is actually offering us two criteria of incommensurability:
(Fl) Two theories T1 and T2 are incommensurable if the respective schemes of categorization are incompatible in the sense that the concepts used in one scheme cannot be formulated in the other.
(F2) Two theories are incommensurable if the respective universal principles are incompatible in the sense that the principles of one theory forbid the expression of the concepts that are expressible in terms of the principles of the other theory.
In the later 1980s, Kuhn introduced a new bearer of incommensurability, i.e., the lexicons/lexical structures of scientific languages. Notice that Kuhn uses ‘lexicon’, ‘lexical structure’, ‘taxonomy’, or ‘taxonomic structure’ interchangeably in his writings. By ‘lexicon’, Kuhn does not mean a lexicon in a linguistic sense, but rather refers to the categorical structure of a scientific language. By focusing on the lexical structure of language, the later Kuhn modifies his early contextual theory of reference by associating the determination and the change of reference of kind-terms with the nature of taxonomization. In my reading, Kuhn, by doing this, presents a modified, finer-grained, and coherent theory of reference, which I will call the lexical theory of reference of kind-terms (hereafter LTRK).11
The main concern of Kuhn’s LTRK is about the determination and change of the reference of a kind-term. The question is: How do kind-terms actually attach to nature? What are the references of kind-terms? Here Kuhn is concerned with actual referents themselves, namely, with the set of objects or situations to which kind-terms actually attach. According to Kuhn’s logic, the basic process of how the references of kind-terms are established in a language community can be summarized as follows.
Kind-terms, as category terms, are attached by a language community to the existing categories of objects or situations in a world to which they apply. It is natural to use the extensions of these corresponding categories of objects or situations as the actual referents of the terms.
Furthermore, kinds are taxonomic categories. Kinds, especially scientific kinds, can be arranged taxonomically. Each taxonomy has its unique structure that can be specified by a process of taxonomization. Usually, taxonomization is a categorizing process that can be divided conceptually into two stages: (i) individuation: both individuating entities into distinguishable items and categorizing the domain into taxonomic categories; (ii) distribution: distributing items into pre-existing categories. Corresponding to these two stages, the structure of taxonomy consists of two parts: taxonomic categories or kind-terms and relationships (similarity/dissimilarity) among these categories. More formally, a taxonomic or lexical structure is the vocabulary structure shared by all members of a language community that provides the community with both shared taxonomic categories/kind-terms and shared relationships between them.12
Existing in a taxonomy, interrelated taxonomic categories or kind-terms in a local area cannot be isolated from one another because ‘those categories are interdefmed’, must be learned or relearned together, and laid down on nature as a whole (Kuhn, 1983b, p. 677). More precisely, kind-terms with normic expectations (the expectations with exceptions), which we call ‘high-level theoretical kind-terms’,
must be learned as members of one or another contrast set. To learn the term ‘liquid’, for example, as it is used in contemporary non-technical English, one must also master the terms ‘solid’ and ‘gas’. The ability to pick out referents for any of these terms depends critically upon the characteristics that differentiate its referents from those of the other terms in the set. (Kuhn, 1993a, p. 317)
On the other hand, the kind-terms with nomic expectations (the expectations without exceptions), which we can call ‘low-level empirical kind-terms’, must be learned with other related terms and from situations exemplifying laws of nature. For example, ‘force’ has to be learned both with ‘mass’ and ‘weight’ and with recourse to Hooker’s law and Newton’s three laws of motion.13 Therefore, taxonomy is necessarily locally holistic. Change in one category would affect other categories of a taxonomy in a local area. This means that references of kind-terms, as the extensions of the corresponding categories in a taxonomy, are determined by the taxonomic structure in some local part of the language. In this sense, ‘the criteria relevant to categorization are ipso facto the criteria that attach the names of these categories to the world’ (Kuhn, 1987, p. 20). Categorization is a coinage with two faces, one looking outward to the world to divide it in different ways, the other inward to the world’s reflection in the referential structure of the language to determine the references of kind-terms.
In conclusion, according to Kuhn’s LTRK, the extensions of kind-terms are determined by taxonomic structures of languages, and will change with change of such structures. These reference changes are confined to some selected high-level theoretical kind-terms and leave other terms referentially stable. Therefore, a necessary condition of sameness of reference of shared kind-terms between two languages is that they share the same taxonomic structure.14
Unlike two members of the same language community, speakers of mutually translatable languages need not share terms: ‘Rad’ is not ‘wheel’. But the referring expressions of one language must be matchable to coreferential expressions in the other, and the lexical structures employed by speakers of the languages must be the same, not only within each language but also from one language to the other. Taxonomy must, in short, be preserved to provide both shared categories and shared relationships between them. (Kuhn, 1983b, p. 683)
If so, when two languages with incompatible taxonomic structures confront each other, then the kind-terms from one language cannot be translated into the corresponding terms in the other without violating the non-overlap principle (No two kind-terms in the same level of a taxonomy may overlap in their extensions. For example, there are no dogs that are also cats in the mammal taxonomy). That means that the mutual translation between the two languages is impossible. This combination of a logical theory of taxonomy and a linguistic theory of translation constitutes the later Kuhn’s taxonomic interpretation of incommensurability. ‘Incommensurability thus becomes a sort of untranslatability, localized to one or another area in which two lexical taxonomies differ’ (Kuhn, 1991, p. 5).
It is clear that, as we have shown above, there does not exist, nor does Kuhn or Feyerabend claim, an alleged one-to-one correspondence relationship between a theory and the meanings of its terms. It is not that all changes of a theory would affect the meanings of its terms. The fact that some shared terms in two different theories might have the same meaning is perfectly consistent with Kuhn’s and Feyerabend’s versions of the contextual theory of meaning. Kuhn’s and Feyerabend’s contextual theories of meaning, if correct, really lead to the conclusion that there exists some sort of meaning discontinuity between two competing languages. This sort of discontinuity would lead to untranslatability.
