| { | |
| "paper_id": "C98-1014", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T12:30:46.712767Z" | |
| }, | |
| "title": "Processing Unknown Words in HPSG", | |
| "authors": [ | |
| { | |
| "first": "Petra", | |
| "middle": [], | |
| "last": "Barg", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Seminar ftir Allgemeine Sprachwissenschaft Heinrich~Heine-Universit~it Dfisseldorf Universit", | |
| "location": { | |
| "addrLine": ":dtsstr. 1", | |
| "postCode": "D-40225", | |
| "settlement": "Dtisseldorf", | |
| "country": "Germany" | |
| } | |
| }, | |
| "email": "barg@de" | |
| }, | |
| { | |
| "first": "Markus", | |
| "middle": [], | |
| "last": "Walther", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Seminar ftir Allgemeine Sprachwissenschaft Heinrich~Heine-Universit~it Dfisseldorf Universit", | |
| "location": { | |
| "addrLine": ":dtsstr. 1", | |
| "postCode": "D-40225", | |
| "settlement": "Dtisseldorf", | |
| "country": "Germany" | |
| } | |
| }, | |
| "email": "[email protected]@de" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "The lexical acquisition system presented in this paper increlnenlally updates linguistic properties of unknown words inferred lrom their surrounding con-lexI by parsing senlences wilh an liPSG gramnmr for Gerlmm. We employ a gradual, infornmtionbased concept of \"tlnknowntless\" providing a uniform treatment for the range oi completely known to maximally unknown lexical entries. \"Unknown\" information is viewed as revisable inR)rmation, which is either generalizable or specializable. Updating takes place after parsing, which only requires a modified lexical lookup. Revisable pieces of informalion are idenlilied by grammar-specilied declaralions which provide access pmhs into lhe parse feature slructure. The updating nlechanism revises the col responding places in the lexical lcalure stmclures iff the conlext actually provides new information. For revising generalizable inlbrmalioi< (ype union is required. A worked-out example demonstrates the in-Iercntial capacily of our implemented system. *This work was tan'ted out within the ,S'onde@)rschungsbereich 282 '77worie des Lexikol~s' (project B3), funded by the German Federal Research Agency DFG. We thank James Kilbury and lnembers of the B3 group for fruitful discussion.", | |
| "pdf_parse": { | |
| "paper_id": "C98-1014", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The lexical acquisition system presented in this paper increlnenlally updates linguistic properties of unknown words inferred lrom their surrounding con-lexI by parsing senlences wilh an liPSG gramnmr for Gerlmm. We employ a gradual, infornmtionbased concept of \"tlnknowntless\" providing a uniform treatment for the range oi completely known to maximally unknown lexical entries. \"Unknown\" information is viewed as revisable inR)rmation, which is either generalizable or specializable. Updating takes place after parsing, which only requires a modified lexical lookup. Revisable pieces of informalion are idenlilied by grammar-specilied declaralions which provide access pmhs into lhe parse feature slructure. The updating nlechanism revises the col responding places in the lexical lcalure stmclures iff the conlext actually provides new information. For revising generalizable inlbrmalioi< (ype union is required. A worked-out example demonstrates the in-Iercntial capacily of our implemented system. *This work was tan'ted out within the ,S'onde@)rschungsbereich 282 '77worie des Lexikol~s' (project B3), funded by the German Federal Research Agency DFG. We thank James Kilbury and lnembers of the B3 group for fruitful discussion.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "It is a ,emarkable fact thai humans can often understand sentences containing unknown words, inlet their grammatical properties and incrementally refine hypotheses aboul these words when encountering laler instances. In conlrasl, many current NLP systems still prestlppose a complele lexicon. Notable exceptions include Zernik (1989) , Erbach (199()) , Hastings & Lylinen (1994) . See Zernik lot an introduction to the general issues involved.", | |
| "cite_spans": [ | |
| { | |
| "start": 320, | |
| "end": 333, | |
| "text": "Zernik (1989)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 336, | |
| "end": 350, | |
| "text": "Erbach (199())", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 353, | |
| "end": 378, | |
