Quantitative measures of cognitive synonymy : a corpus study of English moderators (rather, quite, pretty, and fairly)

In this paper, I focus on four degree-modifiers of adjectives: rather, quite, fairly, and pretty. These adverbs form a subclass of intensifiers (Bolinger, 1972; Lyons, 1977) known as 'compromisers' (Quirk et al., 1985) or 'moderators' (Allerton, 1987; Paradis, 1997): (1) I think it's fairly easy for anyone to get anything they want (...). (COCA) (2) He seemed in pretty good form. (COCA) (3) Obama' s whole pitch is quite different. (COCA) (4) I heard a rather odd conversation. (COCA) Paradis (1994: 160; 1997: 71) contends that moderators are cognitive synonyms (after Cruse, 1986) because they all serve to fix the properties that gradable adjectives denote to a moderate level. Although moderators are not interchangeable in all contexts, cognitive synonymy is nevertheless evidenced in corpus by a significant amount of overlapping collocational preferences. I want to test the claim that moderators are cognitive synonyms by contrasting them in their distinctive collocational preferences. I thus hope to avoid the problems faced by previous corpus-based approaches to degree modifiers, which rely too much on raw frequencies or coarse-grained relative frequencies, and therefore underestimate significant context-related differences (Paradis, 1997; Lorenz, 2002; Kennedy, 2003; Simon-Vandenbergen, 2008). I extract all combinations from the 410M-word Corpus of Contemporary American English (Davies, 2008-) and implement two techniques from collostructional analysis. First, I conduct a collexeme analysis to determine which adjectives are most strongly attracted to each moderator (Stefanowitsch & Gries, 2003). My results show that some adjectives, such as good, large, different or clear, display collocational preferences with more than one moderator. At this stage, the cognitive synonymy of moderators is partly verified and cannot be dismissed. Then I perform a multiple distinctive collexeme analysis to determine the adjectives that are distinctively associated with each moderator (Gries & Stefanowitsch, 2004). It turns out that moderators attract different classes of adjectives. For example, fairly attracts adjectives that denote novelty (new, recent), commonness or constancy (common, constant, consistant, regular), whereas rather attracts adjectives that denote problematic identification (odd, unusual, strange, bizarre, etc.) By examining how moderated adjectives fall into distinct semantic classes, my corpus study eventually undermines the claim that moderators are cognitive synonyms. Conversely, my results substantiate and refine two other assumptions. The first assumption is that moderators have undergone subjectification (Nevalainen & Rissanen, 2002; Athanasiadou, 2007). Indeed, not only do moderators index the speaker's assessment of intensity on an abstract scale, but they also index the speaker's perspective differently depending on what property is moderated. For instance, speakers tend to modify easiness with fairly and difficulty with pretty or quite. The second assumption is that moderators are intersubjective (Paradis, 1997: 69). My data confirm that moderators typically hedge epistemic assessments of scalar properties in contexts where their absence would be face-threatening (e.g. rather simplistic/lengthy) or where the speaker lacks evidence for an objective statement (e.g. quite correct, pretty smart)