{ "paper_id": "O08-1002", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T08:02:36.959534Z" }, "title": "A Semantic Composition Method for Deriving Sense Representations of Determinative-Measure Compounds in E-HowNet", "authors": [ { "first": "Chia-Hung", "middle": [], "last": "Tai", "suffix": "", "affiliation": { "laboratory": "", "institution": "Academia Sinica glaxy", "location": {} }, "email": "" }, { "first": "Shu-Ling", "middle": [], "last": "Huang", "suffix": "", "affiliation": { "laboratory": "", "institution": "Academia Sinica glaxy", "location": {} }, "email": "" }, { "first": "Keh-Jiann", "middle": [], "last": "Chen", "suffix": "", "affiliation": { "laboratory": "", "institution": "Academia Sinica glaxy", "location": {} }, "email": "kchen@iis.sinica.edu.tw" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "In this paper, we take Determinative-Measure Compounds as an example to demonstrate how the E-HowNet semantic composition mechanism works in deriving the sense representations for all determinative-measure (DM) compounds which is an open set. We define the sense of a closed set of each individual determinative and measure word in E-HowNet representation exhaustively. We then make semantic composition rules to produce candidate sense representations for any newly coined DM. Then we review development set to design sense disambiguation rules. We use these heuristic disambiguation rules to determine the correct context-dependent sense of a DM and its E-HowNet representation. The experiment shows that the current model reaches 88% accuracy in DM identification and sense derivation.", "pdf_parse": { "paper_id": "O08-1002", "_pdf_hash": "", "abstract": [ { "text": "In this paper, we take Determinative-Measure Compounds as an example to demonstrate how the E-HowNet semantic composition mechanism works in deriving the sense representations for all determinative-measure (DM) compounds which is an open set. We define the sense of a closed set of each individual determinative and measure word in E-HowNet representation exhaustively. We then make semantic composition rules to produce candidate sense representations for any newly coined DM. Then we review development set to design sense disambiguation rules. We use these heuristic disambiguation rules to determine the correct context-dependent sense of a DM and its E-HowNet representation. The experiment shows that the current model reaches 88% accuracy in DM identification and sense derivation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Building knowledge base is a time consuming work. The CKIP Chinese Lexical Knowledge Base has about 80 thousand lexical entries and their senses are defined in terms of the E-HowNet format. E-HowNet is a lexical knowledge and common sense knowledge representation system. It was extended from HowNet [1] to encode concepts. Based on the framework of E-HowNet, we intend to establish an automatic semantic composition mechanism to derive sense of compounds and phrases from lexical senses [2] [3] .", "cite_spans": [ { "start": 300, "end": 303, "text": "[1]", "ref_id": "BIBREF0" }, { "start": 492, "end": 495, "text": "[3]", "ref_id": "BIBREF1" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Determinative-Measure compounds (abbreviated as DM) are most common compounds in Chinese. Because a determinative and a measure normally coin a compound with unlimited versatility, the CKIP group does not define the E-HowNet representations for all DM compounds. Although the demonstrative, numerals, and measures may be listed exhaustively, their combination is inexhaustible. However their constructions are regular [4] . Therefore, an automatic identification schema in regular expression [4] and a semantic composition method under the framework of E-HowNet for DM compounds were developed.", "cite_spans": [ { "start": 418, "end": 421, "text": "[4]", "ref_id": "BIBREF2" }, { "start": 492, "end": 495, "text": "[4]", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "In this paper, we take DMs as an example to demonstrate how the E-HowNet semantic composition mechanism works in deriving the sense representations for all DM compounds.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "The remainder of this paper is organized as follows. The section 2 presents the background knowledge of DM compounds and sense representation in E-HowNet. We'll describe our method in the section 3 and discuss the experiment result in the section 4 before we make conclusion in the section 5.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "There are numerous studies on determinatives as well as measures, especially on the types of measures. 