| "text": "Hastings & Lylinen (1994)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "11fis paper describes an HPSG-based system which can incrementally learn and reline properlies of unknown words after parsing individual sen-tences. It focusses oll extracting linguislic properties, as compared to e.g. general concept learning (Hahn, Klenner & Schnatiinger 1996) . Unlike Erbach (1990) , however, it is not conlined to siulpie morpho-synlactic information but can also han-(lie selectional reslriciions, senianlic types and argument slructure. Finally, while statistical approaches like Brenl (1991) can gather e.g. valence inloruialion lioni large corpora, we rote more interested in full gralmnatical processing of individual sentences to maximally exploil each context.", | |
| "cite_spans": [ | |
| { | |
| "start": 244, | |
| "end": 279, | |
| "text": "(Hahn, Klenner & Schnatiinger 1996)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 289, | |
| "end": 302, | |
| "text": "Erbach (1990)", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 504, | |
| "end": 516, | |
| "text": "Brenl (1991)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The following three goals serve to struclure our model. It should i) incorporate a gradual, informalion-based conceptualization of \"unknownness\". Words are not unknown as a whole, but may contain ux~k~mw~t, i.e. revisable pieces of b(for-,la*iom Consequently, even known words can under(,o> revision 1o e.>.<) acquire new senses. This view replaces the binm-y distinclion belween open and closed class wordx. 1l should it) maxinmlly exph)il lhe rich represeniaiions and nlodelling convenlions of IiPSG and associaled formalisnls, wiih esseniially the same grammar anti lexicon as compared (o closed-lexicon approaches. This is important bolh lo facilitate reuse of existing grammars and li) enable meaningful feedback for linguistic theorizing. Finally, it should iii) possess donlain-independenl inference and lexicon-updating capabililies. The grammar writer must be able to fully declae which pieces of inlbrmalion are open to revision. The system was implemenled using MicroCUF, a simplilied version of the CUF typed unification lormalisin (D6rre & Dorna 1993) thai we implemented in SICStus Prolog. It shares both the feature logic and the definite clause extensions with ils big brother, but substitutes a closed-world type system li)r CUF's open-world regime. A fealure of our lype system impletnenlation lhaI will be signilicant later on is (hat type infornmtion in internal IEalure sittic-tures (FSs) can be easily updated.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The HPSG grammar developed with MicroCUF models a fragment of German. Since our focus is on the lexicon, the range of syntactic variation treated is currently limited to simplex sentences with canonical word order. We have incorporated some recent developments of HPSG, esp. the revisions of Pollard & Sag (1994, ch. 9) , Manning & Sag (1995) 's proposal for an independent level of argument structure and Bouma (1997) 's use of argument structure to eliminate procedural lexical rules in *:avour of relational constraints. Our elaborate ontology of semantic types -useful for non-trivial acquisition of selectional restrictions and nominal sorts -was derived from a systematic corpus study of a biological domain (Knodel 1980, 154-188) . The grammar also covers all valence classes encountered in the corpus. As for the lexicon format, we currently list full forms only. Clearly, a morphology component would supply more contextual information from known affixes but would still require the processing of unknown stems.", | |
| "cite_spans": [ | |
| { | |
| "start": 292, | |
| "end": 319, | |
| "text": "Pollard & Sag (1994, ch. 9)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 322, | |
| "end": 342, | |
| "text": "Manning & Sag (1995)", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 406, | |
| "end": 418, | |
| "text": "Bouma (1997)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 714, | |
| "end": 736, | |
| "text": "(Knodel 1980, 154-188)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "When compared to a previous instance, a new sentential context can supply either identical, more special, more general, or even conllicting inIormation along a given dimension. Example pairs illustrating the latter three relationships are given under (1)-(3) (words assumed to be unknown in bold face).