The critics contend (a) that it is possible to locate some kind of meaning continuity between two competing languages, and (b) that the kind of meaning continuity identified would render mutual translation between them possible. However, the critics’ efforts to identify meaning continuity are either unsuccessful, or, if successful, the kinds of meaning continuity identified cannot be used to disprove the thesis of untranslatability.
Do the critics succeed in identifying the desired meaning continuity between two competing languages? For the answer, let us examine the meaning (sense) alternative first. D. Shapere (1984) contends that although meaning (sense) will change with the variance of beliefs or theories, we can still locate meaning (sense) continuity between the languages of two theories by identifying a chain-of-reasoning connecting the two theories. According to Shapere, the rationality of science can be preserved as long as we can locate such a chain-of-reasoning during the development of a scientific theory. However, the attempts to retain the rationality of science in terms of explaining the rationality of conceptual change or meaning change are doomed to failure for the following reason. It is true that a historian of science can make the shift of scientific belief systems intelligible by tracing back the chain of change for each concept. However, historical explanation itself is not a philosophical justification. There is nothing the philosopher can add to what the historian has already done to show that this intelligible and plausible course of development is a rational one. The result would be that everything that has happened in the history of science would be rational, as long as we can reconstruct its actual process of change. This inference is valid only if we accept a presupposition that science itself is a rational enterprise. It begs the question; since whether or not science is rational is the issue at stake that is what Shapere wants to prove through the introduction of the concept of a chain-of-reasoning in the first place.
In addition, it is reference variance rather than meaning (sense) variance that has the relativistic and anti-realistic implications allegedly promoted by the thesis of incommensurability. Historical relativism and scientific anti-realism cannot follow from meaning variance alone. What motivates many philosophers to oppose the thesis of incommensurability is the defense of scientific realism. Scientific realism is a doctrine about reference (the notion of reality and truth), not about meaning (sense). Even we assume that meaning (sense) can be separated from reference, meaning continuity still cannot avoid the consequences brought about by radical change of reference (Leplin, 1979). Therefore, even if the opponents following the meaning alternative succeed in proving the existence of stable meaning or continuity of meaning, they still cannot defeat the thesis of incommensurability and avoid its undesirable implications.
The critics who take the route of informal semantics within the framework of the reference alternative argue that the meanings (sense) of terms may change with the variance of a theory in which the terms occur, but that their references remain stable. This alternative seems to be more promising than the meaning alternative. Even if the meaning (concept) of a term in a theory cannot survive the change of a whole belief system, since they are intertwined inextricatively, it is still possible that the reference of the term survives the change. One possible way to explain the stability of reference is to follow G. Frege’s initiative. I. Scheffler (1967) argues that by separating sense from reference following Frege’s approach, we can keep reference unchanged while meaning varies. This is a strange proposal considering that Frege explicitly adopts and promotes the description theory of meaning. From the fact that some terms different in meaning (sense) have the same reference, it does not follow that a term can always keep its reference unchanged while changing its meaning. Frege’s theory of meaning provides no basis for expecting such stability of reference at all. On the contrary, as long as we accept Frege’s theory, the task of identifying stable reference is hopeless. According to Frege, the only way we have of identifying the reference of a term is by the criteria formulated in its sense; so even if it were true that reference remained the same when sense changed, we would have no (a priori) means of knowing it.
Another possible way to explain the stability of reference is more revolutionary, namely, to apply S. Kripke’s (1972) causal theory of names to general terms, as H. Putnam (1973) does. According to this theory, there is a causal chain of reference-preserving links between the naming ceremony of a term and its use on some occasion. The reference of a term, either an individual or a general term, cannot change after it is fixed in the initial naming ceremony by ostension. However, it is well known that the causal theory of reference faces many difficulties. As G. Evans (1973) points out, the existence of a causal chain is neither sufficient nor necessary for an individual to identify the reference of a name. Especially, the references of many scientific terms (such as ‘positron’, ‘photon’, ‘heat’, or ‘water’) were not fixed by any so-called naming ceremony by ostension. Actually, their initial references were fixed by their roles in a certain theoretical context, rather than on the basis of some experience attributed to the causal agency of the alleged referent (Leplin, 1979). So the causal theory at most tells us what a theory of reference could be, not what a theory of reference actually is. More importantly, the references of many scientific terms do change during the development of scientific theories, as Kuhn argues convincingly in many places, such as the difference in the references of the kind-term ‘compound’ in pre-Dalton and post-Dalton chemistry. Besides, it is perfectly conceivable that the reference of a name will change in some circumstances (imagine that two babies are switched without notice right after their mothers give them names). Therefore, any theory of reference that cannot accommodate the possibility of reference variance is essentially inadequate.
The failure of informal semantics seems to encourage a move toward formal semantics. According to formal semantics, although the reference of a term does change with the variance of the theory in which it is embedded, there exist some sort of commensurable parts between two different references of the same term embedded in two alleged incommensurable theories. This reminds us of Shapere’s chain-of-reasoning within the meaning alternative. However, formal semantics is not in better shape than informal semantics. To see the chief drawback of formal semantics, we need to consider the second issue concerning referential continuity: Does the existence of referential continuity identified by formal semantics between two alleged incommensurable languages disprove the thesis of incommensurability as untranslatability?
