How abstract is language?

Abstract

Linguists typically argue in favor of abstract representations for language, but many recent approaches to language within the cognitive science framework have stressed the importance of surface forms. In these approaches, abstract representations emerge as secondary properties at best, and at worst, are absent entirely. One domain in which linguists invoke a particularly abstract level of representation is Semitic word formation. In Semitic languages (e.g. Arabic, Hebrew), a word consists of a three-consonant root (e.g. Arabic /ktb/ signifies the semantic domain of writing) which gets interleaved into fixed templates: KaTaB means ‘wrote’, KuTiB means ‘was written’ and maKTaB means ‘desk’. Many of the templates themselves bear consistent meaning of the sort associated with inflectional morphology in English (e.g. Arabic /a-a/ signifies the perfective, /u-i/ the passive). Additional grammatical meanings can be added through prefixes and suffixes. This analysis requires at least three abstract elements in the representation of even a simple word: the root, the template, and the rules (phonological and semantic) for combining the two. An alternative analysis is simply to say that people who speak these languages know the words katab, kutib, maktab, etc. as independent units and the abstract analysis reflects a level of meta-analysis necessary for linguists but not the average speaker.In a recent paper, Prunet et al.1xThe mental representation of Semitic words. Prunet, J.-F et al. Linguist. Inq. 2000; 31: 609–648CrossrefSee all References1 address the question of the psychological validity of abstract roots by examining the speech errors of an Arabic–French bilingual aphasic patient. Using a variety of different production tasks (including reading aloud, repetition and picture naming), they found that their subject made a large number of root errors in Arabic in which he re-ordered root consonants, but almost no errors in the ordering of the template. Thus, if the target word were ‘kutib’, he might say ‘tukib’, but would not say ‘kitub’. Similarly, he showed almost no re-ordering errors involving consonants from the prefixes or suffixes; only consonants from the root were affected. Thus, if the target were ‘maktab’ (ma is a prefix), he might say ‘matkab’ but not ‘kamtab’. When the same patient was tested in French, he showed over 10 times fewer errors of any sort, and the re-orderings most often involved switches between adjacent elements as opposed to the discontinuous switches found in Arabic. The authors note that there are other known instances of root consonant re-ordering in Semitic languages: in speech errors of non-aphasic speakers, in childrens language games, and in the historical development of the Semitic languages. Moreover, they review parallel evidence from a different aphasic patient, normal speech errors, and language games showing disruptions to the template that leave the root consonants intact. Taken together, the evidence favors an analysis of Semitic languages that includes the root and template as psychologically real parts of the grammar. This article demonstrates, therefore, the importance of such abstract representations, not only for accounts of language itself, but for understanding how human beings represent and access their language.