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Incremental Lexical Acquisition", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "( ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Incremental Lexical Acquisition", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "In contrast to (la), which provides the inff)rmation that the gender of Axon is not feminine (via ira), the context in (lb) is more specialized, assigning neuter gender (via das). Conversely, (2b) differs from (2a) in providing a more general selectional restriction for the subject of reagiert, since sense organs include ears as a subtype. Finally, the adjective sensibel is used predicatively in (3a), but attributively in (3b). The usage types must be formally disjoint, because some German adjectives allow for just one usage (ehemalig 'former, attr.', schuld 'guilty, pred.').", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Incremental Lexical Acquisition", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "On the basis of contrasts like those in (1)-(3) it makes sense to statically assign revisable information to one of two classes, namely specializable or generalizable, l Apart from the specializable kinds 'semantic type of nouns' and 'gender', the inllectional class of nouns is another candidate (given a morphological component). Generalizable kinds of information include 'selectional restrictions of verbs and adjectives', 'predicative vs attributive usage of adjectives' as well as 'case and lbrm of PP arguments' and 'valence class of verbs'. Note that specializable and generalizable inlbrmation can cooccur in a given lexical entry. A particular kind of intbrmation may also figure in both classes, as e.g. semantic type of nouns and selectional restrictions of verbs are both drawn from the same semantic ontology. Yet the former must be invariantly specialized-independent of the order in which contexts are processed -, whereas selectional restrictions on NP complemenls should only become more general with further contexts.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Incremental Lexical Acquisition", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "We require all revisable or updateable information to be expressible as formal types. 2 As relational clauses can be defined to map types to FSs, this is not much of a restriction in practice. 1The different behaviour underlying this classification has previously been noted by e.g. Erbach (1990) and Hastings & Lytinen (1994) hut received either no implementational status o1' no systematic association with arbitrary kinds of information.", | |
| "cite_spans": [ | |
| { | |
| "start": 283, | |
| "end": 296, | |
| "text": "Erbach (1990)", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 301, | |
| "end": 326, | |
| "text": "Hastings & Lytinen (1994)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "2In HPSG types are sometimes also refmTed to as sorts. ~:eeneralizable inlormalion requires 0'Pe union (e.~,,. pred V (Lttr = prd). qqle latter might pose problems tot type systems requiring the explicit delinition el all possible unions, corresponding to least common supertypes. However, type union is easy lot (Micro)CUF and similm: systems which allow for arbitrary boolean combinations of types. Generalizable inlormation exhibits another peculiarity: we need a disjoint auxiliary type u_g Io colTectly mark the initial unknown inlormalion slate. 3 This is because 'content' types like prd, pred, atlr are to be interpreted as recording whal contextual information was encountered in the past. Thus, using any of lhese to prespecify the initial value--, either as the side-effect of a feature appropriateness declaration (e.g. prd) or through gramlnar-conlrolled specification (e.g. pred, attr) --would be wrong (of. prdi,,ti, l V al:l,r == prd, but tt-.(,lir~iti,l V aLl, r\" :: lt_.q V a,l,l,r) .", | |
| "cite_spans": [ | |
| { | |
| "start": 917, | |
| "end": 1000, | |
| "text": "(of. prdi,,ti, l V al:l,r == prd, but tt-.(,lir~iti,l V aLl, r\" :: lt_.q V a,l,l,r)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Generalizable inlormalion evokes another queslion: can we simply have types like those in fig. 1 within HPSG signs and do in-place type union, just like type unification? The answer is no, li)r essentially two reasons. First, we slill want to rule out ungran]malicaI constructions through (type) unificalion failure of coindexed wdues, so lhat generalizable types cannot always be combined by nonfailing type union (e.g. *tier sensible Geruch 'the sensitive smell' must be ruled out via ,~c'n:~,:_org.';~ A amcll --Z.).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 90, | |
| "end": 96, | |
| "text": "fig. 1", | |
