Categoriality and object extraction in Cantonese serial verb

Categoriality and object extraction in Cantonese serial verb constructions Elaine J Francis and Stephen Matthews. 2006. Natural Language and Linguistic Theory 24: 751 -801.

Cantonese coverbs Coverbs have preposition-like functions and commonly occur as V 1 of a [VP 1 VP 2] serial verb construction, as in (1): (1) Keoi jigaa hai go gaan sezilau gin haak S/he now at that CL office see client ‘She is meeting clients in that office now. ’ (2) *[[Keoi jigaa hai __i gin haak RC] go gaan sezilaui NP] hou daai gaan S/he now at __i see client that CL officei very big CL Intended: ‘The office that she is meeting clients in is very big’ – Are “coverbs” better characterized as verbs or prepositions? – What is the nature of the apparent constraint on extraction of the coverb’s object, shown by the unacceptable relative clause in (2)? 2

Acceptability rating task • • Rating scale of 1 (‘highly acceptable’) to 4 (‘highly unacceptable’) Six verbs/coverbs were selected as V 1: 1. pui 4 陪 ‘accompany’ 2. doi 6 代 ‘replace, in place of’ 3. wan 2 搵 ‘seek’ 4. gan 1 跟 ‘follow, with’ 5. hai 2 � ‘(be) at’ 6. tung 4 同‘with’ • Three factors of two levels each were manipulated to create eight sentence types. There were six sets of eight experimental sentences (one for each verb) plus eighty filler sentences. 1. aspect vs. no aspect (between-groups factor) 2. one-verb vs. two-verb clauses (within-group factor) 3. object extraction vs. no object extraction (within-group factor) • 40 participants, HKU undergraduates ages 18 -21, native speakers of Cantonese, non-linguists, assigned randomly to one of two groups – aspect or no aspect Experimental and filler sentences presented through headphones as recorded speech, using a computer interface to collect responses. Ordering was random for each session. • 3

Sentence materials- no aspect condition One-verb Basic clause Keoi jigaa hai go gaan sezilau S/he now at that CL office ‘She is in that office now. ’ One-verb Object-extracted relative clause Keoi jigaa hai __i go gaan sezilaui hou daai gaan S/he now at __i that CL officei very big CL ‘The office that she is now in is very big’ Two-verb (Coverb) Basic clause Keoi jigaa hai go gaan sezilau gin haak S/he now at that CL office see client ‘She is meeting clients in that office now. ’ Two-verb (Coverb) Object-extracted relative clause Keoi jigaa hai__i gin haak go gaan sezilaui hou daai gaan S/he now at __i see client that CL officei very big CL ‘The office that she is now meeting clients in is very big’ 4

Results: sentence type and aspect • Aspect marking had no significant main effect. Coverbs pattern as verbs. • Object-extracted two-verb sentences were rated worst of the four sentence types, in support of some type of adjunct island constraint on extraction. • Object-extracted one-verb constructions were rated worse than expected. • Aspect marking improved extraction for two-verb sentences and had the opposite effect for one-verb sentences, effectively neutralizing the distinction. 5

Results: sentence type by participant • • Here is a graph of participant 102’s responses. Although no two participants gave the same ratings, the general pattern of sentence types was similar across participants. 6

Results: sentence type by participant • Although no two participants gave the same ratings, the general pattern of sentence types was similar across participants. 7

Results: coverb obj. extraction by verb • All coverbs showed strong extraction effect. • However, acceptability of extraction differed for different coverbs: hai 2 ‘at’ and tung 4 ‘with’ were less acceptable than the other four verbs. 8

Results: coverb vs. simple verb obj. extraction • Although two-verb sentences were generally rated worse, four of these verbs showed an extraction effect even when used in a simple one-verb clause, perhaps due to their common usage in coverb constructions. • One verb doi 6 ‘replace’ showed the opposite pattern from what we expected. • No aspect group is shown here. Distinction was mostly neutralized for aspect group! 9

Challenges and weaknesses • Problems reconciling theory and data – Adjunct island effect is stronger with certain verbs: not predicted by any syntactic theory – Additional extraction effect for one-verb sentences, albeit a weaker one: not predicted by any syntactic theory – Cantonese lacks classic CED effect with subordinate clause adjuncts, making adjunct island account more complicated • Experiment could have been better controlled by including regular verbs (non-coverbs) in the design and a larger number of token sets. • Some theoretical arguments were based only on constructed examples and informal judgments of our consultants (e. g. , 39 -40 on p. 783). • Construction-specific adjunct island constraint is not exactly an elegant theoretical solution. • Article could have been shorter and less redundant. 10

Strengths • Unifies preposition stranding and adjunct island constraints under a single processing principle (Avoid Competing Subcategorizors) while maintaining category distinctions • Provides a challenge for generative theories of adjunct island constraint • Formal experiment makes data more reliable: – With 40 participants and six sentence sets, we can be more confident in our conclusion that coverbs allow verbal aspect and that the extraction effect is genuine – Original hypotheses based on informal judgments had to be revised • Formal experiment makes data more informative: – Adjunct island effect appears to be stronger with certain verbs-- evidence that coverbs are in transition, with some having become more preposition-like than others. – Interesting interaction between aspect marking and extraction – Raises questions for further research, such as regarding frequency of use for different coverbs and processing constraints on object-extracted relative clauses – Evidence that aspect marking can occur on a non-head in an SVC (or that the relevant notion of headedness is semantic). This was unexpected but interesting. 11

Discussion • 1. How might the phenomena discussed in this paper be captured in terms of synchronic consequences of diachronic change? What do these kind of transitional states mean for the synchronic grammar? • 2. How can we account for the fact that extraction caused degraded acceptability judgments even for simple one-verb sentences? Is this the grammar, or something else? • 3. How could the relevant extraction constraint be formulated in a theory such as Minimalism? • 4. What should we make of the differences between different verbs? Is it theoretically significant that there was no adjunct island effect for one of the verbs? What additional data would be helpful for answering this question? • 5. How might the methodology have been improved? 12