1 Tai [5] asserts that in the literature on general grammar as well as Chinese grammar, classifiers and measures words are often treated together under one single framework of analysis. Chao [6] treats classifiers as one kind of measures. In his definition, a measure is a bound morpheme which forms a DM compound with the determinatives enumerated below.", "cite_spans": [ { "start": 103, "end": 104, "text": "1", "ref_id": "BIBREF0" }, { "start": 109, "end": 112, "text": "[5]", "ref_id": "BIBREF3" }, { "start": 294, "end": 297, "text": "[6]", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "He also divides determinatives word into four subclasses: i. Demonstrative determinatives, e.g. \u9019\" this\", that\"\u90a3\"\u2026", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "ii. Specifying determinatives, e.g. \u6bcf\"every\", \u5404\" each\"\u2026", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "iii. Numeral determinatives, e.g. \u4e8c\"two\", \u767e\u5206\u4e4b\u4e09\"three percentage\", \u56db\u767e\u4e94\u5341\" four hundred and fifty\"\u2026 iv. Quantitative determinatives, e.g. \u4e00\" one\", \u6eff\" full\", \u8a31\u591a\" many\"\u2026 Measures are divided into nine classes by Chao [6] . Classifiers are defined as 'individual measures', which is one of the nine kinds of measures.", "cite_spans": [ { "start": 212, "end": 215, "text": "[6]", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "i. classifiers, e.g. \u672c\"a (book)\",", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "ii. classifier associated with V-O constructions, e.g. \u624b \"hand\",", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "iii. group measures, e.g. \u5c0d\"pair\", iv. partitive measures, e.g. \u4e9b\"some\", v. container measures, e.g. \u76d2\"box\", vi. temporary measures, e.g. \u8eab\"body\", vii. Standard measures, e.g. \u516c\u5c3a\"meter\", viii. quasi-measure, e.g. \u570b\"country\", ix. Measures with verb, e.g. \u6b21\"number of times\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "As we mentioned in the section of introduction, Chao considers that determinatives are listable and measures are largely listable, so D and M can be defined by enumeration, and that DM compounds have unlimited versatility. However, Li and Thompson [7] blend classifiers with measures. They conclude not only does a measure word generally not take a classifier, but any measure word can be a classifier. In Tai's opinion [5] , in order to better understand the nature of categorization in a classifier system, it is not only desirable but also necessary to differentiate classifiers from measure words. These studies on the distinction between classifiers and measures are not very clear-cut. In this paper, we adopt the CKIP DM rule patterns and Part-of-Speeches for morpho-syntactic analysis, and therefore inherit the definition of determinative-measure compounds (DMs) in [10] . Mo et al. define a DM as the composition of one or more determinatives together with an optional measure. It is used to determine the reference or the quantity of the noun phrase that co-occurs with it. We use the definition of Mo et al. to apply to NLP and somewhat different from traditional linguistics definitions.", "cite_spans": [ { "start": 248, "end": 251, "text": "[7]", "ref_id": "BIBREF5" }, { "start": 420, "end": 423, "text": "[5]", "ref_id": "BIBREF3" }, { "start": 875, "end": 879, "text": "[10]", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2." }, { "text": "Due to the infinite of the number of possible DMs, Mo et al. [10] and Li et al. [4] propose to identify DMs by regular expression before parsing as part of their morphological module in NLP. For example, when the DM compound is the composition of one determinative, e.g.", "cite_spans": [ { "start": 61, "end": 65, "text": "[10]", "ref_id": "BIBREF8" }, { "start": 80, "end": 83, "text": "[4]", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "for numerals in (1), roughly rules (2a), (2b) or (2c) will be first applied, and then rules (2d), NO1 = {\u25cb,\u4e00,\u4e8c,\u5169,\u4e09,\u56db,\u4e94,\u516d,\u4e03,\u516b,\u4e5d,\u5341,\u5eff,\u5345,\u767e,\u5343,\u842c,\u5104,\u5146,\u96f6, \u5e7e}; b. NO2 = {\u58f9,\u8cb3,\u53c3,\u8086,\u4f0d,\u9678,\u67d2,\u634c,\u7396,\u62fe,\u4f70,\u4edf,\u842c,\u5104,\u5146,\u96f6,\u5e7e}; c. NO3 = {\uff11,\uff12,\uff13,\uff14,\uff15,\uff16,\uff17,\uff18,\uff19,\uff10,\u767e,\u5343,\u842c,\u5104,\u5146}; The major semantic roles played by determinatives and measures are listed in the Table 1 .", "cite_spans": [], "ref_spans": [ { "start": 98, "end": 201, "text": "NO1 = {\u25cb,\u4e00,\u4e8c,\u5169,\u4e09,\u56db,\u4e94,\u516d,\u4e03,\u516b,\u4e5d,\u5341,\u5eff,\u5345,\u767e,\u5343,\u842c,\u5104,\u5146,\u96f6, \u5e7e}; b. NO2 = {\u58f9,\u8cb3,\u53c3,\u8086,\u4f0d,\u9678,\u67d2,\u634c,\u7396,\u62fe,\u4f70,\u4edf,\u842c,\u5104,\u5146,\u96f6,\u5e7e};", "ref_id": "TABREF4" }, { "start": 325, "end": 332, "text": "Table 1", "ref_id": null } ], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "The basic feature unification processes [12] :", "cite_spans": [ { "start": 40, "end": 44, "text": "[12]", "ref_id": "BIBREF10" } ], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "If a morpheme B is a dependency daughter of morpheme A, i.e. B is a modifier or an argument of A, then unify the semantic representation of A and B by the following steps.