| "ref_id": "FIGREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "We would ideally like to order all type tmilicalions pertaining Io a value before all unions, but this violates the order independence of constraint solving. Secondly, we already know that a given inlbrmational token can ,vimulta~wously be generalizable and specializable, e.g. by being coindexed through HPSG's valence principle, tlowever, silnullaneous in-place union and unilication is conlradiclory.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "To avoid these problems and keep the declarative monotonic setting, we employ two independent features gen and ctxt. ctxt is the repository of contextually unified inlbrmation, where conllicts result in ungrammalicality, gen holds generalizable information. Since all gen values contain u_g as a type disjunct, they are always unifiable and lhus not restriclive during the parse. To nevertheless get correct gen values we perform type union after parsing, i.e. during lexicon update. We will see below how this works out.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "3Actually, the situatkm is more symmetrical, as we need a dual type uor t(} con'ectly mark \"unknown\" sT)ecializable infl)rmarion. This In'events incon'cct updating i~t known inforlmition. ltowever, uov is tlnllccessaly li~i' the examples presented below. Step 1 projects the parse FS derived from the whole sentence onto all parlicipaling word lokens. This resulls in word FSs which are conlexlually enriched (as compared to their original lexicon state) and disambiguated (choosing the compatible disjuncl per parse solution if the entry was disjunctive). It then filters the set of word FSs by unification with the right-hand side of revisability chmses like in (4). The oulput of step 1 is a list of update candidates for those words which were unifiable.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Slop 2 determines concrete update values lor each word: for each matching generalizable chmse we lake the type tmion oflhe gen value of the old, lexical state of the word (Lca:Cc~ 0 with lhe ctxt value of its parse projection (Ctzt): 5l'17 = Lcz(;eTzuCt:c~. For each matching specializabIe(Spec) chmse we take the parse value Spec.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Step 3 checks whether updating would make a difference w.r.t, the original lexical entry of each word. The condition to bc met by generalizable information is that 7'U D L :xGeu, lot specializal~le inlormation we similaly require Spec C l, exSpcc.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "In step 4 the lexical entries of words surviving slep 3 are actually modilied. We retract the old lexical entry, revise the entry and re-asserl it. For words never encountered before, revision lnUSt obviously be preceded by making a copy of the generic unknown entry, but wilh the new word's phonology. Revision itself is the destructive modilicalion of type informa-tion according to the values determined in step 2, at the places in a word FS pointed to by the revisability clauses. Tiffs is easy in MicroCUE as types are implemented via the attributed variable mechanism of SICStus Prolog, wlffch allows us to substitute the type in-place. In comparison, general updating of Prolog-encoded FSs would typically require the traversal of large structures and be dangerous if structure-sharing between substituted and unaffected p',u-ts existed. Also note that we currently assume DNF-expanded entries, so that updates work on the contextually selected disjunct, qTtis can be motivated by the advantages of working with presolved structures at run-time, avoiding description-level operations and incremental grammar recompilation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Representation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "We will illustrate how incremental lexical revision works by going through the examples under (5)-(7). Afler parsing (5) the gender of the unknown noun Nase is instantiated to fern by agreement with the determiner die. As the special&able clause (4b) matches and the gend parse value differs from its lexical value gender, gender is updated to fern. Furthermore, the object's semantic type has percolated to the subject ]Vase. Since the object's sense~grgan type differs from generic initial nora_rein, Nase's ctxt value is updated as well. In place of the still nonexisting entry for perzipiert, we have displayed the relevant part of the generic unknown verb entry.