<?xml version="1.0" standalone="yes"?>
<Paper uid="C96-1032">
  <Title>Syntactic Analyses for Parallel Grammars: Auxiliaries and Genitive NPs</Title>
  <Section position="3" start_page="0" end_page="182" type="metho">
    <SectionTitle>
2 The Formalism
</SectionTitle>
    <Paragraph position="0"> The architecture of LFG assumed here is the &amp;quot;traditional&amp;quot; architecture described in Bresnan (1.982), as well as the newer advances within LFG (Dalrynq)te et al. , 1995). A grammar is viewed as a set of correspondences expressed in ternls of projections fl'om one level of representation to nilother. Two fundamental levels of representations within LFG are the c(onstitutent)-structure and the f(unctional)-structure. The c-structure encodes idiosyncratic phrase structural prope.rties of a given language, while the f-structure provides  son (1991), Kaplan mtd Wedekind (1993), Butt (1994) for fllrther work on MT withii, LFG.</Paragraph>
    <Paragraph position="1">  a language mfiversal rel)resenl;adon of Gramm;rl,i(:a\] funcl;ions (e.g., Sllll3(~(:\[;, OiLI(~(:(;), COml)h,m(:nl;a.i;ion, l,(:nsc, binding, ('.(x:. The (:orrcsl)ond(:nc(: |)(:l;w(:Oll (&gt;,'-;trl1(;(;IIre ;l.Ii( |f-sl;rtlctur(! is llOt; oIll;o or O/l(&gt;(;o-ollc~ blll; in~_t/lyq;o-()ii(:~ ~-tllowillg 0Al 'tl, bs|;r~t(:(;ion ow:r idiosyncra.(;ic c-structure proper(Jes of a la.nguage ((:.g., discon(;inuous cons(;ituenl:s).</Paragraph>
    <Paragraph position="2"> In addil;ion, severM proi)osals ext)loring t)ossihie rO, l)r(~s(;lli;;~l;ioll,q of &amp; s(0Ill&amp;Ill;ic)-sl;rllt:l;llr(! h;LVe l)ce.n ma.dc ovt'.r tim y(;ars (e.g. \[lalvo)'s(m and t(alJan (1988), l)a,h',ymplc el; al. (1993)). As (;he r(:-Miz~l;ion of a SCl)a)~(;(: s(:m;m(;ic compon(:n(; is only l)lmmed fl)r (;h(,. lal;l;(:r stages wil;hin I'AIIGIIAM, llo \['llrtht;r discussion of possible f()vma, limns will l;ake l)la.t'.e ht'.rc, t(; should b(, noi,ed, how(;ver, l;ha.l, rudim(:nLary s(:mm~ti(: inform;t(;ion, su('.h a.s a.rtgmt(~nI; S(;I'llCI,llI'O illf()I'lllil,\[;iOll (\]exical StHIIDAII,iCS), iS (;Ii.(:oded within l;he t'-s(xuc(;m'e,~ in ord(:r Ix) fa.cilil;ate la'a,nsfl~r in s()m(,, cases. A case in t&gt;oin(, is t)res(m(,(~d in l;he s(:c(;ion on (\](!rman g(;ni(;ive Nl's.</Paragraph>
  </Section>
  <Section position="4" start_page="182" end_page="185" type="metho">
    <SectionTitle>
3 Auxiliaries- -a flat approach
3.11 The Received Wisdom
</SectionTitle>
    <Paragraph position="0"> Auxiliaries have given li,qe. 1,&lt;) lively debai;es con(:(w/ring (,heir (:xa, t:(, synl:a,c(:ic si,al;us (e.g. (\]h()m-sky (1957), \]h&gt;ss (1967), Pulhtn, and Wilson (1.()77), Al(m;kjia, ii el, al. (1979), Gazda, (!1; M. (1!)82)): are (;hey simply main verbs wit,h Sl)ecial l)rOlmrl,ies , or should Lhey insl,and;1ol:e a stmcia\] c;1o(;egory AUX'? WiLhin cmren(; lexica.1 n.i)proacht~s (\[a xl( M--\[ lllI( tlOtl;tl-(, l ;!l'~lIll I,l (1,1 G), Hea.d-driven I)hrase S\[;ruc, tm'(', (~laml m,~ (IIPSG)), auxiliaries (e.g. h, rzve, be',) and modal,q (e.g. musl,, ,sh, o',,hl) arc (;rt:a, Led ;ts r,,i,sin9 verbs, which are marked as special iu some way: in tlI'SG through ;m \[AtrX: -I\] feal,m'e (l'ollard and Sag 1994), in IAVG (llresmm 1.C/)82) by a ditl'erence ill Pill'\]l) Va,ltl(L 2 Ih)wew~r, newt, work wil;hin I,f,'(I (13resnan 1995, T.II. King 1995) has been moving ;~xvay from (;he r;fising ;~l)lntm,ch I,ow;uds ;m ;ma, l 3' sis where auxiliaries are O,\](!Ill(;lll;,q which conl;rilm(;c 1,o (;hi' dause only I,(ms(',/aSl)ecl, , ;~greenmn(,, or voice informal;ion, 1)ul; no(; a sul)t:;Lt;egorizat;ion f'rmno. This view is also in line wil,h at)proaches wil;hin GB (Govermneni-Binding), which s(:e mtxilim'ies simply as possit)le insl;;m(,iat;ions of t, he l'uncl;ionat (:~(;(;gory i (st;(: ;tlsC/) Iia,lh; a,nd Mmmll,z (1 :)!):3)).</Paragraph>
