On comparative concepts and descriptive categories: A reply to Haspelmath

In his discussion note (Haspelmath 2010), Martin Haspelmath (henceforth ‘H’) proposes a radical disconnect between the grammatical analyses of individual languages and crosslinguistic (typological) generalizations about grammatical patterning: ‘The analysis of particular languages and the comparison of languages are thus independent of each other as theoretical enterprises’ (p. 682). In an earlier publication (Haspelmath 2007), H had argued, along with Dryer (1996, 1997), Croft (2001), and others, that it is impossible to provide unambiguous and reliable tests for identifying a particular grammatical entity in one language (‘subject’, ‘noun’, ‘dative’, or whatever) as being ‘the same construct’ as in another language. Hence, ‘[e]ach language has its own categories, … [which] are often similar across languages, but the similarities and differences between languages cannot be captured by equating categories across languages’ (p. 664). In his view, typology is carried out by means of universally applicable ‘comparative concepts’, which include conceptual-semantic concepts, general formal concepts such as ‘precede’, ‘identical’, and ‘overt’ (which he assumes ‘to be unproblematic’, p. 670), and other comparative concepts. Interestingly, H claims that ‘in practice typologists generally work with such special comparative concepts’ (p. 663), an assertion that applies equally to generative typologists, who ‘have in practice used comparative concepts, rather than crosslinguistic categories whose instantiations in individual languages are well motivated’ (p. 676).

I argue below that universal comparative concepts and language-particular descriptive categories are each highly problematic in and of themselves. It is only by means of working out the interplay between the language-particular and the language-independent that we can hope to understand either. This is just the sort of interplay that H steadfastly rejects. The great irony is that the more successful H is at proposing a set of useful comparative concepts, the more he undercuts his assertion that they ‘cannot be right or wrong’, but can ‘only be more or less productive’ (p. 678). If the comparative concepts that enter into a typological generalization are ‘productive’, then we want to know why they are productive. The answer might possibly be due to an innate universal grammar (UG), or possibly be a consequence of a complex interplay of cognitive and functional factors. But there has to be an answer—productive hypotheses and lines of research derive their productivity from the fact that they shed light on the essential nature of the phenomena being investigated, while nonproductive hypotheses and lines of research do not do so. In simple terms, a productive hypothesis is closer to being right than a nonproductive one. H’s outlook reminds me of that of the post-Bloomfieldian linguists of the 1940s and 1950s. They insisted that their procedures for linguistic analysis were arbitrary, yet, as Chomsky and Halle pointed out in many publications, they were careful to choose only those procedures that led to results that were revealing about how grammars actually work. Their practice was better than their theory, as I would argue is the case for H.

H devotes space in his discussion note to repeating arguments that Subject is not a crosslinguistic category. The problem, as H sees it, is that subjects in English are not definable by the same criteria as subjects in Tagalog, which are not definable by the same criteria as subjects in whatever other language one might name. Since the criteria devised for identifying subjects differ from language to language, we need to ‘[return] to the structuralist position that each language has its own categories and that there are no crosslinguistic categories’ (p. 667).And yet, two of his typological generalizations—one about adjective order (p. 670) and one involving reflexive pronouns (p. 673)—refer to the notion ‘subject’ (and one of them refers to the notion ‘object’ as well). I see no way to avoid the conclusion that H is guilty of extreme inconsistency. If H is correct that no single criterion identifies subjects crosslinguistically, then how can he go ahead and formulate such a criterion, which he must do in order to capture the two generalizations referred to above? Here and elsewhere, H undermines his own position that there is a total disconnect...