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Step 1: Identify semantic relation between A and B to derive relation(A)={B}. Note: the possible semantic relations are shown in Table 1 .", "cite_spans": [], "ref_spans": [ { "start": 129, "end": 136, "text": "Table 1", "ref_id": null } ], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Step 2: Unify the semantic representation of A and B by insert relation(A)={B} as a sub-feature of A.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "It seems that a feature unification process can derive the sense representation of a DM compound, as exemplified in (7) and 8 Figure 1 . The dependent relations of \u4e00\u7897\u9eb5\"a bowl of noddle\".", "cite_spans": [ { "start": 116, "end": 119, "text": "(7)", "ref_id": "BIBREF5" } ], "ref_spans": [ { "start": 126, "end": 134, "text": "Figure 1", "ref_id": null } ], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "After feature unification process, the semantic representation of \"\u4e00 def: quantity={1}\" becomes the feature of its dependent head \"\u7897 def: container={bowl|\u7897} and derives the feature representation of \"one bowl \u4e00\u7897 def: container={bowl| \u7897 :quantity={1}}\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Similarly, \"one bowl \u4e00\u7897\" is the dependent daughter of \"noodle|\u9eb5 def:{noodle|\u9eb5}\". After unification process, we derive the result of (9). (9)one bowl of noodle|\u4e00\u7897\u9eb5 def:{noodle|\u9eb5:container={bowl|\u7897:quantity={1}}}", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "The above feature unification process written in term of rule is expressed as (10). If a DM has more than one determinative, we can consider the consecutive determinatives as one D and the feature representation of D is a coordinate conjunction of the features of all its determinatives. For instance, \"this one|\u9019\u4e00\" and \"this one|\u9019\u4e00\u672c\" both are expressed as \"quantifier={definite|\u5b9a\u6307}; quantity={1}\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Omissions of numeral determinative are occurred very often while the numeral quantity is \"1\". For instance, \"\u9019\u672c\" in fact means \"this one|\u9019\u4e00\u672c\". Therefore the definition of (8)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "should be modified as: \u9019\u672c def: quantifier={definite|\u5b9a\u6307}; quantity={1};", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "The following derivation rules cover the cases of omissions of numeral determinative.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "(12) If both numeral and quantitative determinatives do not occur in a DM, then the feature quantity={1} is the default value of the DM.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Another major complication is that senses of morphemes are ambiguous. The feature unification process may produce many sense representations for a DM compound. Therefore sense disambiguation is needed and the detail discussions will be in the section 3.1.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Members of every type of determinatives and measures are exhaustively listable except numeral determinatives. Also the formats of numerals are various. For example, \"5020\" is equal to \"\u4e94\u96f6\u4e8c\u96f6\" and \"\u4e94\u5343\u96f6\u4e8c\u5341\" and \"\u4e94\u5343\u4e8c\u5341\". So we have to unify the numeral representation into a standard form. All numerals are composition of basic numeral as shown in the regular expressions (2). However their senses are not possible to define one by one. We take a simple approach. For all numeral, their E-HowNet sense representations are expressed as themselves. For example, is expresses as quantity={5020} and will not further define what is the sense of 5020. Furthermore all non-Arabic forms will be convert into Arabic expression, e.g. \"\u4e94\u5343\u96f6\u4e8c\u5341\" is defined as quantity={5020}.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "The other problem is that the morphological structures of some DMs are not regular patterns.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Take \"\u5169\u500b\u534a two and half\" as an example. \"\u534a half\" is not a measure word. So we collect those word like \"\u591a many, \u534a half, \u5e7e many, \u4e0a up, \u5927 big, \u4f86 more\" for modify the quantity definition. So we first remove the word \"\u534a\" and define the \"\u5169\u500b\" as quantity={2}.