As I have mentioned, the motivation behind the attempts to identify commensurable reference is to restore mutual translation between two alleged incommensurable languages. However, formal semantics, if successful, fails to give us translatability. The reason is simple. The thesis of incommensurability as untranslatability is a negative thesis. To argue for untranslatability between two competing languages, one only needs to show that some necessary conditions of intertranslation cannot be met between two alleged incommensurable languages. Although controversy exists about what should count as necessary conditions of translation, it is safe to say that a successful translation requires one to formulate any expression es in the source language (the language to be translated) into an appropriate expression et, a semantic equivalent of es, in the target language (the language into which the translation is made). Sameness of reference and meaning (sense) are two obvious desirable semantic equivalents that should be preserved during translation. As long as Kuhn shows that it is impossible to formulate some crucial expressions in the source language (say, the term ‘mass’ in Newtonian theory) into an appropriate expression in the target language (say, the term ‘mass’ in relativity theory) while preserving the same reference and meaning of the original term, he succeeds in arguing for the thesis of untranslatability. But in order to show the failure of the thesis, the critics have to demonstrate either that mutual translation between two alleged incommensurable languages is possible or, at least, that fully identical reference or meaning of expressions es in the source language (say, ‘dephlogisticated air’ in phlogiston theory) can be preserved by formulating es into the corresponding expression et in the target language (say, ‘oxygen’ in modern chemistry). Unfortunately, the advocates of formal semantics fail to do either. On the one hand, formal semantics only shows us the possibility of preserving partial or overlapping references, instead of identical references, between the corresponding terms in two alleged incommensurable languages. On the other hand, the existence of overlapping references or partial identical references (even the existence of full identical references) between the shared terms in two alleged incommensurable languages is not sufficient for translation. Therefore, even if the critics succeeded in showing that there exists some kind of overlap between the references of the shared terms between two alleged incommensurable languages, they would still not disprove the thesis of untranslatability.
A similar argument can be applied to informal semantics as well. According to informal semantics, it is possible for references of scientific terms to remain stable while their senses change with the variance of the theory containing these terms. Therefore, it is possible to identify the full reference of an expression in the source language in terms of an appropriate expression in the target language. For instance, P. Kitcher (1978) argues that the language of modern chemistry can be used to identify the referents of the terms or expressions of phlogiston theory, at least to the extent that those terms or expressions actually do refer in each specific context (context-sensitive reference or token-reference). For example, ‘dephlogisticated air’ refers in some contexts to oxygen itself, or refers in some contexts to an oxygen-enriched atmosphere. However, as Kuhn (1983b) points out, the crucial issue concerning translation is not whether the referent of a term in the source language can be identified by the appropriate expression in the target language, which is often possible; but whether the term can be formulated by the corresponding expression(s) in the target language to the extent that those expressions can capture the beliefs, intentions, and implications conveyed by the original terms (such as the whole belief system associated with the terms ‘phlogiston’). Reference-identification alone is not a sufficient condition of translation. Hence, reference-identification is not translation.
To sum up, the critics are either unsuccessful in providing us with the meaning continuity that they promise us or fall short of what they intend to achieve; namely, mutual translation between alleged incommensurable languages.
The previous two sections might give readers the impression that I am trying to defend the standard interpretation of incommensurability as untranslatability. This is not the case at all. It is not my intention to make a judgment about which side, Kuhn’s and Feyerabend’s thesis of incommensurability or the above influential criticism of it, holds more water. Instead, I believe that both sides are on a wrong track: It is unproductive and futile to identify incommensurability as untranslatability. The notion of incommensurability (commensurability) does not depend on the notion of untranslatability (translatability).
There is a common assumption underlying both the arguments for and against the thesis of incommensurability as untranslatability: Incommensurability has something to do with a specific kind of semantic relation, i.e., the meaning relation, between the terms of two radically disparate languages:
(MR) It is the meaning relationship between the terms (rather than sentences) of two competing languages that constitutes the determinant semantic relationship in the case of incommensurability.
(MR) is supported by the following fallacious reasoning:
(Ml) Mutual translation is necessary (and sufficient) for effective mutual understanding and successful communication between two scientific language communities.
(M2) Meaning continuity between the terms of two competing languages is necessary for mutual translation between them.
Therefore,
(M3) Meaning continuity is necessary for effective mutual understanding and successful communication between any two distinct language communities.
Attributing the premise M2 to both the advocates and opponents of incommensurability as untranslatability is obvious; for no matter what notion of translation is adopted, mutual translation between two languages is possible only if at least two basic semantic values of expressions (sense and reference) in the translated language can be preserved in the translating language. To justify the premise Ml, we need to realize that although there is no general agreement concerning the nature of incommensurability, there is one basic intuitive sense of incommensurability which is likely to be acceptable to all; namely, incommensurability as the communication breakdown between two competing languages. To say that two languages are incommensurable is another way of saying that successful cross-language communication breaks down between them.
Based on this shared line of reasoning, the views of the advocates and opponents of incommensurability as untranslatability diverge due to conflicting views about the possibility of the existence of meaning continuity between the terms occurring in two allegedly incommensurable languages. The advocates of incommensurability argue, in terms of the contextual theory of meaning, that there is a lack of the meaning continuity between the terms of two competing languages due to radical meaning variance. The two languages are hence not mutually translatable. Consequently, a communication breakdown between the two language communities is inevitable. The two languages are therefore incommensurable. On the contrary, the critics argue that some degree of meaning continuity must exist between two allegedly incommensurable languages. Furthermore, if the meaning continuity is not only a necessary, but also a sufficient condition for translation, then the thesis of untranslatability simply fails.
However, although the critics argue against the untranslatability thesis, they agree with the advocates that the issue of incommensurability is conceptually linked with the issue of translation: Incommensurability is equivalent to untranslatability and commensurability to translatability.15 In particular, the critics accept also that it is the link between untranslatability and incommensurability that makes the meaning relation the determinate semantic relation in the case of incommensurability. In this sense, the critics agree with the advocates on the translational-failure approach to incommensurability.