    <Paragraph position="1"> The &amp;quot;tradit, ional&amp;quot; tle;ttmenl, ot' ;mxilim'ies in 1)o(;h Ill S(, (l)ollm'(t and Sag 1994) a ttd I,F(', has il;s roo|,s in l{oss's (1967) tlrol)osa.l 1,o I;re;~l; mtxilim'ies and moda,ls on a, par wil;\]l main v(!rlm. :~ US(!(; Falk (1984) for im ea, rly I,FCI l,l(:tloI;lil(;Ii(; of %; in line wil;h l;h;1o(; l)lOpos(:d herr:, ;md Abeilh',' and Godm'd (71994) for a similm' I;rt!.qPSint!nt in l~}(:nt:lt.</Paragraph>
    <Paragraph position="2"> 3t\]'ll(~ (;(~l'lll awxiliary has ofl,t:n been tAtk(!ll to Slll)sume bol;h modMs ~md (;\](?lll(~ll\[;S Sll(;h ;is have ;tn(l be. lh)wev(w, (;he disl;incl;ion l)(~l;wt:(:n (.he Lwo is nt!t:t!ssmy not only semanl;i(:ally, bul, Mso synl;a(:t;i(',ally. In ('er-IIt I)m'l;icular, mlxiliaries are (a'ea.t;od as a sul)class of raising vcrl)s (e.g. I%ll;u'd ;rod Sag (t994), Falk (1984)). l)br exmnph',, a simple seni;ence like, (\]) would correspond l;o (,he c-structure and f sl;rucl;ur(; shown in (2) ;rod (3), respecdvt',ly. Note (;hat the level of elnlmdtting in the f-structure exacl;ly mirrors t, he c-st, ructure: each verbM (:|cnwnt takes a c(/mpl(:m(:n(;.</Paragraph>
    <Paragraph position="3">  The triain reasons (,o (;r(:a(; a.uxilim'i(:s as (:ompl(&gt; lllCll(; taking verbs in English axe: 1) ;m m:(;oun(; of VP-(:llipsis, VP-l;ol)i(:aliz~i;ion, etc. follows imm(:diaAx:ly; 2) l'C,';l;l'iCl;i()IlS ()ll (,lit: lDl,{;l/l'(~ Of l;hc VClhal COml)h;m(:nL (t)rogressive , t)as(; t)ard(:iph!, e((:.) following the auxiliary c;m l)e sl;al;(:d six;right;for wardly (l&gt;ullum mM Wilson (1977), Akmajian (:I; hohls for (\]Cl'IH~-LII }1o,'4 well, ;rod in fact,, wil;hout man and (some di;1olect;s of) Fatglish modals (:;1oil lm sl;a.cked, while (;he disl;ribul, ion of mtxilim'it'.s is more resi;ricl;ed. Also, assuming thai; scream;it interpret;tdon is driven l)rinmrily off' of l;he fsl;ru(:lmr(:, l;he relative embedding of modals mnsI; be preserved ~1oI; that; level in order to allow an inl;erl)rei:a(Aon of their scope a, tld S(:lll&amp;lltic fol'c(!,  some sort of a hierar(:hieal stru(:ture, stating well-formedness conditions on a string of multiple auxiliaries become.s wellnigh impossible, in light of the greater ordering possibilities gra.nted hy the flexible German word or(ler. There are a,lso major reasons, however, tbr nol; adopting this analysis: 1) linguistic adequacy; 2) unmotivated structural complexity; 3) lion-parallel analyses for predicationally equivalent sentences. Consider the French e{luivalent of (\]) in (4).</Paragraph>
    <Paragraph position="4"> (4) Le {:ondu(:teur aura tourn6 le levier the driver will have turned the lever 'The driver will have turned the lever.' As argued by Akmajian el; al. (1979), erosslinguistic evidence indicates that elements bearing only tense, mood, or voice should helong to a distinct syntacti(: category. In many languages, like Fr(;nch or Japalmse, the infornmtioll (:arried by 'will (future), or have (perfect) is realized mort)hologically rather than i)eriphrastically. The analysis in (4) thus effcetiw?ly claims that there exists a dee t) difference in the pr{;{lieational structure of auxiliaries like will and have and the l~ench aura. 4 This is not {lesirable flom ~L crosslinguisti{: point, of view, nor is it, helpful f{}r MT.</Paragraph>
