| { | |
| "paper_id": "N12-1030", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T14:04:55.742252Z" | |
| }, | |
| "title": "The Challenges of Parsing Chinese with Combinatory Categorial Grammar", | |
| "authors": [ | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Tse", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Sydney", | |
| "location": { | |
| "country": "Australia" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "James", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Curran", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Sydney", | |
| "location": { | |
| "country": "Australia" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "We apply Combinatory Categorial Grammar to wide-coverage parsing in Chinese with the new Chinese CCGbank, bringing a formalism capable of transparently recovering non-local dependencies to a language in which they are particularly frequent. We train two state-of-the-art English \uf763\uf763\uf767 parsers: the parser of Petrov and Klein (P&K), and the Clark and Curran (C&C) parser, uncovering a surprising performance gap between them not observed in English-72.73 (P&K) and 67.09 (C&C) F-score on \uf770\uf763\uf774\uf762 6. We explore the challenges of Chinese \uf763\uf763\uf767 parsing through three novel ideas: developing corpus variants rather than treating the corpus as fixed; controlling noun/verb and other \uf770\uf76f\uf773 ambiguities; and quantifying the impact of constructions like pro-drop.", | |
| "pdf_parse": { | |
| "paper_id": "N12-1030", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "We apply Combinatory Categorial Grammar to wide-coverage parsing in Chinese with the new Chinese CCGbank, bringing a formalism capable of transparently recovering non-local dependencies to a language in which they are particularly frequent. We train two state-of-the-art English \uf763\uf763\uf767 parsers: the parser of Petrov and Klein (P&K), and the Clark and Curran (C&C) parser, uncovering a surprising performance gap between them not observed in English-72.73 (P&K) and 67.09 (C&C) F-score on \uf770\uf763\uf774\uf762 6. We explore the challenges of Chinese \uf763\uf763\uf767 parsing through three novel ideas: developing corpus variants rather than treating the corpus as fixed; controlling noun/verb and other \uf770\uf76f\uf773 ambiguities; and quantifying the impact of constructions like pro-drop.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Automatic corpus conversions from the Penn Treebank (Marcus et al., 1994) have driven research in lexicalised grammar formalisms, such as \uf76c\uf774\uf761\uf767 (Xia, 1999) , \uf768\uf770\uf773\uf767 (Miyao et al., 2004) and \uf763\uf763\uf767 (Hockenmaier and Steedman, 2007) , producing the lexical resources key to wide-coverage statistical parsing.", | |
| "cite_spans": [ | |
| { | |
| "start": 52, | |
| "end": 73, | |
| "text": "(Marcus et al., 1994)", | |
| "ref_id": "BIBREF18" | |
| }, | |
| { | |
| "start": 143, | |
| "end": 154, | |
| "text": "(Xia, 1999)", | |
| "ref_id": "BIBREF26" | |
| }, | |
| { | |
| "start": 162, | |
| "end": 182, | |
| "text": "(Miyao et al., 2004)", | |
| "ref_id": "BIBREF20" | |
| }, | |
| { | |
| "start": 208, | |
| "end": 223, | |
| "text": "Steedman, 2007)", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The Chinese Penn Treebank (\uf770\uf763\uf774\uf762; Xue et al., 2005) has filled a comparable niche, enabling the development of a Chinese \uf76c\uf774\uf761\uf767 (Xia et al., 2000) , a wide-coverage \uf768\uf770\uf773\uf767 parser (Yu et al., 2011) , and recently Chinese CCGbank (Tse and Curran, 2010) , a 750 000-word corpus of Combinatory Categorial Grammar (\uf763\uf763\uf767; Steedman, 2000) derivations.", | |
| "cite_spans": [ | |
| { | |
| "start": 33, | |
| "end": 50, | |
| "text": "Xue et al., 2005)", | |
| "ref_id": "BIBREF28" | |
| }, | |
| { | |
| "start": 125, | |
| "end": 143, | |
| "text": "(Xia et al., 2000)", | |
| "ref_id": "BIBREF27" | |
| }, | |
| { | |
| "start": 174, | |
| "end": 191, | |
| "text": "(Yu et al., 2011)", | |
| "ref_id": "BIBREF29" | |
| }, | |
| { | |
| "start": 223, | |
| "end": 245, | |
| "text": "(Tse and Curran, 2010)", | |
| "ref_id": "BIBREF24" | |
| }, | |
| { | |
| "start": 310, | |
| "end": 325, | |
| "text": "Steedman, 2000)", | |
| "ref_id": "BIBREF23" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "We train two \uf763\uf763\uf767 parsers, Clark and Curran (C&C; , and the Petrov and Klein (P&K; \uf770\uf763\uf766\uf767 parser, on Chinese CCGbank. We follow Fowler and Penn (2010), who treat the English CCGbank (Hockenmaier and Steedman, 2007) grammar as a \uf763\uf766\uf767 and train and evaluate the P&K parser directly on it.", | |
| "cite_spans": [ | |
| { | |
| "start": 26, | |
| "end": 48, | |
| "text": "Clark and Curran (C&C;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 59, | |
| "end": 81, | |
| "text": "Petrov and Klein (P&K;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 196, | |
| "end": 211, | |
| "text": "Steedman, 2007)", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "We obtain the first Chinese \uf763\uf763\uf767 parsing results:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "F -scores of 72.73 (P&K) and 67.09 (C&C) on labelled dependencies computed over the \uf770\uf763\uf774\uf762 6 test set. While the state-of-the-art in Chinese syntactic parsing has always lagged behind English, this large gap is surprising, given that Fowler and Penn 2010found only a small margin separated the two parsers on English CCGbank (86.0 versus 85.8). Levy and Manning (2003) established that properties of Chinese such as noun/verb ambiguity contribute to the difficulty of Chinese parsing. We focus on two factors within our control: annotation decisions and parser architecture.", | |
| "cite_spans": [ | |
| { | |
| "start": 343, | |
| "end": 366, | |
| "text": "Levy and Manning (2003)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Existing research has varied parsers whilst keeping the corpus fixed. We vary the corpus whilst keeping the parsers fixed by exploring multiple design choices for particular constructions. By exploiting the fully automatic CCGbank extraction process, we can immediately implement these choices and assess their impact on parsing performance.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Secondly, we contrast the performance of C&C, with its tagging/parsing pipeline, with P&K, a parser which performs joint tagging and parsing, and establish that P&K is less sensitive to the greater lexical category ambiguity in Chinese CCGbank.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "We demonstrate that Chinese \uf763\uf763\uf767 parsing is very difficult, and propose novel techniques for identifying where the challenges lie.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "\u88ab \uf770\uf773\uf773 \u56f0 trap \u7684 \uf764\uf765 \u516c\u4e3b princess \u6211 I \u89e3\u6551\u4e86 rescued (S[dcl]\\NP)/((S[dcl]\\NP)/NP) (S[dcl]\\NP)/NP (N/N)\\(S[dcl]/NP) N NP (S[dcl]\\NP)/NP \u227b \u227bT S[dcl]\\NP S/(S\\NP) \u227a \u227bB N/N S[dcl]/NP \u227b N \u2212\u2192 NP T S/(S/NP) \u227b S[dcl]", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Figure 1: 3 types of non-local dependencies in 6 words: \"(As for) the trapped princess, I rescued (her).\"", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Bikel and Chiang (2000) developed the first \uf770\uf763\uf774\uf762 parser, demonstrating that Chinese was similar enough to English for techniques such as a Collinsstyle head-driven parser or \uf774\uf761\uf767 to succeed. Later \uf770\uf763\uf774\uf762 parsers used Tree Insertion Grammar (Chiang and Bikel, 2002) , \uf770\uf763\uf766\uf767s (Levy and Manning, 2003) , the Collins models (Bikel, 2004) and transition-based discriminative models (Wang et al., 2006; Zhang and Clark, 2009; Huang et al., 2009) . These systems also established the relative difficulty of parsing Chinese and English; while \uf770\uf761\uf772\uf773\uf765\uf776\uf761\uf76c scores over 92% are possible for English (McClosky et al., 2006) , systems for Chinese have achieved only 87% (Zhang and Clark, 2009) on the same metric. Non-local dependencies (\uf76e\uf76c\uf764s) are lexical dependencies which hold over unbounded distances. Guo et al. (2007) observed that despite the importance of \uf76e\uf76c\uf764s for correct semantic interpretation, and the fact that Chinese syntax generates more \uf76e\uf76c\uf764s than English, few parsers in Chinese are equipped to recover the traces which mark \uf76e\uf76c\uf764s. For instance, extraction, a common \uf76e\uf76c\uf764 type, occurs more frequently in \uf763\uf770\uf774\uf762 sentences (38%) compared to \uf770\uf774\uf762 (17%).", | |
| "cite_spans": [ | |