My criticism of the translational-failure approach to incommensurability focuses on two questions. First: Can the notion of untranslatability carry ‘the weight’ of clarifying the nature of incommensurability? The answer to this question depends on the answers to the following two related questions: Is there a tenable and integrated notion of translation that we can use to clarify the notion of incommensurability? Is mutual translation necessary (and sufficient) for effective cross-language understanding and communication? Second: Is the meaning relation between the terms of two competing languages the determinant semantic relation between two allegedly incommensurable languages? The answer to this question depends in turn upon the answer to the following question: Is meaning continuity necessary for cross-language translation?
Presumably, whether or not translation could be used to clarify the notion of incommensurability depends on whether we can have an adequate notion of translation to work with. It is noteworthy that what counts as an acceptable translation is vague and full of confusion. Here are just a few examples, (a) A distinction could be made between literal (strict, or exact) translation into a home language as it is already constituted—i.e., a translation without change at all to the home language on the one hand and liberal (loose) translation that achieves a degree of success by modifying the existing usages, and hence the existing meanings of words of the home language, on the other, (b) The previous distinction roughly corresponds to the distinction between translation into the home language in the normal sense and interpretation, or a far looser sort of reinterpretative reconstruction of the alien language into the home language.16 (c) Intensional translation focusing on meaning (sense) could be distinguished from extensional translation focusing on references. W. V. Quine (1981, p. 42) fully realizes that translatability is a flimsy notion, unable to fulfill its overloaded mission of distinguishing alternative languages and thereby ‘unfit to bear the weight of the theories of cultural incommensurability that Davidson effectively and justly criticizes’. In fact, Quine (1970) tries to construct a very strict notion of translation to measure the conceptual remoteness of two alternative languages: i.e., a translation in a purely extensional sense, or word-by-word isomorphic mapping between the extensions of all respective general terms and the designata of all corresponding singular terms. Nevertheless, at the same time, he clearly allows awkward and baffling translation between languages embodying alternative conceptual schemes (Quine, 1981, p. 41).
To illustrate Quine’s point, let us take a close look at the notion of translation used by Kuhn in his explication of incommensurability. Kuhn has employed two different notions of translation during the development of his thoughts on the thesis of incommensurability as untranslatability. The earlier Kuhn employs a concept of translation that is used by professional translators in practice. According to this concept, the final goal of a translator, like Quine’s radical translator or interpreter, is to achieve a better understanding of an alien text. For this purpose, the translator ‘must find the best available compromises between incompatible objectives’. This is because ‘translation, in short, always involves compromises which alter communication. The translator must decide what alterations are acceptable’ (Kuhn, 1970b, p. 268).
Translation in this practical sense, so-called liberal translation, has the following features: (a) The target language is permitted to be changed by introducing new concepts and more or less by subtly changing the old concepts, (b) It does not require a systematic replacement with or mapping of words or word groups in the source language to the corresponding words or word groups in the target language, (c) The translation is a matter of degree; no exact translation is available, (d) Most importantly, the translation is closely connected with the process of language learning, and thereby involves a strong interpretative component. Actually, liberal translation is the enterprise practiced by historians and anthropologists when they try to understand an old text or break into an alien culture. In such cases, in order to make the source language intelligible, the translator (strictly speaking, the interpreter) has to learn the source language in the process of translation, and afterward to look for the closest counterparts of expressions of the source language in the target language (Kuhn, 1983b, pp. 672-3).
An immediate problem with the explication of incommensurability in terms of liberal translation is that if we accept this concept of translation, then it is always possible to translate any different text, no matter how alien it is to the translator’s home language; for it is always possible theoretically to make an alien context intelligible by language-learning. Thus, there would be no incommensurable texts at all eventually. The notion of untranslatability leads to the dissolution of the issue of incommensurability. This line of reasoning makes it too easy for the opponents of incommensurability to claim a victory.
A further problem with the notion of liberal translation is related to Kuhn’s treatment of the relation between translation and the existence of a common language. The earlier Kuhn explicitly regards the existence of a common language as a prerequisite for commensurability or translatability.
In applying the term ‘incommensurability’ to theories, I had intended to insist that there was no common language within which both could be fully expressed and which could therefore be used in a point-by-point comparison between them. (Kuhn, 1976, p. 191)
As Kuhn and Feyerabend argue repeatedly, there is not any common language available between two incommensurable languages. In this way, Kuhn makes his argument for the nonexistence of any neutral language the foundation of his thesis of untranslatability. However, untranslatability of two languages L1 and L2 into a common language L is not equivalent to mutual untranslatability between L1 and L2. On the one hand, the requirement for a neutral language as a necessary condition for mutual translation or commensurability is too strict. It cannot be met. On the other hand, the requirement for a common language as a sufficient condition for mutual translation is too loose. We can represent various mathematical theories T1 and T2 in set theory, but we may not be able to represent T1 in T2. The problem is that a translation into set theory does not necessarily give us a translation from set theory, which is what we are going to need to get from T1 to T2 via set theory.17 Especially, this notion of translation based on the existence of common languages is directly contradictory with the concept of loose translation Kuhn employs at this stage; for the former is so strict as to render the translatables untranslatable while the latter is so loose as to render the untranslatables translatable.
After 1983, by virtue of introducing the new tool of lexicon and its structure, Kuhn continues to rely heavily on the notion of untranslatability in his explication of incommensurability, and further links untranslatability closely with change of the taxonomic structures of scientific languages. At this stage, the concept of incommensurability is directly connected with the notion of untranslatability without resource to the notion of a neutral language.18 ‘Incommensurability thus equals untranslatability’ (Kuhn, 1988, p. 11). ‘If two theories are incommensurable, they must be stated in mutually untranslatable languages’ (Kuhn, 1983b, pp. 669-70).