    <Section position="1" start_page="183" end_page="185" type="sub_section">
      <SectionTitle>
3.2 Alternative Implementation
</SectionTitle>
      <Paragraph position="0"> The approach adopted here is a fiat analysis of auxiliaries at f-structure ((5)).</Paragraph>
      <Paragraph position="1">  The auxiliaries wird 'will' and haben 'have' now only contribute information as to tit(', overall tense, lint do not subcategorize for complements. Structural phenomena like VP-ellipsis, coordination, or topicalization can, however, still be accounted for ill terms of an apl)roi)riate emt)e, hling at c-structure (cf. (2)). The role of a.uxiliaries in natm'al language is now adequately modeled, ill partieular with respect to a more realistic treatment of tense (compare (3) and (5)), as the 1,Y=ench (4) has essentially the same f-structure as (5)) 4Note that wird 'will' is often analyzed as a modal in accordance with Vater (1975). However, the arguinents i}resented there are not conclusive.</Paragraph>
      <Paragraph position="2"> '~The construction of the wflue for the cmnposed tenses results fl'om a complex interaction between the Ilowever, the fiat f-structure in (5) providc's no room for a statement of selectional requirements, allowing massive overgeneration (e.g. nothing blocks the presence of two haben ill (1)). Ncither (:all the particular order of auxilia.ries be regulated. Our solution takes advantage of LFG's flexiMe projection-based architecture by implementing a projection which models the hierarchical seleetional requirements of auxiliaries, yet does not interfere with the sul)categorizatiomfl prot)erties of verbs, as wouhl he the ease under a raising anal-</Paragraph>
      <Paragraph position="4"> gedreht haben in LFG, the flexihle word order of German is handled via flmctional wnccrtainty, which characterizes long-distance dependeneies without resorting to movement analyses (Netter (1988), Zaenen and Kaplan (1995)). As in (6), which illustrates our alternatiw; solution, functional uneertainty is represented by the Kleene Star (.).6 Tile annotation on the NPs indicates Lhat they could fulfill the role of any possible grammatical flmetion (GF), e.g. SUllJ or oB.I, and that the level of embedding ranges from zero to infinite. With e.very auxiliary subcategorizing for an XCOMP, the two NPs could conceivably be arguments of three different verbs: wird, haben, or gedreht. Titus, the greater structural complexity unnecessarily increases the.</Paragraph>
      <Paragraph position="5"> search space for the deternfination of a verb's arguments. In (6), however, the m-structure is projected from the e-structure parallel to the f-structure through annotations similar to the usual f-structure annotations. 7 Statements about &amp;quot;morle~xical entries. Note that this treatment does not as yet include a fine-grained represention of tense and aspect. This is the subject of ongoing work. The treatment presented here provides th('~ basis needed for a thorougt, erosslinguistic analysis of temi)oral and aspectual phelmmena.</Paragraph>
      <Paragraph position="6"> 6For space reasons, the xc indicates XCOMP, the D a DEP.</Paragraph>
      <Paragraph position="7"> rThe annotation p M* in (6) refers to the m-structure associated with the parent c-structure node, and t t* refers to the m-structure associated with the daughter node. The more familiar t and $ of I,FG are simply shorthand notations of the same idea,</Paragraph>