| { | |
| "start": 10, | |
| "end": 23, | |
| "text": "Chiang (2000)", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 237, | |
| "end": 261, | |
| "text": "(Chiang and Bikel, 2002)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 270, | |
| "end": 294, | |
| "text": "(Levy and Manning, 2003)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 316, | |
| "end": 329, | |
| "text": "(Bikel, 2004)", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 373, | |
| "end": 392, | |
| "text": "(Wang et al., 2006;", | |
| "ref_id": "BIBREF25" | |
| }, | |
| { | |
| "start": 393, | |
| "end": 415, | |
| "text": "Zhang and Clark, 2009;", | |
| "ref_id": "BIBREF31" | |
| }, | |
| { | |
| "start": 416, | |
| "end": 435, | |
| "text": "Huang et al., 2009)", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 581, | |
| "end": 604, | |
| "text": "(McClosky et al., 2006)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 650, | |
| "end": 673, | |
| "text": "(Zhang and Clark, 2009)", | |
| "ref_id": "BIBREF31" | |
| }, | |
| { | |
| "start": 786, | |
| "end": 803, | |
| "text": "Guo et al. (2007)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Background", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "A more satisfying approach is to use a grammar formalism, such as \uf763\uf763\uf767 (Steedman, 2000) , which generates them inherently, enabling a unified parsing model over local and non-local dependencies. This approach is taken in the C&C parser (Clark and Curran, 2007) , which can directly and transparently recover \uf76e\uf76c\uf764s in English (Rimell et al., 2009) .", | |
| "cite_spans": [ | |
| { | |
| "start": 70, | |
| "end": 86, | |
| "text": "(Steedman, 2000)", | |
| "ref_id": "BIBREF23" | |
| }, | |
| { | |
| "start": 235, | |
| "end": 259, | |
| "text": "(Clark and Curran, 2007)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 323, | |
| "end": 344, | |
| "text": "(Rimell et al., 2009)", | |
| "ref_id": "BIBREF22" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Background", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Chinese CCGbank (Tse and Curran, 2010) demonstrates that a parsimonious account of Chinese syntax with \uf763\uf763\uf767 is possible. Many familiar objects of Chinese syntax which generate \uf76e\uf76c\uf764s, including the \u628a ba/\u88ab bei constructions, topicalisation and extraction receive natural \uf763\uf763\uf767 analyses in Chinese CCGbank. Figure 1 shows the CCGbank analysis of passivisation, topicalisation and extraction, creating \uf76e\uf76c\uf764s between \u516c\u4e3b princess and each of \u88ab \uf770\uf773\uf773, \u56f0 trap and \u89e3\u6551 rescue respectively. We take two state-of-the-art parsers and train them to establish the difficulty of parsing Chinese with \uf763\uf763\uf767. The first is the Clark and Curran (C&C; parser, which uses supertagging (Clark and Curran, 2004) , a local, linear-time tagging technique which drastically prunes the space of lexical categories which the polynomial-time parsing algorithm later considers. The second is the coarse-to-fine parser of Petrov and Klein (2007) which iteratively refines its grammar by splitting production rules to uncover latent distinctions. Fowler and Penn (2010) demonstrate that the English CCGbank grammar is strongly context-free, allowing them to treat it as a \uf763\uf766\uf767 and train the Petrov and Klein (2007) parser directly.", | |
| "cite_spans": [ | |
| { | |
| "start": 16, | |
| "end": 38, | |
| "text": "(Tse and Curran, 2010)", | |
| "ref_id": "BIBREF24" | |
| }, | |
| { | |
| "start": 599, | |
| "end": 621, | |
| "text": "Clark and Curran (C&C;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 654, | |
| "end": 678, | |
| "text": "(Clark and Curran, 2004)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 881, | |
| "end": 904, | |
| "text": "Petrov and Klein (2007)", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 1148, | |
| "end": 1171, | |
| "text": "Petrov and Klein (2007)", | |
| "ref_id": "BIBREF21" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 300, | |
| "end": 308, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Background", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(a) Derivational . . . . S/S . . . . . NP . . . . . . . NP/N . . . N (b) Lexical . . . . S/S . . . . . . . (S/S)/N . . . N", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Background", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "The designer of a CCGbank must frequently choose between derivational and lexical ambiguity (Hockenmaier, 2003; Tse and Curran, 2010) . Derivational ambiguity analyses special constructions through arbitrary label-rewriting phrase structure rules, while lexical ambiguity assigns additional categories to lexical items for when they participate in special constructions.", | |
| "cite_spans": [ | |
| { | |
| "start": 92, | |
| "end": 111, | |
| "text": "(Hockenmaier, 2003;", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 112, | |
| "end": 133, | |
| "text": "Tse and Curran, 2010)", | |
| "ref_id": "BIBREF24" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Derivational vs. lexical ambiguity", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Derivational and lexical ambiguity often arise in \uf763\uf763\uf767 because of the form-function distinctionwhen the syntactic form of a constituent does not coincide with its semantic function (Honnibal, 2010) . For instance, in English, topicalisation causes an NP to appear in clause-initial position, fulfilling the function of a sentential pre-modifier while maintaining the form of an NP. Figure 2 shows two distinct \uf763\uf763\uf767 analyses which yield the same dependency edges.", | |
| "cite_spans": [ | |
| { | |
| "start": 180, | |
| "end": 196, | |
| "text": "(Honnibal, 2010)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 381, | |
| "end": 389, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Derivational vs. lexical ambiguity", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Derivational ambiguity increases the parser search space, while lexical ambiguity enlarges the tag set, and hence the complexity of the supertagging task.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Derivational vs. lexical ambiguity", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "We extract three versions of Chinese CCGbank to explore the trade-off between lexical and derivational ambiguity, training both parsers on each corpus to determine the impact of the annotation changes. Our hypothesis is that the scarcity of training data in Chinese means that derivational ambiguity results in better coverage and accuracy, at the cost of increasing time and space requirements of the resulting parser.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Three versions of Chinese CCGbank", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "In the following sentences, the words in bold have often been analysed as belonging to a lexical category localiser (Chao, 1968; Li and Thompson, 1989) .", | |
| "cite_spans": [ | |
| { | |
| "start": 116, | |
| "end": 128, | |
| "text": "(Chao, 1968;", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 129, | |
| "end": 151, | |
| "text": "Li and Thompson, 1989)", | |
| "ref_id": "BIBREF17" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(1) a. \u5c4b\u5b50 house \u91cc\u9762 inside:\uf76c\uf763 the inside of the house/inside the house b. \u5927 big \u6811 tree \u65c1\u8fb9 beside:\uf76c\uf763 (the area) beside the big tree Localisers, like English prepositions, identify a (temporal, spatial, etc.) extent of their complement. However, the combination Noun + Localiser is ambiguous between noun function (the inside of the house) and modifier function (inside the house).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "We consider two possibilities to represent localisers in \uf763\uf763\uf767, which trade derivational for lexical ambiguity. In (2-a), a direct \uf763\uf763\uf767 transfer of the \uf770\uf763\uf774\uf762 analysis, the preposition \u5728 at expects arguments of type LCP. In (2-b), \u5728 at now expects only NP arguments, and the unary promotion LCP \u2192 NP allows LCP-form constituents to function as NPs.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(2) a. \u5728 at \u623f\u5b50 room \u91cc in:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "\uf76c\uf763 PP/LCP NP LCP\\NP \u227a LCP \u227b PP b. \u5728 at \u623f\u5b50 room \u91cc in:\uf76c\uf763 PP/NP NP LCP\\NP \u227a LCP \u2192 NP \u227b PP", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The analysis in (2-a) exhibits greater lexical ambiguity, with the lexical item \u5728 at carrying at least two categories, PP/NP and PP/LCP, while (2-b) trades off derivational for lexical ambiguity: the unary promotion LCP \u2192 NP becomes necessary, but \u5728 at no longer needs the category PP/LCP. The base release of Chinese CCGbank, corpus A, like (2-a), makes the distinction between categories LCP and NP. However, in corpus B, we test the impact of applying (2-b), in which the unary promotion LCP \u2192 NP is available.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The lexical category LC (localiser)", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The most frequent unary rule in English CCGbank, occurring in over 91% of sentences, is the promotion from bare to non-bare nouns: N \u2192 NP. Hockenmaier (2003) explains that the rule accounts for the form-function distinction in determiner-less English nouns which nevertheless have definite reference, while preventing certain over-generations (e.g. * the the car). The N-NP distinction also separates adjectives and noun modifiers (category N/N), from predeterminers (category NP/NP) (Hockenmaier and Steedman, 2005) , a distinction also made in Chinese.", | |
