Ivan A. Sag (sag@csli.stanford.edu)
Tue, 5 Aug 97 16:59:30 PDT
All, We meant to circulate this earlier, but the linguistic institute seems to have gotten in the way. August greetings... ----------------------------------------------------------------------- Memo on Clitic Doubling and Left Dislocation August 2, 1997 From: Anne Abeille, Daniele Godard, Ivan Sag ----------------------------------------------------------------------- This is a tentative anwer to the two questions asked a few weeks ago about how to do clitic doubling and left dislocation in HPSG. The analysis of left dislocation sketched below draws on the analysis of dont relative clauses with a resumptive pronoun in French (Abeille et al 97b, building on earlier work by Maling and Zaenen and by Sells); that of clitic doubling draws on the analysis of French clitics in two recent papers (Sag and Miller 97, Abeille et al 97a) which avoids both traces and lexical rules. 1. Clitic Doubling (CD) A problem we had with the question is that it is not clear to us what clitic doubling exactly means (in terms of distribution) so we have identified at least 4 variants. We assume (at least for Romance) that pronominal clitics are affixes on verbs. The analysis we propose (which rules out clitic doubling) goes like this: A verbal lexeme only specifies its ARG-ST, and does not constrain the synsem type of the ARG-ST members, which can either be canonical, affix (for clitics) or gap (for extraction). The verbal forms inherit information from a lexeme type, but also from an inflectional type and from what we call a realization type. We distinguish for French a plain-word realization type (no clitics) and a clitic-word realization type (+ means list concatenation and o means shuffle): pl-wd : ARG-ST = SUBJ + COMPS o list (gap) cl-wd : ARG-ST = SUBJ + COMPS + nelist (aff) o list(gap) A plain verb realization has all its arguments in SUBJ and COMPS except for gaps; a clitic-verb realization has its arguments in SUBJ and COMPS except for (at least one) affix and possibly some gaps. For a language without clitic doubling, the morphological realization of these verbs cares about this realization type: the function we call F-praf applies only to cl-wd and attaches the right pronominal affixes (in the desired order) to the verbal root. That is how we get the Standard French distribution: Paul la-voit (Paul sees her) = cl-wd Il-voit Paul (He sees P) = cl-wd Il-la-voit (he sees her) = cl-wd Paul voit Marie (Paul sees Marie) = pl-wd * Paul il-voit Marie * Paul la-voit Marie * Il-la voit Marie * Paul il-la-voit The starred sentences are out because in order to get an affix realized in the morphology, the corresponding argument must be in the ARG-ST but not in the value of any valence feature. So you cannot have a corresponding argument syntactically realized in addition to the affix. For a (hypothetical) language with obligatory clitic doubling (of all arguments), we simply say that the morphology attaches a pronominal affix on the verb for any argument in the ARG-ST (whether syntactically realized or not) and you don't need the special synsem type affix. So the morphological function F-praf applies to all verb types. (it is an empirical question whether you also want to have an affix attached or not in the case of a gap argument; both options can be accommodated.) In this language you still have the two types of realization we defined above, because you also want to allow only clitics on the verb. So now you have the reverse distribution: * Paul la-voit * Il-voit Paul * Paul voit Marie * Paul il-voit Marie * Paul la voit Marie Il-la-voit (he sees her) = cl-wd Il-la-voit Marie (he sees Marie) = cl-wd Paul il-la-voit (Paul sees her) = cl-wd Paul il-la-voit Marie (Paul sees Marie) = pl-wd VARIANT1: In a language where you cannot have only clitics on the verb, but where you need a syntactic pronoun in addition to each clitic, we say that you only have the pl-wd realization type. This gives you something like: * il-la-voit Lui il-la-voit elle etc VARIANT2: Now the Romance languages we know seem to require clitic doubling with certain arguments but not others (subject clitics in Quebecois (cf Auger), dative human clitics in Spanish, accusative clitic pe in Romanian ...). For these, we say that we keep our realization types and also the synsem type affix. The F-praf function applies to cl-wd and pl-wd, but it only checks the affix type of the argument (in the verb's ARG-ST list) for the non doubling arguments; if the argument is not typed affix in the verb ARG-ST list it does not attach the corresponding clitic on the verb (it does nothing). For the doubling arguments, F-praf either does not check the synsem type or it checks to make sure that it is not a gap. In all cases, it is a minor change in the morphological function that produces the desired results. VARIANT 3: For cases of optional clitic doubling, F-praf again applies to both cl-wd and pl-wd, but can (optionally) do nothing (have an empty realization slot) when the corresponding argument is typed canonical (or non-affix) on the verb's ARG-ST list. The distribution will be the following: Paul la-voit = cl-wd Il-voit Paul = cl-wd Paul voit Marie = pl-wd Paul il-voit Marie = pl-wd Paul la voit Marie = pl-wd Il-la-voit = cl-wd Il-la-voit Marie = cl-wd Paul il-la-voit = cl-wd Paul il-la-voit Marie = pl-wd We have not addressed the question of clitic doubling for languages with postlexical clitics (clitics which are not affixes). 2. Left dislocation (LD) We analyze left dislocation as a type of extraction with a resumptive pronoun. The idea is to have (personal) pronouns bear an optional SLASH value that can be passed up and unified with a filler. In French the resumptive pronoun does not have to be a clitic; and the LD construction does not obey island constraints (which are obeyed by other types of extraction): Jean, je le connais. (Jean I know him) Jean, je ne connais que lui. (Jean I know only him) Jean, je connais quelqu'un qui l'a vu (Jean, I know someone who saw him) We distingush between two LOC type values: prl (pronominal) and nprl (non-pronominal: for both canonical and gap synsem), we say that islands are those constructions constrained to be [SLASH list(prl)] (that means either SLASH empty or SLASH {prl}). Pronouns and affix synsems have a optional SLASH value of type prl: lui: [SLASH {([prl,noun,acc,3sgm])}] We define a type of clause for LD constructions that checks to ensure that the SLASH of the Head Daughter has a member of type prl, and identifies the INDEX value of the Filler-Daughter with the INDEX of that member. For French you do not want the filler and the resumptive pronoun to share more because you may have case mismatches: Jean, je pense a lui Jean I think about him *A Jean, je pense a lui. For wh-constructions, the construction type checks that the SLASH of the Head Daughter has a member of type nprl and that it unifies with the Filler-Daughter's LOC value.
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:36:08 PST