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Because the word \"\u534a\" means quantity={0.5}, we define the E-HowNet definition for \"\u5169\u500b \u534a\" as quantity={2.5}. For other modifiers such as \"\u591a many, \u5e7e many, \u9918 more, \u4f86 more\", we use a function over() to represent the sense of \"more\", such as \"\u5341\u591a\u500b more than 10\" is represented as quantity={over 10 For an input Chinese sentence, we use the regular expression rules created by Li et al. [2006] to identify all possible DMs in the input sentence. Then, for every DM compound, we segment it into a sequence of determinatives and measures. If any numeral exists in the DM, every numeral is converted into decimal number in Arabic form. For every DM, we follow the feature unification principles to composite semantics of DM in E-HowNet representations and produce possible ambiguous candidates. The final step of sense disambiguation is described in the following section.", "cite_spans": [ { "start": 369, "end": 385, "text": "Li et al. [2006]", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Regular Expression Approach for Identifying DMs", "sec_num": "2.1" }, { "text": "Multiple senses will be derived for a DM compound due to ambiguous senses of its morpheme components. For instance, the measure word \"\u982d head\" has either the sense of {\u982d|head}, such as \"\u6eff\u982d\u767d\u9aee full head of white hairs\" or the null sense in \"\u4e00\u982d\u725b a cow\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "Some DMs are inherent sense ambiguous and some are pseudo ambiguous. For instances, the above example \"\u4e00\u982d\" is inherent ambiguous, since it could mean \"full head\" as in the example of \"\u4e00\u982d\u767d\u9aee full head of white hairs\" or could mean \"one + classifier\" as in the example of \"\u4e00\u982d\u725b a cow\". For inherent ambiguous DMs, the sense derivation step will produce ambiguous sense representations and leave the final sense disambiguation until seeing collocation context, in particular seeing dependent heads. Some ambiguous representations are improbable sense combination. The improbable sense combinations should be eliminated during or after feature unification of D and M. For instance, although the determiner \"\u4e00\" has ambiguous senses of \"one\", \"first\", and \"whole\", but \"\u4e00\u516c\u5c3a\" has only one sense of \"one meter\", so the other sense combinations should be eliminated.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "The way we tackle the problem is that first we find all the ambiguous Ds and Ms by looking their definitions shown in the appendix A. We then manually design content and context dependent rules to eliminate the improbable combinations for each ambiguous D or M types. For instance, according to the appendix A, \"\u982d\" has 3 different E-HowNet representations while functions as determinant or measure, i.e. \"def:{null}\", \"def:{head| \u982d }\", and \"def:ordinal={1}\". We write 3 content or context dependent rules below to disambiguate its senses.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "(13) \u982d\"head\", Nfa, E-howNet: \"def:{null}\" : while E-HowNet of head word is \"\u52d5 \u7269({animate|\u751f\u7269}\" and it's subclass.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "(14) \u982d\"head\", Nff, E-howNet: \"def:{\u982d}\" : while pre-determinant is \u4e00(Neqa)\"one\" or \u6eff\"full\" or \u5168\"all\" or \u6574\"total\". (15) \u982d\"first\", Nes, E-howNet: \"def:ordinal={1}\" : while this word is being a demonstrative determinatives which is a leading morpheme of the compound.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "The disambiguation rules are shown in appendix B. In each rule, the first part is the word and its part-of-speech. Then the E-HowNet definition of this sense is shown, and followed by the condition constraints for this sense. If there is still ambiguities remained after using the disambiguation rule, we choice the most frequent sense as the result.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Sense Disambiguation", "sec_num": "3.1" }, { "text": "We want to know how good is our candidate production, and how good is our disambiguation rule. We randomly select 40628 sentences (7536 DM words) from Sinica Treebank as our development set and 16070 sentences (3753 DM words) as our testing set. We use development set for designing disambiguation rules and semantic composition rules. Finally, we derive 36 contextual dependent rules as our disambiguation rules. We randomly select 1000 DM words from testing set. We evaluate the composition quality of DMs with E-HowNet representation before disambiguation. For 1000 DM words, the semantic composition rules produce 1226 candidates of E-HowNet representation from 939 words. The program fails to produce E-HowNet representations for the rest of 61 words because of undefined morphemes. There are 162 words out of the 939 words having ambiguous senses. The result shows that the quality of candidates is pretty good. Table 2 gives After data analysis, we conclude the following three kinds of error types.", "cite_spans": [], "ref_spans": [ { "start": 918, "end": 931, "text": "Table 2 gives", "ref_id": "TABREF4" } ], "eq_spans": [], "section": "Experiment and Discussion", "sec_num": "4." }, { "text": "\u4e03\u68d2\"7 th batter\", \u4e03\u5c40\"7 th inning\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "A. Unknown domain error:", "sec_num": null }, { "text": "Because there is no text related to baseball domain in development set, we get poor performance in dealing with the text about baseball. The way to resolve this problem is to increase the coverage of disambiguation rules for the baseball domain.