| "cite_spans": [ | |
| { | |
| "start": 139, | |
| "end": 157, | |
| "text": "Hockenmaier (2003)", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 484, | |
| "end": 516, | |
| "text": "(Hockenmaier and Steedman, 2005)", | |
| "ref_id": "BIBREF12" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "While Chinese has strategies to mark definite or indefinite reference, they are not obligatory, and a bare noun is referentially ambiguous, calling into question whether the distinction is justified in \uf763\uf763\uf767:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "(3) a. \u72d7 dog \u5f88 very \u806a\u660e clever Dogs are clever. b. \u6211 \uf644\uf773\uf767 \u770b\u5230 see \u72d7 dog I saw a dog/dogs. c. \u72d7 dog \u8dd1\u8d70 run-away \u4e86 \uf761\uf773\uf770", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The dog/dogs ran away.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The fact that the Chinese determiner is not necessarily a maximal projection of the noun -in other words, the determiner does not 'close off' a level of NPalso argues against importing the English analysis. In contrast, the English CCGbank determiner category NP/N reflects the fact that determiners 'close off' NP -further modification by noun modifiers is blocked after combining with a determiner.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "(4) \u5171\u548c\u515a Republican Party \u8fd9 this", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The bare/non-bare NP distinction", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "To test its impact on Chinese parsing, we create a version of Chinese CCGbank (corpus C) which neutralises the distinction. This eliminates the atomic category N, as well as the promotion rule N \u2192 NP.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u4e3e\u52a8 act this action by the Republican Party", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "While a standard split of \uf770\uf763\uf774\uf762 5 exists, as defined by Zhang and Clark (2008) , we are not aware of a consistently used split for \uf770\uf763\uf774\uf762 6. We present a new split in Table 1 which adds data from the \uf761\uf763\uf765 broadcast section of \uf770\uf763\uf774\uf762 6, maintaining the same train/dev/test set proportions as the \uf770\uf763\uf774\uf762 5 split.", | |
| "cite_spans": [ | |
| { | |
| "start": 55, | |
| "end": 77, | |
| "text": "Zhang and Clark (2008)", | |
| "ref_id": "BIBREF30" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 164, | |
| "end": 171, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "We train C&C using the hybrid model, the bestperforming model for English, which extracts features from the dependency structure (Clark and Curran, 2007) . We use \u03b2 = \u27e80.055, 0.01, 0.05, 0.1\u27e9 during training with a Gaussian smoothing parameter \u03b1 = 2.4 (optimised on the corpus A dev set). We use \u03b2 = \u27e80.15, 0.075, 0.03, 0.01, 0.005, 0.001\u27e9 during parsing, with the maximum number of supercats (chart entries) set to 5,000,000, reflecting the greater supertagging ambiguity of Chinese parsing.", | |
| "cite_spans": [ | |
| { | |
| "start": 129, | |
| "end": 153, | |
| "text": "(Clark and Curran, 2007)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The P&K parser is used \"off-the-shelf\" and trained with its default parameters, only varying the number of split-merge iterations and enabling the Chinesespecific lexicon features. The P&K parser involves no explicit \uf770\uf76f\uf773 tagging step, as the (super)tags correspond directly to non-terminals in a \uf763\uf766\uf767. Since the P&K parser plus generate produce dependencies in the same format as C&C, we can use the standard Clark and Curran (2007) dependencybased evaluation from the \uf763\uf763\uf767 literature: labelled Fscore (LF ) over dependency tuples, as used for \uf763\uf763\uf767 parser evaluation in English. Critically, this metric is also \uf76e\uf76c\uf764-sensitive. We also report labelled sentence accuracy (Lsa), the proportion of sentences for which the parser returned all and only the gold standard dependencies. Supertagger accuracy compares leaf categories against the gold standard (stag).", | |
| "cite_spans": [ | |
| { | |
| "start": 408, | |
| "end": 431, | |
| "text": "Clark and Curran (2007)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "For C&C, we report on two configurations: \uf767\uf76f\uf76c\uf764, evaluated using gold standard \uf770\uf76f\uf773 tags; and \uf761\uf775\uf774\uf76f, with automatic \uf770\uf76f\uf773 tags provided by the C&C tagger (Curran and Clark, 2003) . For P&K, we vary the number of split-merge iterations from one to six (following Fowler and Penn (2010), the k-iterations model is called I-k). Because the P&K parser does not use \uf770\uf76f\uf773 tags, the most appropriate comparison is against the \uf761\uf775\uf774\uf76f configuration of C&C. For C&C, we use the average of the logarithm of the chart size (log C) as a measure of ambiguity, that is, the number of alternative analyses the parser must choose between.", | |
| "cite_spans": [ | |
| { | |
| "start": 161, | |
| "end": 173, | |
| "text": "Clark, 2003)", | |
| "ref_id": "BIBREF8" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Following Fowler and Penn (2010), we perform two sets of experiments: one evaluated over all sentences in a section, and another evaluated only over sentences for which both parsers successfully parse and generate dependencies.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "We define the size of a \uf763\uf763\uf767 grammar as the number of categories it contains. The size of a grammar affects the difficulty of the supertagging task (as the size of a grammar is the size of the supertag set). We also consider the number of categories of each shape, as defined in Table 2 . Decomposing the category in- ventory into shapes demonstrates how changes to the corpus annotation affect the distribution of types of category. Finally, we calculate the average number of tags per lexical item (Avg. Tags/Word), as a metric of the degree of lexical ambiguity in each corpus. Table 3 shows the performance of P&K and C&C on the three dev sets, and Table 4 only over sentences parsed by both parsers. (A is the base release, B includes the unary rule LCP \u2192 NP, and C also collapses the N-NP distinction.) For P&K on corpus A, F -score and supertagger accuracy increase monotonically as further split-merge iterations refine the model. P&K on B and C overfits at 6 iterations, consistent with Fowler and Penn's findings for English. Supertagging and parsing accuracy are not entirely correlated between the parsers -in corpora A and B, \uf761\uf775\uf774\uf76f supertagging is comparable or better than I-3, but F -score is substantially worse.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 278, | |
| "end": 285, | |
| "text": "Table 2", | |
| "ref_id": "TABREF2" | |
| }, | |
| { | |
| "start": 580, | |
| "end": 587, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 652, | |
| "end": 659, | |
| "text": "Table 4", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Shape Pattern V (predicate-like) (S[dcl]\\NP)$ M (modifier) X|X P (preposition-like) (X|X)|Y N (noun-like) N or NP O (all others)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Comparing A and B in Table 3 , C&C receives small increases in supertagger accuracy and coverage, but parsing performance remains largely unchanged; P&K performance degrades slightly. On both parsers, C yields the best results out of the three corpora, with LF gains of 1.07 (P&K), 1.28 (\uf767\uf76f\uf76c\uf764) and 0.63 (\uf761\uf775\uf774\uf76f) over the base Chinese CCGbank. We select C for our remaining parser experiments.", | |