More significantly, at this stage, Kuhn consciously operates with a different notion of translation, literal translation: Translation must be taken in a strict sense as the formulation of expressions within the translating language (the target language) that are semantic equivalents of expressions of the translated language (the source language). Semantic equivalence is not word-to-word synonymy, but it does require that the translator systematically substitute the appropriate expressions in the target language for the corresponding expressions in the source language to produce an equivalent text in the target language. In comparison with the notion of liberal translation, literal translation has the following features: (a) During the process of translation, the target and the source languages should remain unchanged. No addition of new kind-terms and no change of semantic values (sense, reference, and truth-value) of the expressions are permissible, (b) The translation requires a systematic mapping of concepts (kind-terms especially) in the source language to the corresponding concepts in the target language such that each concept is mapped to a concept with exactly the same semantic values, (c) The translation is an all or nothing issue: The translation between two languages is either possible or impossible, (d) The translation is a totally different linguistic activity from language learning or interpretation. The purpose of interpretation is to make an alien text intelligible. We can achieve this by language learning. For this purpose, it suffices to identify the related semantic values of the expressions in the source language. However, the purpose of literal translation is not only to identify the semantic values of the expressions in the source language, but also to formulate semantic equivalents of these expressions within the target language. Translatability in this sense is a function of what can be said or expressed in the target language. It depends upon the potential ability of the target language to produce semantic equivalents of the expressions in the source language without changing its taxonomic structure and without extending its semantic resources. However, the target language’s ability of translating the source language is limited and constrained by the capacity of its taxonomic structure. A translation fails if formulating the semantic equivalents of the expressions of the source language in the target language requires either change of the target language’s taxonomic structure or an extension of its semantic resources by semantic enrichment.19
What constitutes semantic equivalency as sought in a literal translation? Sameness of sense or intension, sameness of references or the extensions of shared kind-terms (including the sameness of the way in which the reference of a term is determined as well), and sameness of truth-values of shared sentences are obvious desiderata. As far as the sufficient conditions of literal translation are concerned, some pragmatic aspects of language, such as conversational implicatures, the speaker’s intentions, the speaker’s meaning, the speaker’s reference, and context specifiers have to be included as well. Unfortunately, it is almost impossible to list all the sufficient conditions for literal translation (in fact, it is doubtful whether there exists such a complete set of sufficient conditions).
According to Kuhn’s notion of literal translation, translation is an all-or-nothing issue. Such a notion of the perfect translation presupposes that there exists, if any, one and only one right translation for an expression in question. However, this presupposition is questionable. Quine has argued convincingly that there can be many mutually incompatible systems of translation consistent with all possible behavioral data. We have no way to determine which of many available translations is the only right translation. Hence, translation is indeterminate. Of course, we can follow Davidson’s suggestion to put enough constraints—both the constraints from the bottom, such as formal constraints and empirical constraints (tested as true), and the constraints from the top, namely, the methodological perception or the principle of charity—on the language to be translated to reduce multiple acceptable translations for the language to a small amount of translations or even to the one best translation. However, a translation so constructed is no longer a literal translation. As Davidson contends, such a translation does not (should not) preserve the detailed propositional attitudes of the speakers (such as the speaker’s beliefs or intentions) and does not, and should not, make use of unexplained linguistic concepts (such as the speaker’s meaning, synonymy). Besides, due to inextricability between meaning and belief, we have to put the speaker in general agreement on beliefs with the interpreter and try to maximize such an agreement. This often inevitably distorts the speaker’s belief system. Even so, there will still be no perfect mapping from sentences held true by the speaker in the source language onto the sentences held true by the translator in the target language.
Recall an example given by Davidson: It is sometimes even impossible to give a literal translation in English of an English speaker’s expression, such as, ‘There’s a hippopotamus in the refrigerator’.
Hesitation over whether to translate a saying of another by one or another of various nonsynonymous sentences of mine does not necessarily reflect a lack of information: it is just that beyond a point there is no deciding, even in principle, between the view that the other has used words as we do but has more or less weird beliefs, and the view that we have translated him wrong. (Davidson, 1984, pp. 100-101)
The inextricable intertwining of the speaker’s belief and the speaker’s meaning establishes that there is no such literal translation available in any case. Therefore, Kuhn’s notion of literal translation is too strict and could never be exemplified. If we accept such a notion of translation, we cannot even translate the speech of others who speak our own language, not to mention translation between two distinct languages.
The conclusion from the above examination of Kuhn’s two notions of translation is that neither the notion of literal translation nor that of liberal translation can help us clarify the notion of incommensurability. In fact, the thesis of incommensurability as untranslatability becomes trapped in a fatal dilemma: If ‘translation’ is construed narrowly and strictly to mean ‘literal translation’, then it indeed follows that two distinct languages (such as Newtonian vs. Einsteinian languages) are untranslatable. But incommensurability in this sense would degenerate into a trivial platitude since there is no literal translation available between any two languages anyway. In this sense, some degree of partial untranslatability marks the relationship of every language to every other. Consequently, it would make the commensurables incommensurable. On the other hand, if ‘translation’ is construed broadly and loosely to mean ‘liberal translation’, then untranslatability will not qualify as a criterion of incommensurability since any two languages would be translatable. Consequently, it would make the incommensurables commensurable. Either way, the thesis of incommensurability would turn out to be an illusion.
Some might reply, ‘Yes, maybe Kuhn’s two notions of translation are trapped in the above dilemma. But this does not exclude the possibility of a more tenable notion of translation that could be used as a criterion of commensurability’. It is true that there might be some more appropriate notion of translation than Kuhn’s two radical notions of translation. But I doubt that we could have a ‘desirable’ notion of translation as the advocates expect, which could be used to clarify the notion of incommensurability. To show this, I am not going to consider every possible notion of translation presented or to be presented by the advocates, such as D. Pearce’s (1987) notion of context-sensitive translation as reduction.20 For my limited purpose, it suffices to point out that the issue of incommensurability, as I will argue later, is essentially associated with the possibility of truth-or-falsehood. It is the issue about the distribution of truth-value status among the sentences of the competing languages, not about the distribution of different truth-values among the sentences of the competing languages. Two languages are incommensurable when both sides disagree about the truth-value status of some substantial number of core sentences in the other language. Such a truth-value gap indicates a linguistic communication breakdown between the two language communities. However, the traditional notion of translation (whether Kuhn’s notion of literal translation or any other alleged more ‘appropriate’ notion of translation), Quine’s indeterminacy of translation, and Davidson’s notion of radical interpretation all start from an idea that an adequate translation should preserve the truth-values of the sentences translated and gain the truth-value matching of sentences between the target and the source languages.