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "A. Unknown domain error:", "sec_num": null }, { "text": "\u6bcf\u4e09\u500b\"each three\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "B. Undefined senses and morphemes:", "sec_num": null }, { "text": "We do not define the sense of \u6bcf \"each\" and we only define \u6bcf \"all\", so we have to add the sense of \"each\" in E-HowNet representation about \u6bcf. \u6709\u4e09\u4f4d \"there are three persons\", \u540c\u4e00\u500b \"the same\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "B. Undefined senses and morphemes:", "sec_num": null }, { "text": "Because \u6709 \"have\" and \u540c \"the same\" do not appear in our determinative list, it is not possible to composite their E-HowNet definitions.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "B. Undefined senses and morphemes:", "sec_num": null }, { "text": "In parsed sentence: NP(property:DM:\u4e0a\u534a\u5834\"first half \"|Head:DM:\u4e8c\u5341\u5206\"twenty minutes or twenty points\") . The E-HowNet representation of \u4e8c\u5341\u5206\"twenty minutes or twenty points\" can be defined as \"def:role={\u5206\u6578:quantity={20}}\" or \"def:time={\u5206 \u9418:quantity={20}}\". More context information is needed to resolve this kind of sense ambiguity.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "C. Sense ambiguities:", "sec_num": null }, { "text": "For unknown domain error and undefined rule, the solution is to expand the disambiguation rule set and sense definitions for morphemes. For sense ambiguities, we need more information to disambiguate the true sense.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "C. Sense ambiguities:", "sec_num": null }, { "text": "E-HowNet is a lexical sense representational framework and intends to achieve sense representation for all compounds, phrases, and sentences through automatic semantic composition processing. In this paper, we take DMs as an example to demonstrate how the semantic composition mechanism works in E-HowNet to derive the sense representations for", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "all DM compounds. We analyze morphological structures of DMs and derive their morphological rules in terms of regular expression. Then we define the sense of all determinatives and measure words in E-HowNet definition exhaustively. We make some simple composition rules to produce candidate sense representations for DMs. Then we review development set to write some disambiguation rules. We use these heuristic rules to find the final E-HowNet representation and reach 88% accuracy.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "The major target of E-HowNet is to achieve semantic composition. For this purpose, we defined word senses of CKIP lexicon in E-HowNet representation. Then we try to automate semantic composition for phrases and sentences. However there are many unknown or compound words without sense definitions in the target sentences. DM compounds are occurring most frequently and without sense definitions. Therefore our first step is to derive the senses of DM words. In the future, we will use similar methods to handle general compounds and to improve sense disambiguation and semantic relation identification processing. We intend to achieve semantic compositions for phrases and sentences in the future and we had shown the potential in this paper. e.g.\u4e00, Neu, def:quantity={1}, while part-of-speech of head word is Na, except the measure word is \u8eab \"body\" or \u81c9 \"face\" or \u9f3b \u5b50 \"nose\" or \u5634 \"mouth\" or \u809a \u5b50\"belly\" or \u8154\"cavity\" . e.g.\u584a,Nfg,def:role={money|\u8ca8\u5e63}, while E-HowNet representation of head word is \"{money|\u8ca8\u5e63}\" or {null}, or head word is \u9322\"money\" or \u7f8e\u91d1\"dollar\" or the s uffix of word is \u5e63\"currency\" and previous word is not D1.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "\u584a,Nfd,def:{null}, otherwise, use this definition.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u9762,Nfa,def:{null}, while part-of-speech of head word is Nab.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "\u9762,Nfh,def:direction={aspect|\u5074}, otherwise use this one.