| "cite_spans": [ | |
| { | |
| "start": 287, | |
| "end": 293, | |
| "text": "(\uf767\uf76f\uf76c\uf764)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 21, | |
| "end": 28, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Both C&C's \uf767\uf76f\uf76c\uf764 and \uf761\uf775\uf774\uf76f results show higher coverage than P&K (a combination of parse failures in P&K itself, and in generate). Since F -score is only computed over successful parses, it is possible that P&K is avoiding harder sentences. In Table 4 , evaluated only over sentences parsed by both parsers shows that as expected, C&C gains more (1.15%) than P&K on the common sentences. Table 7 : Grammar size, categorised by shape", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 242, | |
| "end": 249, | |
| "text": "Table 4", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 386, | |
| "end": 393, | |
| "text": "Table 7", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "To understand why corpus C is superior for parsing, we compare the ambiguity and sparsity characteristics of the three corpora. Examining log C, the average log-chart size (Table 3) shows that the corpus B changes (the addition of the unary rule LCP \u2192 NP) increase ambiguity, while the additional corpus C changes (eliminating the N-NP distinction, resulting in the removal of the unary rule N \u2192 NP) have the net effect of reducing ambiguity. Table 6 shows that the changes reduce the size of the lexicon, thus reducing the average number of tags each word can potentially receive, and therefore the difficulty of the supertagging task. This, in part, contributes to the reduced log C values in Table 3 . While the size of the lexicon is reduced in B, the corresponding log C figure in Table 3 increases slightly, because of the additional unary rule. Table 7 breaks down the size of each lexicon according to category shape. Introducing the rule LCP \u2192 NP reduces the number of V-shaped categories by 10%, while not substantially affecting the quantity of other category shapes, because the subcategorisation frames which previously referred to LCP are no longer necessary. Eliminating the N-NP distinction, however, reduces the number of P and M-shaped categories by over 20%, as the distinction is no longer made between attachment at N and NP.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 172, | |
| "end": 181, | |
| "text": "(Table 3)", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 443, | |
| "end": 450, | |
| "text": "Table 6", | |
| "ref_id": "TABREF7" | |
| }, | |
| { | |
| "start": 695, | |
| "end": 702, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 786, | |
| "end": 793, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 852, | |
| "end": 859, | |
| "text": "Table 7", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Corpus ambiguity", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "The well-known noun/verb ambiguity in Chinese (where, e.g., \u8bbe\u8ba1\u5efa\u8bbe 'design-build' is both a verbal compound 'design and build' and a noun compound 'design and construction') greatly affects parsing accuracy (Levy and Manning, 2003) .", | |
| "cite_spans": [ | |
| { | |
| "start": 205, | |
| "end": 229, | |
| "text": "(Levy and Manning, 2003)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Error analysis", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "However, little work has quantified the impact of noun/verb ambiguity on parsing, and for that matter, the impact of other frequent confusion types. which we saw in Table 3 , we perform an experiment where we corrupt the gold \uf770\uf76f\uf773 tags, by gradually reintroducing automatic \uf770\uf76f\uf773 errors on a cumulative basis, one confusion type at a time.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 165, | |
| "end": 172, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Error analysis", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "The notation X \u25b7\u25c1 Y indicates that the \uf770\uf76f\uf773 tags X and Y are frequently confused with each other by the \uf770\uf76f\uf773 tagger. For example, VV \u25b7\u25c1 NN represents the problematic noun/verb ambiguity, allowing the inclusion of noun/verb confusion errors. Table 8 shows that while the confusion types NR \u25b7\u25c1 NN and JJ \u25b7\u25c1 NN have no impact on the evaluation, the confusions DEC \u25b7\u25c1 DEG and VV \u25b7\u25c1 NN, introduced one at a time, cause reductions in F -score of 1.50 and 1.75% respectively. This is expected; Chinese CCGbank does not distinguish between noun modifiers (NN) and adjectives (JJ). On the other hand, the critical noun/verb ambiguity, and the confusion between DEC/DEG (two senses of the particle \u7684 de) adversely impact F -score. We performed an experiment with C&C to merge DEC and DEG into a single tag, but found that this increased category ambiguity without improving accuracy.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 239, | |
| "end": 246, | |
| "text": "Table 8", | |
| "ref_id": "TABREF9" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Error analysis", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "The VV \u25b7\u25c1 NN confusion is particularly damaging to the \uf763\uf763\uf767 labelled dependency evaluation, because verbs generate a large number of dependencies. While Fowler and Penn (2010) report a gap of 6.31% between C&C's labelled and unlabelled F -score on the development set in English, we observe a gap of 10.35% for Chinese. The sensitivity of C&C to tagging errors, and the higher performance of the P&K parser, which does not directly use \uf770\uf76f\uf773 tags, calls into question whether \uf770\uf76f\uf773 tagging yields a net gain in a language where distinctions such as the noun/verb ambiguity are often difficult to resolve using local tagging approaches. The approach of Auli and Lopez (2011) , which achieves superior results in English \uf763\uf763\uf767 parsing with a joint supertagging/parsing model, may be promising in light of the performance difference between P&K and C&C. Table 9 shows how well the best models of each parser recovered selected local and non-local dependencies. The slot represented by each row appears in boldface. While C&C and P&K perform similarly recovering NP-internal structure, the ability of P&K to recover verbal arguments, unbounded long-range dependencies such as subject and object extraction, and bounded long-range dependencies such as control/raising constructions, is superior.", | |
| "cite_spans": [ | |
| { | |
| "start": 647, | |
| "end": 668, | |
| "text": "Auli and Lopez (2011)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 844, | |
| "end": 851, | |
| "text": "Table 9", | |
| "ref_id": "TABREF11" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Error analysis", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "The C&C \uf761\uf775\uf774\uf76f parser appears to be biased towards generating far more of the frequent dependency types, yet does not typically have a higher recall for these dependency types than P&K.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Non-local dependencies", | |
| "sec_num": "6.1" | |
| }, | |
| { | |
| "text": "One of the most common types of unary rules in Chinese CCGbank, occurring in 36% of Chinese CCGbank sentences, is the subject pro-drop rule S[dcl]\\NP \u2192 S [dcl] , which accounts for the optional absence of the subject pronoun of a verb for pragmatic reasons where the referent can be recovered from the discourse (Li and Thompson, 1989) .", | |
| "cite_spans": [ | |
| { | |
| "start": 154, | |
| "end": 159, | |
| "text": "[dcl]", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 312, | |
| "end": 335, | |
| "text": "(Li and Thompson, 1989)", | |
| "ref_id": "BIBREF17" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Pro-drop and its impact on \uf763\uf763\uf767 parsing", | |
| "sec_num": "6.2" | |
| }, | |
| { | |
| "text": "The subject pro-drop rule is problematic in Chinese parsing because its left hand side, S[dcl]\\NP, is a very common category, and also because several syntactic distinctions in Chinese CCGbank hinge on the difference between S[dcl]\\NP and S [dcl] .", | |
| "cite_spans": [ | |
| { | |
| "start": 241, | |
| "end": 246, | |
| "text": "[dcl]", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Pro-drop and its impact on \uf763\uf763\uf767 parsing", | |
| "sec_num": "6.2" | |
| }, | |
| { | |
| "text": "The latter point is illustrated by two of the senses of \u7684 de, the Chinese subordinating particle. Two categories which \u7684 de receives in the grammar are (NP/NP)\\(S[dcl]\\NP) (introducing a relative clause) and (NP/NP)\\S[dcl] (in the construction S de NP). Because subject pro-drop promotes any unsaturated S [dcl]\\NP to S[dcl] , whenever the supertagger returns both of the above categories for the lexical item \u7684 de, the parser must consider two alternative analyses which yield different dependencies:", | |
