Gosse Bouma (gosse@let.rug.nl)
Thu, 23 Jan 1997 14:00:11 +0100 (MET)
Bob Levine writes: > (1) You cannot avoid putting adjuncts on the ARG-ST list if you're > going to extract them using Ivan's traceless theory of connectivity, > because only by putting them on the ARG-ST list will you be able to > define a single morphosyntactic condition triggering binding domain > effects for languages which display such effect, in the general case. > Verbs which subcategorize for complements containing a gap will have > both COMPS lists and ARG-ST lists which contain elements bearing a > SLASH specification. But the V from whose COMPS list the filler is > missing (via the CELR) will have a SLASHed element only on its ARG-ST > list, obviously; inspection of the COMPS list will contain no > information which could trigger the appearance of binding domain > effects---that's built into the whole approach. So the only common > property of the two cases that allows one to define a unitary property > of heads' lexical entries which triggers binding domain morphosyntax > is that the ARG-ST list contains a SLASHed synsem object. In that > case, adjuncts have to go on ARG-ST lists and the cataphora facts > can't be evaded, unless you change the binding theory fairly > dramatically. I just want to check whether I understand the implications of the pt. above correctly: There are languages (Chamorro etc) where certain markings on the (verbal) head only show up if this head is in fact the head of an `incomplete constituent' (i.e. its projection is on the path of an unbounded dependency). Hukari and Levine show that these effects also show up in cases of `adjunct extraction'. The most promising way to account for this so far, I think, is Sag's reformulation of the NLFP as a lexical constraint. (Note that the reformulation of GPSG's theory of UDC's (i.e. SLASH is not a head feature, and metarules can have phrasal heads) as proposed in Hukari and Levine does account for adjunct extraction, but does not make it possible to distinguish between heads of UDC paths and heads which are not on such a path.) So, unless we find another way to account for such data, this is additional evidence for putting adjuncts on ARG-ST. > To invert the argumentation once again, the success of the binding > theory in effectively predicting the cataphora and antireconstruction > facts for unselected element might well be taken as a sign that such > elements *are* in fact unselected. Following this line of reasoning, > one *wouldn't* wish to adopt a theory of connectivity which redefined > unselected elements as selected simply so one could get rid of traces. Note also that the crucial pt. in accounting for Chamorro etc is lexical slash amalgamation, not the elimination of traces. (I.e. in theory, one could have lexical slash amalgamation and still have traces.) Of course, this leaves the anaphoric binding data to be explained by `whatever residual differences remain' between selected and unselected elements. However, I think the data from Dutch and Japanese might provide independent evidence for such a reconsideration of the binding theory. If adjuncts are simply never on ARG-ST (or SUBCAT, if we want to translate this into the terminology of the HPSG II binding conditions), there is no way in which the `long distance reflexives' in the examples below can be `o-bound': Mitio ga zibun o hihan sita ato de taihen kanpai sita Michio NOM self ACC criticize after great.deal worried 'After Michio_i criticized self_#i/j, [I/she/...]_j worried a lot.' Eva_i zag mij plotseling voor zich_i staan Eva saw me suddenly before ZICH standing Thus, selecting adjuncts by means of a head-adjunct scheme cannot be the solution either. Of course, there is a middle way here, which is to say that adjuncts are sometimes introduced by means of a head-adjunct rule, and sometimes as complements. It seems to me, however, that if the two different modes of selection have such far reaching consequences for binding and for extraction, it should be relatively easy to check whether this is a coherent position. (I.e. the cataphora effects observed by Bob should only be possible with unselected, and therefore non-extractable, adjuncts.) Tibor Kiss writes: > Since the adverbs are added lexically, their scope is also defined > lexically and hence by definition is always narrower than the scope of > nominal complements of the verb. At least for German (and I assume for > Dutch, too), the prediction is not borne out by the facts, witness > (3). > > (3) Er hatte oft einen Jungen verletzt. > He has often a boy hurt > > (3) can only have the interpretation often(a boy(he hurt him)). The > LR-account would predict the reading a boy(often(he hurt him)) > which is not possible. The assumption that adverbs introduced by a LR must always have narrower scope than the complements of a verb is not correct, it seems to me. Given a Pollard/Yoo stype of analysis of q-scope, a verb may assign the scope of its nominal arguments lexically (there is even a fn in Pollard/Yoo citing German data provided by Kiss as evidence for this possibility). An Adjunct LR, semantically, works as follows: [ comps #1 ] ==> [ comps < #1 | Adv > ] [ cont #2 ] [ cont Adv'(#2) ] If the q-store meaning of the object is retrieved lexically, example (3) will receive the interpretation often(a boy(he hurt him)). If the q-store meaning of the object is retrieved later on, we will obtain the meaning a boy(often(he hurt him)). The latter, in Dutch, is unlikely but not impossible, according to my informants. Note that obtaining the two readings does not require adding adverbs to q-store. Gosse. -- Gosse Bouma, Alfa-informatica, RUG, Postbus 716, 9700 AS Groningen gosse@let.rug.nl tel. +31-50-3635937 fax +31-50-3636855
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:36:03 PST