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u982d,Nfa,def:{null}, while head word is Nab and E-HowNet representation of head word is \"\u52d5\u7269{animate|\u751f\u7269}\" and it's subclass.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "\u982d,Nfh,def:direction={EndPosition|\u7aef} , if part-of-speech of head word is Na, do not use this definition. The previous word usually are \u9019 \"this\" or \u90a3 \"that\" or \u53e6\"another\". e.g.\u5206,Nfg,def:time={minute|\u5206\u9418}, if the sentence contains the word \u6642\"hour\" or \u9418 \u982d\"hour\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u5169,Nfg,def:weight={\u5169}, if the sentence contains the word \u91cd\"weight\" or \u91cd \u91cf\"weight\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "\u5169,Nfg,def:role={money|\u8ca8\u5e63}, if the sentence contains the word \u9280\"sliver\" or \u9322\"money\" or \u9ec3\u91d1\"gold\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "pre-determinant-based rule e.g.\u982d, Nff,def:{head|\u982d}, while pre-determinant is \u4e00(Neqa)\"one\" or \u6eff\"full\" or \u5168\"all\" or \u6574\"total\". e.g.\u8173, Nff,def:{leg|\u8173}, while pre-determinant is \u4e00(Neqa)\"one\" or \u6eff\"full\" or \u5168 \"all\" or \u6574\"total\" and part-of-speech of head word is not Na.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "\u8173, Nfi,def:frequency={}, while part-of-speech combination is V+D4,D15+\u8173.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u9ede,Nfg, def:time={ \u9ede }, while part-of-speech of pre-determinant is D4 or D15(1~24) and part-of-speech of previous word is not D1 or previous word is not \u6709\"have\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u8f2a,Nfg,def:time={\u8f2a}, while pre-determinant is \u7b2c \" a function word placed in front of a cardinal number to form an ordinal number\" or \u9996\"first\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "e.g.\u4e00\u3001\u4e8c...1\u30012...\u5169..., Neu, def:ordinal={}, the determinant of word is \u7b2c, \u6c11\u570b, \u516c\u5143, \u897f\u5143, \u5e74\u865f, \u4e00\u4e5d XX or 12XX, (four digits number). \u4e00\u3001\u4e8c...1\u30012...\u5169..., Neu,def:quantity={}, otherwise use this definition. e.g.\u982d,Nes,def:ordinal={1},the word \u982d\"head\" is determinant word. e.g.\u5169,Neu,def:quantity={}, the word \u5169\"a unit of weight equal to 50 grams\" is determinant word. measure word based rule e.g.\u4e00,Neqa,def:quantity={all|\u5168}, the part-of-speech of the measure word behind \u4e00 is Nff, or the suffix of the measure word is \u5b50, (for example,\u6ac3\u5b50\" cabinet\", \u74f6 \u5b50\"bottle\")or \u7c6e\u7b50\" large basket\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "determinative-based rule", "sec_num": null }, { "text": "Chao [6] and Li and Thompson[7] detect measures and classifiers. He[8] traces the diachronic names of measures and mentions related literature on measures. The dictionary of measures pressed by Mandarin Daily News Association and CKIP[9] lists all the possible measures in Mandarin Chinese.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The symbol \"Neu\" stands for Numeral Determinatives. Generation rules for numerals are partially listed in (2).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "This research was supported in part by the National Science Council under a Center Excellence Grant NSC 96-2752-E-001-001-PAE and Grant NSC96-2221-E-001-009.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgement:", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "HowNet and the Computation of Meaning", "authors": [], "year": 2006, "venue": "Zhendong Don & Qiang Dong", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Zhendong Don & Qiang Dong, 2006, HowNet and the Computation of Meaning. World Scientific Publishing Co. Pte. Ltd.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "E-HowNet-an Expansion of HowNet, The First National HowNet Workshop", "authors": [ { "first": "Shu-Ling", "middle": [], "last": "Huang", "suffix": "" }, { "first": "You-Shan", "middle": [], "last": "Chung", "suffix": "" }, { "first": "Keh-Jiann", "middle": [], "last": "Chen", "suffix": "" } ], "year": 2008, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shu-Ling Huang, You-Shan Chung, Keh-Jiann Chen, 2008, E-HowNet-an Expansion of HowNet, The First National HowNet Workshop, Beijing, China.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "A Probe into Ambiguities of Determinative-Measure Compounds", "authors": [ { "first": "", "middle": [], "last": "Li", "suffix": "" }, { "first": "Su-Chu", "middle": [], "last": "Shih-Min", "suffix": "" }, { "first": "", "middle": [], "last": "Lin", "suffix": "" } ], "year": 2006, "venue": "International Journal of Computational Linguistics & Chinese Language Processing", "volume": "11", "issue": "3", "pages": "245--280", "other_ids": {}, "num": null, "urls": [], "raw_text": "Li, Shih-Min, Su-Chu Lin, Chia-Hung Tai and Keh-Jiann Chen, 2006. A Probe into Ambiguities of Determinative-Measure Compounds, International Journal of Computational Linguistics & Chinese Language Processing, Vol. 11, No. 3. pp.245-280.