| "cite_spans": [ | |
| { | |
| "start": 306, | |
| "end": 324, | |
| "text": "[dcl]\\NP to S[dcl]", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Pro-drop and its impact on \uf763\uf763\uf767 parsing", | |
| "sec_num": "6.2" | |
| }, | |
| { | |
| "text": "(5) a. t i t i \u51fa\u6765 come out \u7684 \uf764\uf765", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Pro-drop and its impact on \uf763\uf763\uf767 parsing", | |
| "sec_num": "6.2" | |
| }, | |
| { | |
| "text": "\u95ee\u9898 i question i the questions which arise English Chinese PTB/PCTB-based 92.1% (McClosky et al., 2006) 86.8% (Zhang and Clark, 2009) CCGbank-based 86.0% (Fowler and Penn, 2010) 72.7% (this work) 85.8% (Clark and Curran, 2007) 67.1% (this work) ", | |
| "cite_spans": [ | |
| { | |
| "start": 79, | |
| "end": 102, | |
| "text": "(McClosky et al., 2006)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 109, | |
| "end": 132, | |
| "text": "(Zhang and Clark, 2009)", | |
| "ref_id": "BIBREF31" | |
| }, | |
| { | |
| "start": 201, | |
| "end": 225, | |
| "text": "(Clark and Curran, 2007)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Pro-drop and its impact on \uf763\uf763\uf767 parsing", | |
| "sec_num": "6.2" | |
| }, | |
| { | |
| "text": "\u95ee\u9898 question the question of (him, her) coming out 38.1% of sentences in the development set contain at least one instance of pro-drop. The evaluation over only these sentences is given in Table 12 . This restricted evaluation shows that while we cannot conclude that pro-drop is the causative factor, sentences with pro-drop are much more difficult for both parsers to analyse correctly, although the drops in Fscore and supertagging accuracy are largest for P&K. Critically, the fact that supertagging performance on these more difficult sentences is reasonably comparable with performance on the full set suggests that the bottleneck is in the parser rather than the supertagger. One measure of the complexity of prodrop sentences is the substantial increase in the log C value of these sentences. This suggests that a key to bringing parser performance on Chinese in line with English lies in reining in the ambiguity caused by very productive unary rules such as pro-drop.", | |
| "cite_spans": [ | |
| { | |
| "start": 459, | |
| "end": 463, | |
| "text": "P&K.", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 188, | |
| "end": 196, | |
| "text": "Table 12", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "\uf764\uf765", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Using Chinese CCGbank (Tse and Curran, 2010), we have trained and evaluated the first \uf763\uf763\uf767 parsers for Chinese in the literature: the Clark and Curran (C&C; and Petrov and Klein (P&K; parsers. The P&K parser substantially outperformed (72.73) C&C with automatic \uf770\uf76f\uf773 tags (67.09). Table 11 summarises the best performance of parsers on \uf770\uf774\uf762 and CCGbank, for English and Chinese. We observe a drop in performance between English and Chinese \uf763\uf763\uf767 parsers which is much larger than, but consistent with, \uf770\uf774\uf762 parsers. To close this gap, future research in Chinese parsing should be informed by quantifying the aspects of Chinese which account most for the deficit.", | |
| "cite_spans": [ | |
| { | |
| "start": 133, | |
| "end": 155, | |
| "text": "Clark and Curran (C&C;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 160, | |
| "end": 182, | |
| "text": "Petrov and Klein (P&K;", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 279, | |
| "end": 287, | |
| "text": "Table 11", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "We start by using corpus conversion to compare different linguistic representation choices, rather than for generating a single immutable resource. This can also be exploited to develop syntactic corpora parameterised for particular applications. We found that collapsing categorial distinctions motivated by theory can yield less ambiguous corpora, and hence, more accurate parsers. We have also taken a novel approach to investigating the impact of noun/verb and other \uf770\uf76f\uf773 ambiguities on parsing.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "The large gap between Chinese C&C and P&K is surprising, given that Fowler and Penn (2010) found only a small gap for English. We found that C&C is very sensitive to \uf770\uf76f\uf773 tagging performance, which leads to its inferior performance given automatically assigned \uf770\uf76f\uf773 tags. This suggests that joint supertagging/parsing approaches, as performed by P&K, are more suitable for Chinese. Finally, we have shown that pro-drop is correlated with poor performance on both parsers, suggesting an avenue to closing the Chinese-English parsing gap.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "While developing the first wide-coverage Chinese \uf763\uf763\uf767 parsers, we have shed light on the nature of the Chinese-English parsing gap, and identified new and significant challenges for \uf763\uf763\uf767 parsing.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "7" | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "We thank our anonymous reviewers for their insightful and detailed feedback. James R. Curran was supported by Australian Research Council (\uf761\uf772\uf763) Discovery grant DP1097291 and the Capital Markets Cooperative Research Centre.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgements", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A comparison of loopy belief propagation and dual decomposition for integrated ccg supertagging and parsing", | |
| "authors": [ | |
| { | |
| "first": "Michael", | |
| "middle": [], | |
| "last": "Auli", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Adam", | |
| "middle": [], | |
| "last": "Lopez", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "49th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "470--480", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Michael Auli and Adam Lopez. 2011. A comparison of loopy belief propagation and dual decomposition for integrated ccg supertagging and parsing. In 49th Annual Meeting of the Association for Computational Linguistics, pages 470-480. Association for Computa- tional Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "On the parameter space of generative lexicalized statistical parsing models", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Daniel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Bikel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Daniel M. Bikel. 2004. On the parameter space of genera- tive lexicalized statistical parsing models. Ph.D. thesis, Citeseer.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Two statistical parsing models applied to the Chinese Treebank", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Daniel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Bikel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Chiang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Second workshop on Chinese language processing", | |
| "volume": "12", | |
| "issue": "", | |
| "pages": "1--6", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Daniel M. Bikel and David Chiang. 2000. Two statisti- cal parsing models applied to the Chinese Treebank. In Second workshop on Chinese language processing, vol- ume 12, pages 1-6. Morristown, NJ, USA.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Parser evaluation: a survey and a new proposal", | |
| "authors": [ | |
| { | |
| "first": "John", | |
| "middle": [], | |
| "last": "Carroll", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ted", | |
| "middle": [], | |
| "last": "Briscoe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Antonio", | |
| "middle": [], | |
| "last": "Sanfilippo", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Proceedings of the 1st International Conference on Language Resources and Evaluation", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "447--454", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "John Carroll, Ted Briscoe, and Antonio Sanfilippo. 1998. Parser evaluation: a survey and a new proposal. In Pro- ceedings of the 1st International Conference on Lan- guage Resources and Evaluation, pages 447-454.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "A grammar of spoken Chinese", | |
| "authors": [ | |
| { | |
| "first": "Yuen-Ren", | |
| "middle": [], | |