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "A Worked-Out Example", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "Having parsed (6) the system then knows thai perzipiert can be used intransitively with a nominative subject referring to ears. Formally, an HPSG mapping principle was successful in mediating between surface subject and complement lists and the argument list. Argument list instantiations are themselves related to corresponding types by a further The first (subject) element on the args list itself is targeted by another revisability clause. Tiffs has the side-effect of further instantiating the underspecified lexical FS. Since selectional restrictions on nominal subjects must become more general with new contextual evidence, the union of ear and the old value u_g is indeed appropriate. Sentence (7) first of all provides more specific evidence about the semantic type of partially known ,Vase by way of attributive modification through verschnupfte. The system detects tiffs through the difference between lexical ctxt value sense_organ and the parse value nose, so that the entry is specialized accordingly. Since the subject's synsem value is coindexed with tim first args element, [ctxt ,,ose] simultaneously appears in the FS ofperzipiert. However, the revisability clause matching there is of class generalizable, so union takes place, yielding ear\" V nose = sense_organ (w.r.t. the simplified ontology of iig. 1 used in this paper). An analogous match with the second element of args identifies the necessary update to be the unioning-in of smell, the semantic type of Gestank. Finally, the system has learned that an accusative NP object can cooccur withperzipiert, so the argument structure type of gen receives another update through union with npnom_npacc.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "A Worked-Out Example", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "The inccemenlal lexical acquisition approach described above attains the goals stated earlier. It realizes a gradual, infornmtion-based conceptualiza~ lion of unknownness by providing updateable lbrmal types -classilied as either generalieable or special-<able -logelher with grammar-delined revisability clauses, it maximally exploits standard HPSG red resenlalions, requiring moderate rearrangements in grammars at best wtfile keeping with the standard assumptions of typed unification formalisms. One noteworlhy demand, however, is the need for a type union operation. Parsing is conventional modulo a modilied lexical lookup. The actual lexical revision is done in a domain-independent poslprocessing step guided by lhe revisability clauses.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Of course Ihere are areas requiring furlher consideration. In conlrasl to humans, who seem to leap to conclusions based on incomplete evidence, our approach employs a conservalive form of generalizalion, taking file disjunction of actually observed valties only. While this has the advantage of not leading lo overgeneralization, lhc requirement of having to encounter all subtypes in order lo infer their con> men superlype is not realistic (sparse-da|a problem).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "In (2) se~zse_orgatt as lhe senlanlic type of lhe tirsl argument olperzil)iert is only acquired because tile siulplilied hierarchy in lig. l has ~to.s'e and ear as its only subtypes. Here lhe work of Li & Abe (1995) who use the MDL principle to generalize over the slots of observed case frames nfight prove li-uitful.", | |
| "cite_spans": [ | |
| { | |
| "start": 200, | |
| "end": 215, | |
| "text": "Li & Abe (1995)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "An important question is how to administrate allernative parses and lheir update hypotlteses. In Dax Aktiot~Sl)Oterttial erreicht del~ l)ettdt'itelt 'the aclion polenlial reaches lhe dendrite(s)', l)endriten is' ' alnblguous between acc.sg, and dat.pl., giving rise to lwo wtlence bYt)olheses Itpzlomotpacc anti tqmomatpdat for errz'.ic:ht. Details relnain to be worked out on how to delay the choice between such alternative hypoll~eses until flmher contexts provide enough inforulation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Another topic concerns the treatment of 'cooccurrence reslriclions'. In lig. 2 llle system has indepettdetttly generalized over the selectional reslriclions for subject and object, yet there are clear cases where this overgenerales (e.g. *Das Ohr perzipiert de~ Ge.rtcmk 'the ear perceives the stench'). An idea worth exploring is to have a partial, extensible list of lype reoccurrences, which is traversed by a recursive principle at parse lime.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "A molc gcnct-al issue is the al)parenl antagonism 95 between the desire 1o have both sharp grammatical predictions and continuing openness to contextual revision. If after parsing (7) we transler the fact that smells are acceptable objects toper~ipiert into tile restricting ctxt feature, a later usage with an objecl of type sound fails. The opposite case concerns newly acquired specializable values. If in a later context these are used to update a 9en value, tile result may be too general. It is a topic of future research when to consider inR~rmation oct-lain and when to make revisable information restrictive.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "3" | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Valmme Alternation without LexL cal Rules", | |
