Cognitive explanations, distributional evidence, and diachrony
University of Pavia
Cognitively oriented approaches to the study of language standardly use synchronic distributional evidence to make assumptions both about the psychological mechanisms that lead speakers to create particular constructions, and about the components of a speaker's mental representation of their language. Yet, as synchronic distributional patterns are a result of specific diachronic processes, any assumption about the psychological mechanisms or types of mental representation underlying particular patterns should take into account the diachronic processes that give rise to these patterns. Based on evidence from different languages and language families, the chapter discusses several diachronic processes pertaining to the development of various types of alignment systems and prototype effects in dependent clauses. It is shown that these processes provide no evidence for a number of assumptions about psychological mechanisms and a speaker's mental representation that have been made on synchronic grounds in order to account for the relevant distributional patterns. It follows that this type of assumptions cannot be inferred directly from synchronic distributional patterns, and should be investigated independently of these patterns.
This chapter addresses the general issue of the relationship between theory and data in cognitive linguistics with regard to what types of linguistic data can be used to make assumptions about the structure of grammatical representation in a speaker's mind, and about the types of psychological mechanisms that lead speakers to create particular constructions. In particular, it is argued that, while such assumptions are usually based on synchronic distributional evidence, they should take into account the diachronic processes that give rise to the relevant distributional patterns.
A central tenet of cognitively oriented approaches to the study of language, including cognitive linguistics and the functional-typological approach, is that the properties of linguistic structure can provide insights on a number of aspects of a speaker's mental organization. These include both psychological mechanisms that may govern a speaker's creation, acquisition, and use of individual constructions, and conceptual entities that are specifically represented in a speaker's mind. For example, linguists working within the functional-typological approach generally assume that a speaker's creation of novel constructions may be driven by a tendency to maximize form-function correspondence (iconicity), or to minimize the amount of information that is encoded explicitly in an utterance (economy). Also, both functionally oriented typologists and cognitive linguists often assume that the grammatical categories and relations that are part of a speaker's knowledge of their language (for example, noun, verb, or subject) have a prototype structure with central and peripheral members. Sometimes, these linguists also assume that grammatical representation in a speaker's mind includes what is called a universal conceptual space, that is, a space where different concepts are always arranged in the same way depending on their relative degree of similarity (see, e.g., Croft (2001) or Haspelmath (2003)).
These assumptions are generally made on synchronic grounds. While research on language change, particularly grammaticalization studies, has provided a considerable body of data on various mechanisms that feature prominently in cognitively oriented theories of grammar (for example, metonymization), the standard way to formulate hypotheses about a speaker's mental organization is to look at the structural properties of individual constructions as manifested at the synchronic level. If these properties are compatible with particular assumptions about a speaker's mental organization, then they are taken as evidence for these assumptions, regardless of how the relevant patterns originated in individual languages. This means that, in many cases, particular psychological principles or models of grammatical representation are assumed to motivate individual patterns, but these principles or models are postulated based on the patterns in themselves, not any specific evidence that they play a role in the mechanisms that lead speakers to create or use these patterns.
Thus, for example, economy is invoked to account for the distribution of zero vs. overt morphological marking across particular categories (e.g., singular vs. plural, or subject vs. object: Croft (2003), among several others), but this is based on the distribution itself, rather than any specific evidence that economy plays a role in the development of the markers. The presumed prototype structure of particular grammatical categories is used to account for the fact that the members of these categories do not display exactly the same range of properties cross-linguistically (Hopper & Thompson (1984), Croft (1991), among others), but this structure is postulated based on the synchronic distribution of the various properties, not the fact that it can be shown to play a role in how they originate in individual languages. The existence of a universal conceptual space that is part of a speaker's mental representation is used to account for the fact that languages display recurrent multifunctionality patterns cross-linguistically, that is, patterns whereby the range of meanings encoded by a single form is typically the same from one language to another. The conceptual space, however, is postulated based on the multifunctionality patterns in themselves, not the fact that it can be shown to play a role in the diachronic processes that give rise to the relevant multifunctionality patterns in the various languages (see Cristofaro (2010) for detailed discussion of this volume).
Over the past decades, however, a number of linguists working within the functional-typological approach, particularly Bybee (1988, 2008), and Dryer (2006), have argued that, since the distributional patterns for particular constructions are a result of specific diachronic processes, the explanations proposed for individual patterns should refer to these processes, rather than the patterns in themselves (see also Newmeyer 1998, 2002, and 2005). For example, Bybee (1988) has argued that a number of word order correlations, such as the correlation between the order of adposition and noun and that of possessor and possessee in possessive constructions, originate because one of the two constructions involved develops from the other and maintains the original order of the latter, not because of (ultimately psychological) principles that may be formulated in order to account for these correlations on synchronic grounds, such as processing ease.
This argument implies that, in principle, any assumption about the psychological mechanisms or types of mental representation underlying particular distributional patterns should take into account the diachronic processes that give rise to these patterns. If it cannot be shown that particular psychological mechanisms or types of mental representation play a role in these processes, then it is not straightforward that the patterns can be taken as evidence for those mechanisms or types of mental representation.
This is in line with a point that has been repeatedly raised by linguists from different theoretical backgrounds over the past decade, namely that distributional evidence is not, per se, evidence about mental representation, because distributional patterns are in fact compatible with several types of mental representation (Croft (1998), Dabrowska (2004), Haspelmath (2004)). In the actual research practice, however, assumptions about psychological mechanisms and mental representation made on synchronic grounds are usually not tested against the diachronic processes that give rise to the relevant distributional patterns. Conversely, even when particular processes of language change point to specific psychological mechanisms, as is often the case with grammaticalization processes, these results are not usually used to test the assumptions that have been made about the relevant distributional patterns on synchronic grounds.
In what follows, various cases will be discussed illustrating a number of ways in which the diachronic processes that give rise to particular distributional patterns may not support the assumptions about a speaker's mental organization that can be made in order to account for these patterns on synchronic grounds. Attention will be focused on two types of phenomena, the organization of alignment systems (Section 2) and so-called prototype effects in parts of speech categories (Section 3). While these phenomena have mainly been investigated within the functional-typological approach, rather than cognitive linguistics proper, they involve a number of synchronic distributional patterns that have been used to make hypotheses about specific psychological mechanisms and types of mental representation. Based on data on the history of several languages from different families, it will be shown that at least some of the diachronic processes that may give rise to the various patterns do not actually provide evidence in support of these hypotheses. The same patterns may be the result of different diachronic processes in different cases, and these processes, while ultimately cognitively based, are not amenable to a unified explanation, and involve different mechanisms than those that can be postulated on synchronic grounds.