      <Paragraph position="9"> tn'cvia, ted) lexil:al /!nl;ry (~Xeml)lifi(~s l:lm pi(~l:cs o| in\['(/rma, l;il)n ne('xh~d. The (lisjm~cl;iv(! h~xi(:a,l ent;ry for wi';Y=l 'will' iu (8) l;akl!s 1,he various (:(nnbinalx)ry l)ossit)ilil;ies of auxilia.ries and maiu verl)s illtO ~L(:(',()lllll,, tLllll \])r(ivi(|(~,s 1;h(', ;q)propria,t;c t;Imse timt, m'e. |,'or i',xamph~, it; r(~lluires l;ha.1; l;he (unt)elld0d VV()IIM \])e BASI';, ,1/.i1(1 (;ha.t I;}l(!l'(t 1)o lto lmssive involved for a, simple ful:ur(~ likl,' wi'rd dvclu',n.</Paragraph>
      <Paragraph position="11"> iTui, l'(,slri,:ted 1;~, l,lm l,r,,i,,ct,i(m \['1'()11ll c stl'll(ttlll'(! 1:() f-sia'uci;ure: I&amp;quot; &lt;t, m*,-1--- &lt;lJ * ~&amp;quot;l';lt;lll'(~s ll('.()(\](,~(l ()lily to (ulSUl'(~ la/lgllaP~e \[)al ti(:ulm wl;llformedu(:s,~ are no h)nger unified inl,o t;lm \[~struclmrl% clut, t(wing ;~ r(q)l'eS(:nl;~don l:ha.t, is m(!ant, tx) be language ind(;llendenl;. In our ana.lysis, only t'(!al;m'(!s nl!(;th;(l for fl~rtlmr stmmnl;ic iu.</Paragraph>
      <Paragraph position="12"> Ix!rl)r(!t, al;i/)n, MT, (n' for the exl)ressh)n of \]all gua.ge univ(~rsa.l synl;;ml.il: gen(,,ra\]iza.l;i()ns are reI)rt!sonl,(~d ~l,I; ft4rll(:tllr(&lt; l,'()r (;Xaml)h~ , morpholog/i(:ally (m(:{/(bd inf'ormalJ(m like case, g(m(|(w, //r al,~r(~(mmnt, is n(~(~ded f(n sl,at(tln(',nt,'-; as t() binding, l)rl',dica.l:t&gt;argmn(ml, relal;i()~ts, or 1;lm det;iwu6- null (l)rew~rlm\] NPs, posl:verlm.1 PPs or NPs), poses a lmrl.imJar chall(m~,~, for a. I)a.ralh~l grammar (h&gt; v(%t)m(mt; l)roj(~(:t, like I'AI((IIIAM. \]11 I:his ira. 1)(!r, w(~ Slll(gesl, a. l,r(!al;l\[l(~nl; ()f muli,iph~ g(midve. NPs whi(:h nol. only a(:c()unt.s for s()me r(;st.ricd()ns on t.lmir (lisl;rilmlion wilJml (Mrman, tml. a.ls()allows a languag(~ iml(!imlMe.t. (mfiversal) r(q)res(m1,alion, thus f;t(:ilil,al;illg M~I '.</Paragraph>
      <Paragraph position="13"> In /,,etmral, I.he distribution of mull:iple NPs wil.hin NPs is an ar(m ol(',e, rman synl:ax whi(:h has IlOI, l'O,C(~iV(~d a sa,l,ist'4cl;()ry &amp;ccolllll, I;o (I;tt;e ((!.g., I'()llztr(l and Sag (t994), llha,~;l; (1990), llai(hw (I !)88)). \]In (;(~1'I11~'111, 11()11IIS getmrally have at most ()n\[~ g(mil;ive which ma.y o{:cu, r in ~ l)r(;l~ominal in i)ostnominal posil;ion adja(:enl; lx) t, lm noun. \]h)l;h kin(is of ge, nidv/~s ha,v(~ the sam(', morlJl(llogi('al slmpe. \[h-)wev(u', I,nninalizat;i(ms thai; a.re ihwiv(~d t'r(Im a l:ransidve verb all()w f(/r l;wo g/'mdv(',s, (/he ill 1;he l)renonfina\], \[h(! ()l:her in l;h(! l)ost.noulinal t)()sidlm.</Paragraph>
      <Paragraph position="14"> Tlm flml:l;iou of a g(!nil;iv(! is g/!n/u'a, lly (!xln(!ssed as in(li('al;in~,~ a Ii(/ssl'~s(ir: p(),~s witJfin Lt,'(;. I\[()w(wer, in dm (:as(~ (if' two gl~nil;ivl',s, l:h(! assignml,.nl ()\[' l,wo poss va, hms vi(/la, tl;s l;tm uniqu(m(;ss-(:tlnllil,i(/n (m f-sl,ru(:l, url~s and is fut'l:h(wm(tr(; insuHi(',ilml; lxl (lim;i,guish t:h(', t,w(I ditiiwing kinds of g(midves. W(~ dmret'ore llrotlosl ,, lJm ul;ilizal;ion (/l' t:w(i flmcl;ions llallle( |(;I,;N\] a, ild (II,;N{2 ill orlhw txl avoid asso(:i:&gt; lJl/n with any st)(~l:ili(' s(~nm,nl;i(: rob. |?ml;h(wmor(;, as g(',nidve, s iu I;h(~ NI ) arc g(Umlally ()l)t;ilma\], t,h(;y are l,aMm 1,o ('xt)ress n() ,t~ov(wned funct3()ns, i.e., l:hey m(~ n(/l, sut)(',al,iWiriz(~d for lly l,tm n(mn. So (;I';N\] ;/lid GI,;N2 ill'(! S(tllla, Ilti(; f'un(:t,i()ns ill \[,t5'(~ on a l)ar with, say, a(ljun(:l,s, q\]m NI' rule for German  then is (9). 8 (9) NP ({DET: q'=;</Paragraph>