| "authors": [ | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Bouma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Papers from the seventh CLIN Meeting", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "25--40", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Bouma, G. (1997). Valmme Alternation without LexL cal Rules. in: Papers from the seventh CLIN Meet- ing 1996, Emdhoven, 25-40.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Automatic Acquisilion of Subcategorization Fr~unes From Untagged Text", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Brent", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1991, | |
| "venue": "Proceedings of 29th ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "209--214", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Brent, M. R. (1991). Automatic Acquisilion of Subcat- egorization Fr~unes From Untagged Text. in: Pro- ceedings of 29th ACL, Berkeley, 209-214.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Computational A.v~ects of Constraint-llased Linguistic Description. IMS, Universit~it Stuttgart", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "& M" | |
| ], | |
| "last": "Dt~rre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Doma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dt~rre, J. & M. Doma (1993). CUF-A Formalism for Linguistic Knowledge Representation. In: J. D(~rre tEd.), Computational A.v~ects of Constraint-llased Linguistic Description. IMS, Universit~it Stuttgart. Deliverable R 1.2.A, DYANA-2 -ESPRIT Project 6852.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Syntactic Processing of Unknown Words", | |
| "authors": [ | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Erbach", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "Institute for Knowledge-Based Systems (IWBS), IBM S tuttg~ut", | |
| "volume": "131", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Erbach, G. (1990). Syntactic Processing of Un- known Words. IWBS Report 131, Institute for Knowledge-Based Systems (IWBS), IBM S tuttg~ut.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Learning from Texts -A Terminological Meta-Reasoning Perspective", | |
| "authors": [ | |
| { | |
| "first": "U", | |
| "middle": [], | |
| "last": "Hahn", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Klenner", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "& K", | |
| "middle": [], | |
| "last": "Schnattinger", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Connectiotzist, Statistical, and Symbolic Approaches to Learning.for Natural Lazzguage Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "453--468", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hahn, U., M. Klenner & K. Schnattinger (1996). Learning from Texts -A Terminological Meta- Reasoning Perspective. In: S. Wennter, E. Riloff & G. Scheler tEd.), Connectiotzist, Statistical, and Symbolic Approaches to Learning.for Natural Lazz- guage Processing, 453-468. Berlin: Springer.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "The Ups ~md Downs of Lexical Acquisition", | |
| "authors": [ | |
| { | |
| "first": "E", | |
| "middle": [ | |
| "M S L" | |
| ], | |
| "last": "Hastings", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Lytinen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Ptv\u00a2'eediltgs oJ' AAAI'94", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "754--759", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hastings, E M. & S. L. Lytinen (1994). The Ups ~md Downs of Lexical Acquisition. in: Ptv\u00a2'eediltgs oJ' AAAI'94, 754-759.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Linder Biologie -Lehrbuch fiir die Oberstufe", | |
| "authors": [ | |
| { | |
| "first": "Ft", | |
| "middle": [], | |
| "last": "Knodel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1980, | |
