Discovering structure

We probe grammatical person differences comparing 3sg with 1sg in actual language use, using Spanish variable subject pronoun expression. We reconfigure the familiar constraint of accessibility to distinguish between clause linking (prosodic and syntactic connectedness) in coreferential contexts and distance from the previous mention (intervening clauses) in non-coreferential contexts. This refinement reveals that accessibility impacts 1sg earlier than 3sg, for which the pronoun rate rises more slowly with increasing distance. At the same time, a greater proportion of 3sg than 1sg pronominal and unexpressed subjects occurs in coreferential contexts. 3sg pronominal and unexpressed subjects thus tend to cluster more closely. By these differences in the workings of accessibility and in contextual distribution, 3sg, unlike speech act participant 1sg, is a transient person in discourse. 1 Grammatical person in cross-linguistic perspective Linguistic structure is not a given but “must be discovered through analysis”, as Ricardo Otheguy has urged (2002, p. 400). Grammatical categories espoused by linguists do not always coincide with actual usage, and though the category of person features in accounts of subject pronoun expression, its effects are not well understood. Admittedly, variable use of subject pronouns in Spanish has been so widely analyzed that we might wonder if there is anything more to be discovered about it. After all, “multiple studies ... across communities, across settings, and across the lifespan reveal the very consistent nature of structured variation” in subject pronoun expression (Carvalho, Orozco & Shin 2015, p. xxii). Grammatical person is consistently found as a probabilistic constraint on subject expression, often the strongest, as in Ricardo Otheguy’s own analyses of Spanish in New York City (Otheguy & Zentella 2012; Otheguy, Zentella & Livert 2007). Person effects are reported for a range of other languages as well, for example, Arabic (Owens, Dodsworth & Kohn 2013, p. 268; Parkinson 1987, p. 356), Auslan (McKee, Schembri, McKee & Johnston 2011, p. 388), Bislama (Meyerhoff 2009, p. 311), Cantonese and Russian (Nagy, Aghdasi, Denis & Motut 2011, p. 141-142), and Turkish (Koban 2011, p. 362). Here we try to explain this effect. A key distinction often made is between first and second person in contrast with (animate) third person, said to hold cross-linguistically. For example, in languages with split ergative marking based on person, the most common configuration is one in which first and second persons are treated differently from other NPs, including (animate) third person (e.g., Delancey 1981, p. 628). In languages with switch reference marking, this is sometimes applied only on third person verbs (e.g., Haiman