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Chinese classifier systems and human categorization", "authors": [ { "first": "J. H-Y ;", "middle": [], "last": "Tai", "suffix": "" }, { "first": "S-Y.", "middle": [], "last": "William", "suffix": "" }, { "first": "", "middle": [], "last": "Wang", "suffix": "" } ], "year": 1994, "venue": "Interdisciplinary Studies on Language and Language Change", "volume": "", "issue": "", "pages": "479--494", "other_ids": {}, "num": null, "urls": [], "raw_text": "Tai, J. H-Y, Chinese classifier systems and human categorization, In Honor of William S-Y. Wang: Interdisciplinary Studies on Language and Language Change, ed. by M. Y. Chen and O J.-L. Tzeng, Pyramid Press, Taipei, 1994, pp. 479-494.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "A grammar of Spoken Chinese", "authors": [ { "first": "Y.-R", "middle": [], "last": "Chao", "suffix": "" } ], "year": 1968, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chao, Y.-R., A grammar of Spoken Chinese, University of California Press, Berkeley, 1968.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Mandarin Chinese: A Functional Reference Grammar", "authors": [ { "first": "C", "middle": [ "N" ], "last": "Li", "suffix": "" }, { "first": "S", "middle": [ "A" ], "last": "Thompson", "suffix": "" } ], "year": 1981, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Li, C. N. and S. A. Thompson, Mandarin Chinese: A Functional Reference Grammar, University of California Press, Berkeley, 1981.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Determinative-Measure Compounds in Mandarin Chinese: Their Formation Rules and Parser Implementation", "authors": [ { "first": "R.-P", "middle": [], "last": "Mo", "suffix": "" }, { "first": "Y.-J", "middle": [], "last": "Yang", "suffix": "" }, { "first": "K.-J", "middle": [], "last": "Chen", "suffix": "" }, { "first": "C.-R", "middle": [], "last": "Huang", "suffix": "" } ], "year": 1991, "venue": "Proceedings of ROCLING IV (R.O.C. Computational Linguistics Conference)", "volume": "", "issue": "", "pages": "111--134", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mo, R.-P., Y.-J. Yang, K.-J. Chen and C.-R. Huang, Determinative-Measure Compounds in Mandarin Chinese: Their Formation Rules and Parser Implementation, In Proceedings of ROCLING IV (R.O.C. Computational Linguistics Conference), 1991, National Chiao-Tung University, Hsinchu, Taiwan, pp. 111-134.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Extended-HowNet-A Representational Framework for Concepts", "authors": [ { "first": "Chen", "middle": [], "last": "Keh-Jiann", "suffix": "" }, { "first": "Shu-Ling", "middle": [], "last": "Huang", "suffix": "" }, { "first": "Yueh-Yin", "middle": [], "last": "Shih", "suffix": "" }, { "first": "Yi", "middle": [ "-" ], "last": "", "suffix": "" } ], "year": 2005, "venue": "OntoLex 2005 -Ontologies and Lexical Resources IJCNLP-05 Workshop", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Chen Keh-Jiann, Shu-Ling Huang, Yueh-Yin Shih, Yi-Jun Chen, 2005a, Extended-HowNet-A Representational Framework for Concepts, OntoLex 2005 -Ontologies and Lexical Resources IJCNLP-05 Workshop, Jeju Island, South Korea", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Concurrent constraint programming in Oz for natural language processing", "authors": [ { "first": "D", "middle": [], "last": "Duchier", "suffix": "" }, { "first": "C", "middle": [], "last": "Gardent", "suffix": "" }, { "first": "J", "middle": [], "last": "Niehren", "suffix": "" } ], "year": 1999, "venue": "Lecture notes", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Duchier, D., Gardent, C. and Niehren, J. (1999a) Concurrent constraint programming in Oz for natural language processing. Lecture notes, http://www.ps.uni-sb.de/~niehren/ oz-natural-language-script.html.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Determinative and measure word in E-HowNet representation \u5b9a\u8a5e(Determinative word) \u5b9a\u6307", "authors": [ { "first": "A", "middle": [], "last": "Appendix", "suffix": "" } ], "year": null, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Appendix A. Determinative and measure word in E-HowNet representation \u5b9a\u8a5e(Determinative word) \u5b9a\u6307 D1-> \u9019\u3001\u90a3\u3001\u6b64\u3001\u8a72\u3001\u672c\u3001\u8cb4\u3001\u655d\u3001\u5176\u3001\u67d0\u3001\u8af8 def: quantifier={definite|\u5b9a \u6307}\uff1b\u9019\u4e9b\u3001\u90a3\u4e9b def: quantifier={definite|\u5b9a\u6307}, quantity={some|\u4e9b} D2-> \u7b2c\u3001\u9996 def: ordinal={D4}", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "def: quantity={1\u30012\u300110000\u30012...