Traditional extensional semantics uses truth-value preservation as a condition for an adequate translation. Kuhn, who adopts intensional semantics on translation, certainly does not regard truth-value preservation as a sufficient condition for literal translation. But the truth-values of the translated sentences are, for Kuhn, an important semantic value that should be preserved in literal translation. Similarly, there is an assumption underlying Davidson’s notion of radical interpretation: Although the speaker and the interpreter in discourse may assign opposite truth-values to some core sentences of a language, they all agree that all the sentences in the competing language are either true or false. The task of the radical interpreter is to design a translation (more precisely, a theory of truth for the language in question), based on Davidson’s principle of charity, that can preserve as much truth as possible. It is my conviction that this kind of truth-preserving translation is irrelevant to the incommensurable texts because what really matters is not redistribution of truth-values between the sentences of two compatible texts, but whether or not both sides still accept the other’s sentences as candidates for truth-or-falsehood. Therefore, using the notion of a truth-preserving translation to clarify the notion of incommensurability is on the wrong track. For this reason, I conclude that there is no tenable and integrated notion of translation that can be used to clarify the notion of incommensurability.
There is still another way to connect the notion of translation with the notion of incommensurability as communication breakdown. Many contend that translation is either sufficient or necessary for effective cross-language understanding and communication. An interpreter can understand an alien language through or only through the relation of the translated language to the interpreter’s own home language. If so, the failure of mutual translation between two languages would indicate the breakdown of cross-language communication between the speakers of two languages, and hence the two languages would be incommensurable. This alleged connection between cross-language communication and translation is a hidden motivation for many philosophers, advocates and opponents of the thesis of incommensurability alike, who identify commensurability with translatability.
Davidson has already observed that a translation manual is to be contrasted with a theory of meaning or interpretation. A theory of interpretation for an unknown language directly describes the way in which the language functions by providing an account for the essential role of each sentence in that language. Since the meaning of each sentence in a language is determined by the essential role of the sentence in the language as a whole, a theory of interpretation for a language does help us understand the language. For Davidson, a theory of interpretation is a theory of understanding. A translation manual tells us only that certain expressions of the translated language mean the same as certain expressions of the translating language, without telling us what, specifically, the expressions of either language mean. It is theoretically possible, Davidson contends, to know of each sentence of a given language that it means the same as some corresponding sentence of another language, without knowing at all what meaning any of these sentences has. Therefore, a translation manual itself does not constitute a theory of meaning or understanding. A translation manual leads to an understanding of the translated language only via the translating language, an understanding that it does not itself supply. In other words, a translation itself is not sufficient for understanding.21
Furthermore, according to the truth-value conditional theory of understanding to be presented in chapter 9, the notion of truth-value status, instead of the notion of truth, plays an essential role in understanding; an interpreter can understand a sentence in a given language only if the sentence has a truth-value from the point of view of the interpreter. Shared notion of truth-value status between the interpreter and the speaker is necessary for the interpreter to understand fully the sentences in the speaker’s language. In this way, understanding an alien text is a matter of recognizing new possibilities for truth-or-falsehood, and of how to appreciate and follow other ways of thinking embedded in other languages that bear on these new possibilities. What we need to capture in communication is not what each word in a sentence or the whole sentence means in an alien language, which can be done by constructing an adequate translation manual, but the factual commitments underlying each language that create new possibilities for truth-or-falsehood. For example, even if a sentence in traditional Chinese medical theory, ‘The combination of yin and rain makes one drowsy’, could be translated into some appropriate expressions in Western medical theory, Western physicians are still at a loss if they do not know the mode of reasoning underlying the sentence. What is true-or-false in the way of reasoning used by Chinese medical theory may not make much sense in another until one has learned how to reason in that way.22 Hence, a truth-preserving or meaning-matching translation, if any, is not sufficient for effective understanding of an alien text.
On the other hand, translation is certainly not necessary for understanding. There is a good reason to think that the whole focus on actual translation in understanding is misguided. The key category with which we should be concerned in understanding is surely not translation but interpretation or language learning. Interpretation is a very different linguistic activity from translation. By interpretation, following many others, I mean the enterprise practiced by historians and anthropologists when they try to understand an old text or break into an alien culture. The interpreter, at the very beginning, only masters his or her own home language. The source language is totally unknown to him or her. The purpose of interpretation is to make an unknown alien text intelligible. The most effective means to make an old or alien text intelligible is to learn the language, instead of to translate it. For this purpose, it suffices to identify the related semantic values of the expressions in the source language. For instance, to identify the referent of an out-of-date kind-term (say, ‘phlogiston’ in phlogiston theory) and the role of the term in the theory is sufficient for understanding the term. However, to understand the theory, the interpreter has to learn the source language by identifying the related semantic values of all the expressions in the source language. By comparison, a translator is supposed to master both the source and the target languages at the very beginning. The purpose of translation is to formulate semantic equivalents of expressions of the source language in the target language. Translatability in this sense is a function of what can be said or expressed in the target language. It depends upon the potential ability of the target language to produce semantic equivalents of the expressions in the source language without change of the taxonomic structure of the target language and without extending its semantic resources. A target language’s ability to translate a source language into a specific language is limited and restrained by the capacity of its underlying taxonomic structure, descriptive universal principles, and mode of reasoning. There can be no genuine translation where the underlying descriptive principles, taxonomic structures, or the modes of reasoning of two languages involved are substantially different.