| "last": "Chao", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1968, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yuen-Ren Chao. 1968. A grammar of spoken Chinese. University of California Press.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Recovering latent information in treebanks", | |
| "authors": [ | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Chiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Daniel", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Bikel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Proceedings of the 19th international conference on Computational linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1--7", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "David Chiang and Daniel M. Bikel. 2002. Recovering la- tent information in treebanks. In Proceedings of the 19th international conference on Computational lin- guistics, volume 1, pages 1-7. Association for Compu- tational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "The importance of supertagging for wide-coverage CCG parsing", | |
| "authors": [ | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "James", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Curran", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the 20th international conference on Computational Linguistics. Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Stephen Clark and James R. Curran. 2004. The impor- tance of supertagging for wide-coverage CCG parsing. In Proceedings of the 20th international conference on Computational Linguistics. Association for Computa- tional Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Wide-Coverage Efficient Statistical Parsing with CCG and Log-Linear Models", | |
| "authors": [ | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "James", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Curran", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Computational Linguistics", | |
| "volume": "33", | |
| "issue": "", | |
| "pages": "493--552", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Stephen Clark and James R. Curran. 2007. Wide- Coverage Efficient Statistical Parsing with CCG and Log-Linear Models. In Computational Linguistics, vol- ume 33, pages 493-552.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Investigating GIS and smoothing for maximum entropy taggers", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "James", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Curran", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Proceedings of the 10th Meeting of the EACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "91--98", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "James R. Curran and Stephen Clark. 2003. Investigating GIS and smoothing for maximum entropy taggers. In Proceedings of the 10th Meeting of the EACL, pages 91-98. Budapest, Hungary.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Accurate context-free parsing with combinatory categorial grammar", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Timothy", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Gerald", | |
| "middle": [], | |
| "last": "Fowler", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Penn", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "335--344", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Timothy A.D. Fowler and Gerald Penn. 2010. Accu- rate context-free parsing with combinatory categorial grammar. Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 335-344.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Recovering non-local dependencies for Chinese", | |
| "authors": [ | |
| { | |
| "first": "Yuqing", | |
| "middle": [], | |
| "last": "Guo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Haifeng", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Josef", | |
| "middle": [], | |
| "last": "Van Genabith", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "EMNLP/CoNLL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "257--266", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yuqing Guo, Haifeng Wang, and Josef Van Genabith. 2007. Recovering non-local dependencies for Chinese. In EMNLP/CoNLL, pages 257-266.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Data and Models for Statistical Parsing with Combinatory Categorial Grammar", | |
| "authors": [ | |
| { | |
| "first": "Julia", | |
| "middle": [], | |
| "last": "Hockenmaier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Julia Hockenmaier. 2003. Data and Models for Statistical Parsing with Combinatory Categorial Grammar. Ph.D. thesis, University of Edinburgh.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "CCGbank: Users' manual", | |
| "authors": [ | |
| { | |
| "first": "Julia", | |
| "middle": [], | |
| "last": "Hockenmaier", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Steedman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Julia Hockenmaier and Mark Steedman. 2005. CCGbank: Users' manual. Technical report, MS-CIS-05-09, Com- puter and Information Science, University of Pennsyl- vania.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "CCGbank: A Corpus of CCG Derivations and Dependency Structures Extracted from the Penn Treebank", | |
| "authors": [ | |
| { | |
| "first": "Julia", | |
| "middle": [], | |
| "last": "Hockenmaier", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Steedman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Computational Linguistics", | |
| "volume": "33", | |
| "issue": "3", | |
| "pages": "355--396", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Julia Hockenmaier and Mark Steedman. 2007. CCGbank: A Corpus of CCG Derivations and Dependency Struc- tures Extracted from the Penn Treebank. Computa- tional Linguistics, 33(3):355-396.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Hat Categories: Representing Form and Function Simultaneously in Combinatory Categorial Grammar", | |
| "authors": [ | |
| { | |
| "first": "Matthew", | |
| "middle": [], | |
| "last": "Honnibal", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Matthew Honnibal. 2010. Hat Categories: Represent- ing Form and Function Simultaneously in Combinatory Categorial Grammar. Ph.D. thesis, University of Syd- ney.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Bilingually-constrained (monolingual) shift-reduce parsing", | |
| "authors": [ | |
| { | |
| "first": "Liang", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Wenbin", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Qun", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "3", | |
| "issue": "", | |
| "pages": "1222--1231", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Liang Huang, Wenbin Jiang, and Qun Liu. 2009. Bilingually-constrained (monolingual) shift-reduce parsing. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Process- ing, volume 3, pages 1222-1231. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Is it harder to parse Chinese, or the Chinese Treebank?", | |
| "authors": [ | |
| { | |
| "first": "Roger", | |
| "middle": [], | |
| "last": "Levy", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christopher", | |
| "middle": [], | |
| "last": "Manning", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "439--446", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Roger Levy and Christopher Manning. 2003. Is it harder to parse Chinese, or the Chinese Treebank? In Annual Meeting of the Association for Computational Linguis- tics, volume 1, pages 439-446. Morristown, NJ, USA.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Mandarin Chinese: A functional reference grammar", | |
| "authors": [ | |
| { | |
| "first": "Charles", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sandra", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Thompson", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Charles N. Li and Sandra A. Thompson. 1989. Mandarin Chinese: A functional reference grammar. University of California Press.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Building a Large Annotated Corpus of English: The Penn Treebank", | |
| "authors": [ | |
| { | |
| "first": "Mitchell", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Marcus", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Beatrice", | |
| "middle": [], | |
| "last": "Santorini", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mary", | |
| "middle": [ | |
| "Ann" | |
| ], | |
| "last": "Marcinkiewicz", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Computational Linguistics", | |
| "volume": "19", | |
| "issue": "2", | |
| "pages": "313--330", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1994. Building a Large Annotated Cor- pus of English: The Penn Treebank. Computational Linguistics, 19(2):313-330.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Effective self-training for parsing", | |
