Stephen M. Wechsler (wechsler@mail.utexas.edu)
Thu, 24 Aug 1995 14:01:06 -0500
A while back Eun Jung (Yoo) Hong pointed out some apparent problems with the weak UDC analysis of the Korean 'ECM' construction. I've been out of town for the past month so I just got the message. Here are some brief replies. >I agree with Steve Wechsler's observation that Korean 'raising' construction >is different from English counterpart. (Actually it appears to me that >English-like examples form only a subset of raising constructions in Korean.) >Moreover, given the example like (i), I found the weak UDC analysis >very interesting. However, a few questions came to my mind. > First, as it stands, the weak UDC analysis seems to violate HPSG >Raising Principle, since despite the presence of a non-role-assigned >argument, an unsaturated phrase does not appear in the SUBCAT list of >a verb. Yes. A similar issue arises for the tough-construction. As discussed in previous postings on this list (and elsewhere) it seems plausible that the subject of the tough construction is role-assigned by the matrix predicate. How can this issue be decided for the Korean ECM? I'm not sure. There are no pleonastics. We can look at idioms (there is some discussion of ECM and idioms in Jeong-Shik Lee's dissertation), but we need to be careful with syntactic arguments from idioms, given the observations of Nunberg et al 1994 (Language). So I'd be interested in any ideas about types of evidence regarding role-assignment. > Second, in the examples like (a) where the second complement is a >PP, it seems hard to assume a null operator or a gap within the >complement: > >(a) Na-nun Mary-lul chinkwu-lo sayngkakha-n-ta. > I-Top Mary-Acc friend-as think-Pres-Dec > 'I regard Mary as a friend.' > The above question about role-assignment is relevant here too: could Mary-lul be an argument of sayngkakha-? > Third, I'm wondering how the following kind of examples can be >explained: > >(b) Ku chayk-ul na-nun [[[hanpen-man ilk-eto] kamtongha-l] saram-i > the book-Acc I-Top once-only read-though is-moved-Rel people-Nom > manh-ta-ko] sayngkakha-n-ta. > many-Dec-Comp think-Pres-Dec > 'I think that there are many people who will be moved even though they > read the book only once.' > >If I understand correctly, in the 'null operator' (weak UDC) analysis >of (b), a null operator (which is coindexed with the NP _ku chayk-ul_) >will move out of the innnermost adjuct clase [hanpen-man ilk-eto] which is >embedded within a relative clause. However, violation of syntactic >island seems free in this case. Moreover, the gap position can be filled with >a resumptive pronoun as in (c): > >(c) Ku chayk-ul na-nun [[[hanben-man kuket-ul ilk-eto] kamtongha-l] > saram-i manh-ta-ko] sayngkak-ha-n-ta. > I'll have to think about this. A general comment: data judgments can vary, so we really need to look at constrasts between 'ECM' and non-ECM constructions, varying case on the relevant NP between Nom and Acc but holding everything else constant. Jae-Gyun Song (1994) did this and found a greater tendency for island effects with the Acc NP. But I don't think he or anyone has done a definitive study. That's all for now. Best, Steve
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:03 PST