After understanding an alien context by learning it, interpreters may be able to make an alien text intelligible to fellow speakers through paraphrasing, ‘explaining’ or the like, the alien text into the nearest ‘corresponding’ (not exact mapping) expressions in their own home language. For example, an English speaker, after learning Chinese, may attempt to interpret the Chinese sentence, ‘Jenni hen fung maan’, as, ‘Jenny is voluptuous, has a full figure, or is full and round, well-developed, full-grown, plump, and smooth-skinned’. But as every Chinese speaker knows, the above English expressions not only do not convey all evaluative elements contained in the Chinese adjective ‘fung maan’, but also cannot even describe exactly the physical appearance of a ‘fung maan’ woman. This is because Chinese speakers categorize the woman’s body differently from the way English speakers do, using different discriminations in doing so. Under this circumstance, ‘fung maan’ remains an irreducibly Chinese term, not translatable into English. But this does not block understanding. An English speaker can understand the term very well by learning Chinese. In general, the process of learning an alien language or text may well involve a complex process of theory building or a loose sort of reinterpretative reconstruction of the alien language. But this is not a process of translation. Translation might be a desideratum, but not a sine qua non necessity for understanding. Interpretation can serve perfectly well in understanding.
Therefore, translation is neither necessary nor sufficient for understanding. To understand is to learn, not to translate.23 In addition, translation does not help understanding, but often sets obstacles. Linguistic studies show that the best way to learn a foreign language is not to learn it by means of word-by-word translation, but by living in the community of native speakers and learning the language from scratch as a child does. The more you forget your native language, the more effectively you learn, and the deeper you understand the foreign language.
Whether or not there is meaning continuity between the terms of two competing languages in the case of incommensurability is supposed to determine whether mutual translation between the two languages is possible. For the opponents of incommensurability as untranslatability, mutual translation is possible because of the existence of a sort of meaning continuity between two languages. For the advocates, mutual translation is impossible due to a lack of meaning continuity. Now, after we have detached the notion of untranslatability from the notion of incommensurability, it is not clear at all how much weight the meaning relation can carry regarding the explication of incommensurability, not to mention how the relation can remain a determinant semantic relation between incommensurable languages.
I would like to point out further that, in general, there is no necessary connection between meaning continuity and translation. Translation does not presuppose sameness of meaning. Whether two languages are mutually translatable depends upon the target language’s capacity for encapsulating substantial expressions of the source language. Such a formulation does not require sameness of meaning. For illustration, let us consider the following artificial case about color classification. Different languages may divide the spectrum in different ways and thereby have different color predicates. Let us imagine two sets, S1 and S2, of color predicates in two different languages L1 and L2. S1 and S2 divide the spectrum in such a way that none of the color predicates of them match up with the color predicates of the other set. Therefore, we have two different color category systems. Let us further suppose that S1 has ‘red’, ‘orange’, ‘yellow’, ‘green’, ‘blue’, and ‘purple’ as its finest-grained color predicates, while S2 has ‘Orange’, ‘Green’, ‘Purple’ at the corresponding level of discrimination. The extensions of S2 can be imagined to be distributed along the spectrum so that ‘Orange’ matches up with the shades of both ‘red’ and ‘orange’, ‘Green’ with both ‘yellow’ and ‘green’, ‘purple’ with ‘blue and purple’ in S1. The conceptual mismatch between S1 and S2 is adjustable since each color predicate of S2 can be defined in S1. The concept ‘Orange’ in S2 can be expressed as ‘both red and orange’ in S1 and so on. More generally, all the color predicates in L2 are translatable into corresponding expressions in L1. Hence, the two languages are translatable. This case shows us that translation does not presuppose the sameness of reference.
Interestingly enough, based on the rule of contraposition, we can easily infer from the above conclusion that meaning continuity is not necessary for translation that the discontinuity or variance of meaning between two competing languages is not sufficient for untranslatability between two languages. If so, Kuhn’s and Feyerabend’s arguments for meaning variance, based on their versions of the contextual theory of meaning, do not entail without further ado the thesis of untranslatability. If this argument against untranslatability holds water (which I think it does), then the same argument can be used against translatability also. Ironically, appeal to the meaning relation (continuity or discontinuity) is a double-edged sword that can be used to hurt both proponents and opponents of the thesis of incommensurability as untranslatability. This establishes that there is no necessary conceptual connection between meaning continuity and the possibility of mutual translation between two languages.
Finally, I would like to point out that untranslatability does not establish anything interesting about many theoretical issues that are supposed to be implicated by the thesis of incommensurability, such as the issues of the rationality of science and scientific progress as well as the debate between relativism and realism. For example, the debate between relativism and rationalism simply does not depend on the possibility of translation. On the one hand, the ineliminable possibility of mutual translation (liberal translation) between two languages does not guarantee either that the speakers of the two languages share criteria of truth, rules of inference, or have a common core of empirical knowledge, or that the two languages enable their speakers to refer to some fundamental common set of referents. On the other hand, the impossibility of mutual translation between two incommensurable languages does not exclude the possibility that the speakers of the two languages may have shared criteria of truth, shared domain of referents, or common criteria of theory evaluation. The alleged semantic obstruction implied by the thesis of incommensurability does not arise from untranslatability.
* * * * *
For the past four decades, the translational-failure interpretation has dominated much of the discussion of incommensurability. Questions of meaning (sense), reference, and translation have been examined both from the standpoint of general philosophical theories as well as from case studies in the history of science. However, in spite of tremendous efforts, little progress has been made toward the goal of understanding and clarifying the thesis of incommensurability. The notion of incommensurability still remains mysterious to us.
