A longstanding problem in the study of syntactic variation is determining the envelope of variation. That is, what are the variants that speakers choose among when they speak? This problem is usually thought of in terms of semantic equivalency: are the variants in question really “different ways of saying the same thing,” or is their selection at least partly based on semantic or pragmatic differences among the variants? But there is another problem facing an analyst of syntactic variation before the work of determining the constraints on variation can begin. To the extent that we consider the statistical tendencies of speakers to use one variant or another part of grammar, we have to ask: Which utterances are things that should be explained via a competence grammar, and which are things that should be explained away via performance factors?

This problem is brought into focus when we study the effect of processing difficulty on variant selection. If one variant is more likely under difficult processing conditions, and another more likely under easy processing conditions, this could be a sign that one variant is an error, more likely to be made under pressure. For example, Staum Casasanto & Sag (2009) found that extra complementizers are more likely to be inserted when the distance between the complement-taking verb and the subject of the complement clause is long. It’s possible to describe this type of pattern in the same terms that we describe other effects on syntactic variant selection, such as style, register, social, semantic, or lexical effects. But to do so misses a critical point: there may be a non-arbitrary, functional relationship between the conditioning factor and the measured outcome. Not only that, but the variant that occurs in the difficult conditions may not be part of any speaker’s idiolect, in terms of grammaticality. If a variant is produced only under conditions of processing difficulty, what aspect of a speaker’s knowledge are we describing when we describe its distribution?

In this talk, I’ll present data from experiments investigating putatively processing-based syntactic variation, propose some ways of distinguishing between grammar and processing, and discuss the limitations of these methods. I’ll argue that although there are strategies we can use to classify variables as inside or outside the purview of grammar, we can only use these once we acknowledge that we need to have different notions of the envelope of variation for different types of analysis of variation.