Introduction

Unlike any other skill that you could care to mention (algebra, tennis, driving, playing the bagpipes), language can work well despite poor execution. Its meaning is recoverable even when its form is incorrect. ‘Me Tarzan, you Jane’, or ‘ET phone home’ are not grammatical, nor particularly subtle, but they get the job done. This useful characteristic of language means that learners can decode and encode the semantic contents of a message without attending to all of its syntactic packaging. Anderson and Lynch (1988), for example, show how contextual and discourse knowledge-sources enable comprehension to proceed for native and non-native speakers without exhaustive bottom-up analysis of a message. Schematic knowledge enables the redundancy in language to be exploited for meaning-extraction. Productively, it is clear that second language (L2) users draw upon a range of communication strategies (Bygate, 1988; Faerch & Kasper, 1983; Kasper & Kellerman, 1997) enabling them to communicate meanings despite limited resources. Individual case studies, e.g., of Wes (Schmidt, 1983) and Schmidt (Schmidt & Frota, 1986), as well as larger scale research such as that evaluating immersion education in Canada (Harley, Allen, Cummins & Swain, 1990; Swain & Lapkin, 1982) also indicates how the need to engage in output does not automatically lead to acquisitional progress. In evolutionary terms, it is entirely plausible that the language system should be robust in this way – requiring crystalline precision in all communication could well have been decidedly dangerous for our ancestors' health and survival.