Too much attention paid to the meaning relation between incommensurable theories is, I believe, to a large extent responsible for the slow progress that has been made toward establishing the integrity and tenability of the notion of incommensurability. This is because the translational approach is not an effective24 way of exploring the essence of incommensurability. It cannot provide us with a tenable and integrated notion of incommensurability. This is mainly because the notions of translatability and untranslatability are sterile without any independent interpretative power in the case of incommensurability. Reference to the notion of translation neither identifies nor resolves the problem of incommensurability.
Therefore, the received interpretation of incommensurability as untranslatability due to radical meaning change has to be rejected. Of course, the failure of the received notion of incommensurability does not indicate the failure of the very notion of incommensurability. It is not the notion of incommensurability that lets us down, but the received notion of it that does so. In fact, the rejection of the translational-failure approach paves the road to a more promising interpretation of incommensurability, namely, the presuppositional interpretation of incommensurability.
1 The term ‘meaning’ could be used in a narrow sense to refer to ‘sense’ or in a broader sense to refer to ‘sense and reference’. In our following discussions, the term ‘meaning’ will be used in the broader sense in most cases. When it is necessary, I will specify whether it refers to sense or reference.
2 2 See Feyerabend, 1962, pp. 29, 57, 59, 68; 1965a, pp. 114-15; 1965b; 1965c, p. 180.
3 Kuhn, 1970a, 1970b, 1976, 1979.
4 H. Sankey identifies three distinct interpretations of Kuhn’s early position on reference. See Sankey, 1994, pp. 153-61.
5 See C. Kordig, 1971, pp. 35-48.
6 Here the description theory refers to the so-called cluster theory of reference defended by John Searle and others. See J. Searle, 1958.
7 Informal semantics gives a realistic interpretation of the references of scientific terms. According to it, (a) a scientific theory is not a logically reconstructed formal system; (b) the reference of a term is identified with the set of ‘real world’ objects to which it applies. Therefore, the intended references of central concepts of a scientific theory are fixed absolutely, not relative to some specific theoretical framework. The references of the terms of a scientific theory should remain unchanged during theory change.
8 Formal semantics gives a relativistic interpretation of reference. A scientific theory is believed to be a logically reconstructed formal system. The issue of meaning and reference can be dealt with within a formal (model-theoretic) framework. Meaning and reference are determined by and relative to a formal framework, instead of being fixed absolutely.
9 For the distinction between determination of the meanings (sense or reference) of terms and formulation of the meanings of those terms and its significance in the explication of incommensurability, please read Kuhn’s (1983b) response to P. Kitcher’s (1978) strategy of translation.
10 J. English (1978, pp. 57-8) attributes this extreme view to Feyerabend.
11 Kuhn’s LTRK is a restricted thesis which mainly concerns the referents of selected kind-terms, especially the high-level theoretical categorical terms, such as ‘planet’, ‘mass’, ‘force’, ‘compound’, ‘element’, ‘phlogiston’, etc., and less concerns the referents of low-level empirical categorical terms such as ‘alloy’, ‘metal’, ‘physical body’, ‘salts’, ‘gold’, ‘water’, etc. LTRK does not concern the referents of names and other singular object designators, such as ‘Earth’, ‘Moon’, etc. For Kuhn, the references of names and the references or extensions of kind-terms (especially, natural kind-terms) are two separate issues. Kuhn partially accepts causal theory of reference for determination of the referents of names (Kuhn, 1979; 1988, p. 25; also see Sankey, 1994, pp. 163-71). However, Kuhn uses his version of a contextual theory of reference to determine the referents of kind-terms.
12 Kuhn, 1983b, pp. 682-3; 1991, pp. 11-12; 1993a, pp. 325, 329.
13 Kuhn, 1983b, pp. 676-9; 1988, pp. 14-21; 1993a, p. 317.
14 Kuhn, 1983b, p. 683; 1993a, pp. 325-6, 329.
15 Many critics of the thesis of incommensurability explicitly accept the identity of incommensurability with untranslatability. D. Pearce (1987) contends that translatability is necessary, and sufficient under suitable conditions, for commensurability. W. Balzer (1989) claims that incommensurability is intimately linked with a particular kind of potential translation. It is well known that D. Davidson (1984) equates incommensurability with untranslatability and argues that it does not make sense to talk about two untranslatable conceptual schemes. In a similar way, H. Putnam (1981, p. 114), M. Devitt (1984), and many others regard incommensurability as being or implying the untranslatability of languages. H. Sankey devotes a book (1994) to defend the thesis of incommensurability as untranslatability.
16 For some analysis of details, please see N. Rescher, 1980, pp. 326-8, M. Forste, 1998, pp. 137-9, and D. Henderson, 1994, section II.
17 The second argument is suggested by S. Lehmann.
18 Kuhn, 1983b, pp. 669-70; 1988, p. 11; 1991, p. 5.
19 Kuhn, 1983b, pp. 672-6, 679-82; Sankey, 1994, pp. 76-7.
20 D. Pearce contends that translatability is necessary and sufficient under suitable conditions for commensurability. A translation can be considered adequate in the semantic sense if it respects an acceptable model-theoretic reduction of the structuralistic sort. Science as a whole is so rich in concepts and so diversified in its patterns of conceptual change and development that one cannot expect to find one general method of translation that can be used to produce adequate specific translation between any two languages. A successful response to the challenge of the doctrine of incommensurability as untranslatability consists in examining concrete examples of theory change in science and providing translation of the right kind that applies there.
21 It seems to me that Davidson’s above argument does not presuppose any specific notion of translation. It provides a general objection to regarding translation as a sufficient condition of understanding.
22 This argument is essentially I. Hacking’s (1982).
23 For a good defense of the distinction between understanding and translation, see H. Sankey, 1997, chapter 6.
24 I intentionally choose the expression ‘not effective’ because I do not wish to imply that the translational-failure approach is completely wrong, although it does lead to misinterpretation.