| "venue": "Stuttgart: J.13. Metzlersche Verlagsbuchhandlung", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Knodel, ft. (1980). Linder Biologie -Lehrbuch fiir die Oberstufe. Stuttgart: J.13. Metzlersche Verlags- buchhandlung.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Generalizing C~tse Fr~unes Using a ~lllesaurus ;rod the MDL Principle", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "&", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Abe", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "Proceedings qf Recent Advantages in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "239--248", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Li, H. & N. Abe (1995). Generalizing C~tse Fr~unes Us- ing a ~lllesaurus ;rod the MDL Principle. in: Pro- ceedings qf Recent Advantages in Natural Lan- guage Processing, Velingrad, Bulgaria, 239-248.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Dissociations between argument structure and grammatical relations", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "& I" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Sag", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Manning, C. & I. Sag (1995). Dissociations between argument structure and grammatical relations. Ms., Stanford University.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Head-Driven Phrase Slructure Grammar", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Pollard", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Sag", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pollard, C. & l. Sag (1994). Head-Driven Phrase Slructure Grammar. Chicago University Press.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Paradigms in Lexical Acquisition", | |
| "authors": [ | |
| { | |
| "first": "U", | |
| "middle": [], | |
| "last": "Zemik", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "Proceedings of the k)rst biternational Lexical Acquisition Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Zemik, U. (1989). Paradigms in Lexical Acquisition. In: U. Zernik ted.), Proceedings of the k)rst biter- national Lexical Acquisition Workshop, Detroit.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "Figure 1shows a relevant fragment. Whereas the combination of special-" | |
| }, | |
| "FIGREF1": { | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "Excerpt from t)7)e hierarchy izable information translates into simple type unification (e.g. nora_fern A ~,eut = rteut), combining" | |
| }, | |
| "FIGREF2": { | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "up nose perceives the stench' The relevant substructures corresponding to the lexical FSs of the unknown noun and verb involved are depicted in fig. 2. The leading feature paths synsemlloc[eont for Nase and synsemlloclcatlarg-st for perzipiert have been onfftted." | |
| }, | |
| "TABREF1": { | |
| "num": null, | |
| "content": "<table><tr><td>(4)</td><td>a. generalizable([j], [23) :=</td></tr><tr><td/><td>gen s> omllooloa, lhead [ot\u00d7,</td></tr><tr><td/><td>b. specializable(W):=</td></tr><tr><td/><td>[cat lhead noun ]] [synseml OC[centl ind J gend[~J]</td></tr><tr><td colspan=\"2\">2.2 Processing</td></tr><tr><td colspan=\"2\">The firsl step in processing sentences wflh unknown</td></tr><tr><td colspan=\"2\">or revisable words consists of conventional parsing.</td></tr><tr><td colspan=\"2\">Any HPSG-compalible parser may be used, subject</td></tr><tr><td colspan=\"2\">to the obvious requiren]ent that lexical lookup must</td></tr><tr><td colspan=\"2\">not lail if a word's phonology is unknown. A canon-</td></tr><tr><td colspan=\"2\">ical entry for such unknowil words is delined as lhe</td></tr><tr><td colspan=\"2\">disjunction of maximally underspecilied generic lex-</td></tr><tr><td colspan=\"2\">ical entries for nottns, adjectives anti verbs.</td></tr><tr><td colspan=\"2\">The actual updating of lexical enlries consists of</td></tr><tr><td colspan=\"2\">lout major steps.</td></tr></table>", | |
| "text": "ql~e last representalional issue is how to identify revisable infornlation in (subslrnctures el) tile parse FS. D)r this purpose lhe grammar delines revisabilily clauses like the following:", | |
| "type_str": "table", | |
| "html": null | |
| }, | |
| "TABREF2": { | |
| "num": null, | |
| "content": "<table><tr><td>Nase</td><td colspan=\"2\">perzipiert</td></tr><tr><td>after (5)</td><td/></tr><tr><td>gendfem</td><td>]</td></tr><tr><td colspan=\"2\">gen u.g ctxt sense~organJ [</td><td>ctxt arg~rtntcJ</td></tr><tr><td>after (6)</td><td/></tr><tr><td colspan=\"2\">gend/em gen u..g ctxt sense~grgan[ ] I</td><td>gen u~,Vnpnom ctxt arg.vtrue .rg../[,oo I cont [ gen It-gVear]] 1_ I] \\ L L taxt n\u00b0m\"renuJ</td></tr><tr><td>alter (7) gend fern ] gen u_g / ctxt noseJ</td><td colspan=\"2\">gen l,t_gVttpnornVnpnom~lpacc ctxt arg_rtruc r, , .rgen u_gVsense~vrgan]] I 1 I /[,oo loom [ot~t.,o,,,_~., JJ'\\/ args I \\[Ioccont rgenu-gvs'ellTl' /)| L L [~t~t,,o.,_,~,,, JJ I_ J</td></tr><tr><td colspan=\"2\">Figure 2:</td></tr></table>", | |
| "text": "Updates on Iexical FSs mapping. On the basis of tiffs type classification of argument structure patterns, the parse derived the clxt value npnom. Since gen values are generalizable, this new value is unioned with the old lexical gen value. Note that ctxt is properly unaffected.", | |
| "type_str": "table", | |
| "html": null | |
| } | |
| } | |
| } | |
| } |