| "authors": [ | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Mcclosky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eugene", | |
| "middle": [], | |
| "last": "Charniak", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Johnson", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the main conference on Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "152--159", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "David McClosky, Eugene Charniak, and Mark Johnson. 2006. Effective self-training for parsing. In Proceed- ings of the main conference on Human Language Tech- nology Conference of the North American Chapter of the Association of Computational Linguistics, pages 152-159. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "Corpus-Oriented Grammar Development for Acquiring a Head-Driven Phrase Structure Grammar from the Penn Treebank", | |
| "authors": [ | |
| { | |
| "first": "Yusuke", | |
| "middle": [], | |
| "last": "Miyao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Takashi", | |
| "middle": [], | |
| "last": "Ninomiya", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "684--693", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yusuke Miyao, Takashi Ninomiya, and Jun'ichi Tsujii. 2004. Corpus-Oriented Grammar Development for Ac- quiring a Head-Driven Phrase Structure Grammar from the Penn Treebank. pages 684-693.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Improved inference for unlexicalized parsing", | |
| "authors": [ | |
| { | |
| "first": "Slav", | |
| "middle": [], | |
| "last": "Petrov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dan", | |
| "middle": [], | |
| "last": "Klein", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proceedings of NAACL HLT 2007", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "404--411", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Slav Petrov and Dan Klein. 2007. Improved inference for unlexicalized parsing. In Proceedings of NAACL HLT 2007, pages 404-411.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Unbounded dependency recovery for parser evaluation", | |
| "authors": [ | |
| { | |
| "first": "Laura", | |
| "middle": [], | |
| "last": "Rimell", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Steedman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "813--821", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Laura Rimell, Stephen Clark, and Mark Steedman. 2009. Unbounded dependency recovery for parser evaluation. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 2- Volume 2, pages 813-821. Association for Computa- tional Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "The Syntactic Process", | |
| "authors": [ | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Steedman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mark Steedman. 2000. The Syntactic Process. MIT Press. Cambridge, MA, USA.", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "Chinese CCGbank: extracting CCG derivations from the Penn Chinese Treebank", | |
| "authors": [ | |
| { | |
| "first": "Daniel", | |
| "middle": [], | |
| "last": "Tse", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "James", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Curran", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1083--1091", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Daniel Tse and James R. Curran. 2010. Chinese CCG- bank: extracting CCG derivations from the Penn Chi- nese Treebank. Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010), pages 1083-1091.", | |
| "links": null | |
| }, | |
| "BIBREF25": { | |
| "ref_id": "b25", | |
| "title": "A fast, accurate deterministic parser for Chinese", | |
| "authors": [ | |
| { | |
| "first": "Mengqiu", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kenji", | |
| "middle": [], | |
| "last": "Sagae", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Teruko", | |
| "middle": [], | |
| "last": "Mitamura", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "425--432", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mengqiu Wang, Kenji Sagae, and Teruko Mitamura. 2006. A fast, accurate deterministic parser for Chinese. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meet- ing of the Association for Computational Linguistics, pages 425-432. Association for Computational Lin- guistics.", | |
| "links": null | |
| }, | |
| "BIBREF26": { | |
| "ref_id": "b26", | |
| "title": "Extracting tree adjoining grammars from bracketed corpora", | |
| "authors": [ | |
| { | |
| "first": "Fei", | |
| "middle": [], | |
| "last": "Xia", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "Proceedings of Natural Language Processing Pacific Rim Symposium '99", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "398--403", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fei Xia. 1999. Extracting tree adjoining grammars from bracketed corpora. In Proceedings of Natural Lan- guage Processing Pacific Rim Symposium '99, pages 398-403.", | |
| "links": null | |
| }, | |
| "BIBREF27": { | |
| "ref_id": "b27", | |
| "title": "Comparing lexicalized treebank grammars extracted from Chinese, Korean, and English corpora", | |
| "authors": [ | |
| { | |
| "first": "Fei", | |
| "middle": [], | |
| "last": "Xia", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chung-Hye", | |
| "middle": [], | |
| "last": "Han", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aravind", | |
| "middle": [], | |
| "last": "Joshi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2000, | |
| "venue": "Proceedings of the second workshop on Chinese language processing: held in conjunction with the 38th Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "12", | |
| "issue": "", | |
| "pages": "52--59", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fei Xia, Chung-hye Han, Martha Palmer, and Aravind Joshi. 2000. Comparing lexicalized treebank grammars extracted from Chinese, Korean, and English corpora. In Proceedings of the second workshop on Chinese lan- guage processing: held in conjunction with the 38th Annual Meeting of the Association for Computational Linguistics, volume 12, pages 52-59. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF28": { | |
| "ref_id": "b28", | |
| "title": "The Penn Chinese TreeBank: Phrase structure annotation of a large corpus", | |
| "authors": [ | |
| { | |
| "first": "Nianwen", | |
| "middle": [], | |
| "last": "Xue", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Fei", | |
| "middle": [], | |
| "last": "Xia", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Fu-Dong", | |
| "middle": [], | |
| "last": "Chiou", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martha", | |
| "middle": [], | |
| "last": "Palmer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Natural Language Engineering", | |
| "volume": "11", | |
| "issue": "02", | |
| "pages": "207--238", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Nianwen Xue, Fei Xia, Fu-Dong Chiou, and Martha Palmer. 2005. The Penn Chinese TreeBank: Phrase structure annotation of a large corpus. Natural Lan- guage Engineering, 11(02):207-238.", | |
| "links": null | |
| }, | |
| "BIBREF29": { | |
| "ref_id": "b29", | |
| "title": "Analysis of the difficulties in chinese deep parsing", | |
| "authors": [ | |
| { | |
| "first": "Kun", | |
| "middle": [], | |
| "last": "Yu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yusuke", | |
| "middle": [], | |
| "last": "Miyao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Takuya", | |
| "middle": [], | |
| "last": "Matsuzaki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xiangli", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "12th International Conference on Parsing Technologies", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kun Yu, Yusuke Miyao, Takuya Matsuzaki, Xiangli Wang, and Jun'ichi Tsujii. 2011. Analysis of the dif- ficulties in chinese deep parsing. In 12th International Conference on Parsing Technologies, page 48.", | |
| "links": null | |
| }, | |
| "BIBREF30": { | |
| "ref_id": "b30", | |
| "title": "A tale of two parsers: investigating and combining graph-based and transition-based dependency parsing using beamsearch", | |
| "authors": [ | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "562--571", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yue Zhang and Stephen Clark. 2008. A tale of two parsers: investigating and combining graph-based and transition-based dependency parsing using beam- search. In Proceedings of the Conference on Empir- ical Methods in Natural Language Processing, pages 562-571. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF31": { | |
| "ref_id": "b31", | |
| "title": "Transition-based parsing of the Chinese treebank using a global discriminative model", | |
