Elaine J. Francis and Stephen Matthews
The ‘coverb’ construction in Cantonese, a serial verb construction (SVC) in which the first verb (the ‘coverb’) has a preposition-like meaning and function, presents a challenging theoretical puzzle. Extracting the object of a coverb typically results in marginal or low acceptability, as illustrated for the coverb doi ‘replace’ in (1).
(1) ??[[ngo doi __i gaau syu RC] go go jani NP] jau beng-zo (Relativization)
I replace teach book that CL person again sick-ASP
‘The person in whose place I teach the class is sick again.’
In studies on Mandarin, such facts are typically explained in terms of a constraint on ‘preposition stranding’, whereby coverbs are analyzed as prepositions rather than verbs (Zhang 1990, McCawley 1992). However, Cantonese coverbs display the typical morphosyntactic properties of verbs, suggesting that they are not prepositions. For example, similar to other verbs, coverbs allow aspect marking, as in (2).
(2) ngo doi-gwo ni go jan gaau syu (Aspect marking)
I replace-ASP this CL person teach book
‘I’ve taught a class for (in place of) this person.’
In this paper, we report on an experimental study in which an acceptability rating task was used to compare coverbs with transitive verbs in SVCs in which the object of V1 is extracted. Participants rated sentences containing one of six verbs/coverbs in V1 position and varying along three orthogonal dimensions: (a) one-verb vs. two-verb clauses; (b) aspect vs. no aspect on V1; and (c) extraction vs. no extraction. The experimental results show no distinction between coverbs and (other) transitive verbs, contradicting a preposition-stranding analysis. Ratings are marginal to low in extraction conditions for all verbs tested, and differences among individual verbs are not predictable based on their purported status as either ‘verb’ or ‘preposition’.
A structural analysis following Y. Li (1993) and Law (1996) would attribute adjunct status to VP1 in cases where extraction from VP1 is ungrammatical (as per Huang’s Condition on Extraction Domain). Unlike the preposition-stranding analysis, this approach could accommodate the observed variability in acceptability: those verbs that are interpretable only as adjuncts in the experimental sentences receive significantly worse ratings than those verbs which may be interpreted as heads. Furthermore, this analysis would predict aspect to be incompatible with sentences for which V1 is interpretable only as an adjunct, since aspect marking occurs only on the head verb of a Chinese SVC (Dechaine 1993, Y. Li 1991). However, our results show no significant effect of aspect marking.
In view of these data, we
propose an alternative analysis from the perspective of a multi-modular theory
of grammar (Sadock 1991, Jackendoff
1997, 2002), re-formulating the relevant constraint as a semantic constraint on extraction from modifying phrases. This
analysis is similar to the structural analysis but allows for the possibility
of ‘mismatch’ between syntactic structure and semantic structure. Thus, a verb that is the syntactic head of
the SVC (allowing aspect) but a semantic modifier is predicted to disallow
extraction of its object, and vice versa, in a manner consistent with the full
range of experimental results.