      <Paragraph position="16"> If the head-noun is not derived from, say, a verb, the single genitive ill either position is interpreted as a possessor. In case of a derived nominal, however, a genitive is interpreted according to tile thematic roles assigned to tile arguments of the verbal base. That means the functions GFN1 and sc gen2 have to be linked to the approt)riate roles. Neither of tile two functions is in principle restricted to any specific roh;. But if both genitives are present they must be interpreted according to a thematic role hierarchy.</Paragraph>
      <Paragraph position="17"> As (10) shows, if only one genitive is present, its prenominal interpretation may be as agent or as patient. A postnominal (single) genitive is interpreted as agent if the head noun is derived fronl an intransitive, and as a patient/theme if derived from a transitive.</Paragraph>
      <Paragraph position="18">  However, if two genitives occur, as in (11), the prenominal genitive is restricted to an agent, and the postnominal one to patient. This restriction must be encoded at some lew;1, but does not follow from tile distiction between GEN\] and (IEN2, wtfich are flmctions that do not bear any semantic content on the.ir own.</Paragraph>
      <Paragraph position="19">  To obtain the correct linking, tile argument structure of the verbal base must be available.</Paragraph>
      <Paragraph position="20"> Since MT is based on f-structu,'es within PAtt,(IRAM~ the argument structure has to be present at; this level of representation. '9 Nominalization is therefore implemented as a morphologically driven process (lexical rule) which eliminates suI~a and Ol&lt;J fl'om the verb's subcategorization frame and enters tile verb's argument structure into tile 8Abstracting away fl'om bar-level considerations and further ot)tional constituents, this ruh; captures the restrictions that determine the dislocation of a genitive in the matrix Nit 9If a semantic or argulnelll; projection is assumed (e.g., Halvorsen and Kaplan, 1988), this information may be represented at another independent projection.</Paragraph>
      <Paragraph position="21"> lexical entry of the noun. This yields the optionality of genitives while preserving tile underlying semantics, as shown in (12). The association of CEN1 and GEN2 then is determined according to a hierarchical order of arguments (Bresnan, 1995). This approach also provides a means of handling certain eases of categorial stfift. For instance, in German temporal and conditional adjuncts mw be realized as PPs dominating an NP headed by a deverbal noun. English does not have this option, but employs an adjunct-clause instead. Ih;re, the GEN1 and GFN2 functions of the Germml fstructure have to be related correctly to tile SUlIJ and OBJ functions of tile English f-structure.</Paragraph>
      <Paragraph position="22">  'when Karl reported the accident everyone had to laugh' Here the linking of the GEN1 and (;EN2 functions to the appropriate thematic rote in the German f-structure driw;s tile transfer of these functions to the SllllJ and oBJ time, lions of tile English f-structure.</Paragraph>
      <Paragraph position="23">  Under this approach, languages now only differ with respct to the categorial realisation of the flmction by ease-marked NP or PP. Thus, an application of this treatnmnt not only provides an adequate grammatical analysis of the NP in Ger,nan, but also facilil, ates MT.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>