| "authors": [ | |
| { | |
| "first": "Yue", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephen", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 11th International Conference on Parsing Technologies", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "162--171", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yue Zhang and Stephen Clark. 2009. Transition-based parsing of the Chinese treebank using a global discrim- inative model. In Proceedings of the 11th International Conference on Parsing Technologies, pages 162-171. Association for Computational Linguistics.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "uris": null, | |
| "text": "Two types of ambiguity", | |
| "type_str": "figure" | |
| }, | |
| "TABREF0": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Fowler and Penn (2010) use the C&C tool generate to convert P&K output to the C&C evaluation dependency format. generate critically does not depend on the C&C parsing model, permitting a fair comparison of the parsers' output.", | |
| "content": "<table><tr><td>\uf770\uf763\uf774\uf762 5</td><td>+\uf770\uf763\uf774\uf762 6</td><td>#sents</td></tr><tr><td colspan=\"3\">Train Test 816-885, 1137-1147 3030-3145 1-815, 1001-1136 2000-2980 22033 2758 Dev 900-931, 1148-1151 2981-3029 1101</td></tr></table>" | |
| }, | |
| "TABREF1": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "\uf770\uf763\uf774\uf762 5 and 6 dev/train/test splits", | |
| "content": "<table><tr><td>4.1 Evaluation</td></tr><tr><td>Carroll et al. (1998) argued against \uf770\uf761\uf772\uf773\uf765\uf776\uf761\uf76c in favour of a dependency-based evaluation. Rimell et al. (2009) focus on evaluating \uf76e\uf76c\uf764 recovery, proposing a dependency-based evaluation and a \uf767\uf772</td></tr></table>" | |
| }, | |
| "TABREF2": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Shapes of categories", | |
| "content": "<table><tr><td/><td>model</td><td>LF Lsa %</td><td>stag cov log C</td></tr><tr><td>A B C</td><td colspan=\"3\">I-3 68.97 I-6 71.67 \uf767\uf76f\uf76c\uf764 75.45 16.70 89.43 99.4 14.55 13.45 83.64 95.7 -15.70 85.00 96.4 -\uf761\uf775\uf774\uf76f 66.32 12.81 83.88 98.6 14.69 I-3 69.75 14.15 84.07 96.0 -I-5 71.40 14.83 84.97 96.4 -\uf767\uf76f\uf76c\uf764 75.41 16.67 89.50 99.6 14.74 \uf761\uf775\uf774\uf76f 66.24 12.61 83.95 98.7 14.75 I-3 70.22 16.49 84.37 96.5 -I-5 72.74 18.59 85.61 96.5 -\uf767\uf76f\uf76c\uf764 76.73 20.56 89.66 99.5 13.58 \uf761\uf775\uf774\uf76f 66.95 14.62 83.90 99.2 13.86</td></tr></table>" | |
| }, | |
| "TABREF3": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "", | |
| "content": "<table/>" | |
| }, | |
| "TABREF4": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "The \u223c9% drop in F -score between the \uf767\uf76f\uf76c\uf764 and \uf761\uf775\uf774\uf76f figures shows that C&C is highly sensitive to \uf770\uf76f\uf773 tagging accuracy (92.56% on the dev set, compared to 96.82% on English). ConsideringTable 4, each best P&K model outperforms the corresponding \uf761\uf775\uf774\uf76f model by 3-5%. However, while P&K is substantially better without gold-standard information, gold \uf770\uf76f\uf773 tags allow C&C to outperform P&K, again", | |
| "content": "<table><tr><td/><td>model</td><td>LF Lsa %</td><td>stag</td><td>cov</td></tr><tr><td>A B C</td><td colspan=\"4\">I-6 71.74 \uf761\uf775\uf774\uf76f 67.50 15.36 84.52 100.0 15.87 85.29 100.0 I-5 71.40 14.97 85.26 100.0 \uf761\uf775\uf774\uf76f 67.72 14.97 84.68 100.0 I-5 72.84 18.69 86.04 100.0 \uf761\uf775\uf774\uf76f 68.43 16.17 84.57 100.0</td></tr><tr><td colspan=\"5\">Table 4: Dev set evaluation for P&K and C&C on \uf770\uf763\uf774\uf762 6 sentences parsed by both parsers</td></tr><tr><td colspan=\"2\">model</td><td>LF Lsa %</td><td colspan=\"2\">stag cov log C</td></tr><tr><td>C</td><td colspan=\"4\">I-5 72.73 \uf767\uf76f\uf76c\uf764 76.89 22.90 89.63 99.1 14.53 20.28 85.43 97.1 -\uf761\uf775\uf774\uf76f 67.09 15.28 83.95 98.7 14.89</td></tr></table>" | |
| }, | |
| "TABREF5": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Test set evaluation for P&K and C&C demonstrating the impact of incorrect \uf770\uf76f\uf773 tags.", | |
| "content": "<table/>" | |
| }, | |
| "TABREF6": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "shows that the behaviour of both parsers on the test section is consistent with the dev section.", | |
| "content": "<table><tr><td>corpus</td><td>Avg. tags/word</td><td colspan=\"2\">Grammar size all f \u2265 10</td></tr><tr><td>A B C</td><td>1.84 1.83 1.79</td><td>1177 1084 964</td><td>324 303 274</td></tr></table>" | |
| }, | |
| "TABREF7": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "", | |
| "content": "<table><tr><td>: Corpus statistics</td></tr></table>" | |
| }, | |
| "TABREF8": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "To quantify C&C's sensitivity to \uf770\uf76f\uf773 tagging errors, NR \u25b7\u25c1 NN 76.72 -0.01 89.64 99.37 JJ \u25b7\u25c1 NN 76.60 -0.12 89.57 99.37 DEC \u25b7\u25c1 DEG 75.10 -1.50 89.07 98.83 VV \u25b7\u25c1 NN 73.35 -1.75 87.68 98.74", | |
| "content": "<table><tr><td>Confusion</td><td>LF \u2206LF</td><td>stag</td><td>cov</td></tr><tr><td colspan=\"2\">Base (\uf767\uf76f\uf76c\uf764) 76.73</td><td colspan=\"2\">89.66 99.50</td></tr><tr><td colspan=\"2\">All (\uf761\uf775\uf774\uf76f) 66.95</td><td colspan=\"2\">83.90 99.20</td></tr></table>" | |
| }, | |
| "TABREF9": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Corrupting C&C gold \uf770\uf76f\uf773 tags piecemeal on \uf770\uf763\uf774\uf762 6 dev set of corpus C. \u2206LF is the change in LF when each additional confusion type is allowed.", | |
| "content": "<table/>" | |
| }, | |
| "TABREF10": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "breaks down the 8,414 false positives generated by C&C on the dev set, according to whether the head of each dependency was incorrectly \uf770\uf76f\uf773-tagged and/or supertagged. The top-left cell shows that despite the correct \uf770\uf76f\uf773 and supertag, C&C makes a large number of pure attachment location errors. The vast majority of false positives, though, are", | |
| "content": "<table><tr><td>C&C \uf761\uf775\uf774\uf76f LF freq LF P&K I-5 freq</td><td>category</td><td colspan=\"2\">\uf76e\uf76c\uf764? dependency function</td></tr><tr><td colspan=\"3\">0.78 4204 0.78 3106 NP/NP 0.73 2173 0.81 1765 (S[dcl]\\NP)/NP 0.65 1717 0.72 1459 (S[dcl]\\NP)/NP 0.68 870 0.74 643 (S[dcl]\\NP)/(S[dcl]\\NP) 0.70 862 0.67 697 S[dcl]\\NP 0.60 670 0.69 499 (S[dcl]\\NP)/(S[dcl]\\NP) \u2713 0.55 626 0.54 412 (NP/NP)/(NP/NP) 0.57 370 0.68 321 (NP/NP)\\(S[dcl]\\NP) 0.59 343 0.70 314 (NP/NP)\\(S[dcl]\\NP) \u2713 0.59 110 0.69 84 (NP/NP)\\(S[dcl]/NP) 0.63 106 0.75 86 (NP/NP)\\(S[dcl]/NP) \u2713</td><td>noun modifier attachment transitive object transitive subject control/raising S complement intransitive subject control/raising subject noun modifier modifier attachment subject extraction S complement subject extraction modifier attachment object extraction S complement object extraction modifier attachment</td></tr></table>" | |
| }, | |
| "TABREF11": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "", | |
| "content": "<table><tr><td colspan=\"2\">: Accuracy per dependency, for selected dependency types</td></tr><tr><td>correct \uf770\uf76f\uf773</td><td>incorrect \uf770\uf76f\uf773</td></tr><tr><td colspan=\"2\">correct stag 2307 (27.42%) incorrect stag 4493 (53.40%) 1563 (18.58%) 51 (0.61% )</td></tr></table>" | |
| }, | |
| "TABREF12": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Analysis of the 8,414 false positive dependencies from C&C on \uf770\uf763\uf774\uf762 6 dev set caused by supertagging errors (the bottom row), but most of these are not a result of incorrect \uf770\uf76f\uf773 tags, demonstrating that supertagging and parsing are difficult even with correct \uf770\uf76f\uf773 tags.", | |
| "content": "<table/>" | |
| }, | |
| "TABREF13": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Summary of Chinese parsing approaches", | |
| "content": "<table><tr><td/><td>model</td><td colspan=\"2\">LF Lsa %</td><td>stag cov log C</td></tr><tr><td>C</td><td colspan=\"2\">\uf767\uf76f\uf76c\uf764 74.99 (76.73 \uf761\uf775\uf774\uf76f 65.42 (66.95 I-5 70.67 (72.74</td><td>7.42 89.36 98.6 18.35 20.56 89.66 99.5 13.58) 4.82 83.73 97.9 18.67 14.62 83.90 99.2 13.86) 8.62 84.99 93.8 -18.59 85.61 96.5 -)</td></tr></table>" | |
| }, | |
| "TABREF14": { | |
| "type_str": "table", | |
| "html": null, | |
| "num": null, | |
| "text": "Dev set evaluation for C&C over pro-drop", | |
| "content": "<table><tr><td colspan=\"2\">sentences only (and over full set in parentheses)</td></tr><tr><td>pro come out b. pro \u51fa\u6765</td><td>\u7684</td></tr></table>" | |
| } | |
| } | |
| } | |
| } |