} or def:ordinal={1\u30012\u3001 10000\u30012", "authors": [ { "first": "", "middle": [], "last": "D3-> \u4e0a \u3001 \u524d Def: Qualification={preceding| \u4e0a \u6b21 } \u3001 \u4e0b \u3001 \u5f8c Def ; -> \u4e00\u3001\u4e8c\u3001\u842c\u3001\u96d9", "suffix": "" } ], "year": null, "venue": "qualification={next|\u4e0b\u6b21}\u3001\u982d\u3001\u9996 def:ordinal={1}\u3001\u672b def: qualification={last| \u6700 \u5f8c}\u3001\u6b21 def:ordinal={2} \u4e0d\u5b9a\u6307 D4", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "D3-> \u4e0a \u3001 \u524d def: qualification={preceding| \u4e0a \u6b21 } \u3001 \u4e0b \u3001 \u5f8c def: qualification={next|\u4e0b\u6b21}\u3001\u982d\u3001\u9996 def:ordinal={1}\u3001\u672b def: qualification={last| \u6700 \u5f8c}\u3001\u6b21 def:ordinal={2} \u4e0d\u5b9a\u6307 D4-> \u4e00\u3001\u4e8c\u3001\u842c\u3001\u96d9... def: quantity={1\u30012\u300110000\u30012...} or def:ordinal={1\u30012\u3001 10000\u30012...} D5-> \u7532\u3001\u4e59... def: ordinal={1\u30012...} D6-> \u5176\u4ed6\u3001\u5176\u9918\u3001\u5225\u3001\u65c1\u3001\u4ed6\u3001\u53e6\u3001\u53e6\u5916 def: qualification={other|\u53e6} D7-> \u6bcf\u3001\u4efb\u4f55\u3001\u4e00\u3001\u5168\u3001\u6eff\u3001\u6574\u3001\u4e00\u5207 def: quantity={all|\u5168} D8-> \u5404 def: qualification={individual|\u5206\u5225\u7684} D9-> \u82e5\u5e72\u3001\u6709\u7684\u3001\u4e00\u4e9b\u3001\u90e8\u4efd\u3001\u6709\u4e9b def: quantity={some|\u4e9b} D10-> \u534a def: quantity={half|\u534a} D11-> \u591a\u5c11\u3001\u5e7e\u591a def: quantity={.Ques.} D12-> \u4f55\u3001\u5565\u3001\u4ec0\u9ebc def: fomal={.Ques.} D13->\u6578\u3001\u8a31\u591a\u3001\u5f88\u591a\u3001\u597d\u591a\u3001\u597d\u5e7e\u3001\u597d\u4e9b\u3001\u591a\u3001\u8a31\u8a31\u591a\u591a\u3001\u591a\u6578\u3001\u5927\u591a\u6578\u3001 \u4e0d\u5c11\u3001\u6cf0\u534a\u3001\u534a\u6578\u3001\u8af8\u591a def: quantity={many|\u591a}\u3001\u5c11\u8a31\u3001\u5c11\u6578\u3001\u5e7e\u8a31\u3001\u500b\u628a def: quantity={few|\u5c11} D14->\u9918\u3001\u8a31\u3001\u4e4b\u591a def: approximate()\u3001\u8db3\u3001\u6574\u3001\u6b63 def: exact()\u3001\u51fa\u982d\u3001\u597d\u5e7e\u3001 \u958b\u5916\u3001\u591a def: over();", "links": null } }, "ref_entries": { "FIGREF0": { "num": null, "uris": null, "text": "Determinative + Measure (D+M) def: semantic-role(M) = {Sense-representation(M): Representation(D)} The rule (10) says that the sense representation of a DM compound with a determinative D and a measure M is a unification of the feature representation of D as a feature of the sense representation of M as exemplified in (9). However a DM compound with a null sense measure word, such as 'this copy|\u9019\u672c' , ' a copy|\u4e00\u672c', or without measure word, such as 'this three|\u9019\u4e09', will be exceptions, since the measure word cannot be the semantic head of DM compound. The dependent head of determinatives become the head noun of the NP containing the DM and the sense representation of a DM is a coordinate conjunction of the feature representations of its morphemes of determinatives only. For instance, in (8), 'copy' has weak content sense; we thus regard it as a null-sense measure word and only retain the feature representation of the determinative as the definition of \"this copy|\u9019\u672c\". The unification rule for DM with null-sense measure is expressed as (11). (11) Determinative + {Null-sense Measure} (D+M) def: Representation(D);", "type_str": "figure" }, "FIGREF1": { "num": null, "uris": null, "text": "some examples of the result. For testing the correctness of our candidates, we manually check the format of 1226 candidates. Only 5 candidates out of 1226 are wrong or meaningless representations. After disambiguation processes, the resulting 1000 DM words in E-HowNet representation are judged manually. There are 880 correct E-HowNet representations for 1000 DM words in both sense and format. It is an acceptable result. Among 120 wrong answers, 57 errors are due to undefined morpheme, 28 errors are unique sense but wrong answer and the number of sense disambiguation errors is 36. Therefore accuracy of sense disambiguation is (162-36)/162=0.778.", "type_str": "figure" }, "FIGREF2": { "num": null, "uris": null, "text": "Measure word) \u6709\u8a9e\u610f\u91cf\u8a5e(Measures with content sense ) Nff-> \u66ab\u6642\u91cf\u8a5e-\u8eab\u3001\u982d\u3001\u81c9\u3001\u9f3b\u5b50\u3001\u5634\u3001\u809a\u5b50\u3001\u624b\u3001\u8173 def:{\u8eab,\u982d, \u2026} Nff-> \u66ab\u6642\u91cf\u8a5e-\u684c\u5b50\u3001\u9662\u5b50\u3001\u5730\u3001\u5c4b\u5b50\u3001\u6c60\u3001\u8154\u3001\u5bb6\u5b50 def: position={\u684c\u5b50, \u9662\u5b50...:quantity={all|\u5168}} Nfe-> \u5bb9\u5668\u91cf\u8a5e-\u76d2(\u5b50)\u3001\u5323(\u5b50)\u3001\u7bb1(\u5b50)\u3001\u6ac3\u5b50\u3001\u6ae5(\u5b50)\u3001\u7c43(\u5b50)\u3001\u7c0d(\u5b50)\u3001\u7210 \u5b50\u3001\u5305(\u5152)\u3001\u888b(\u5152)\u3001\u6c60\u5b50\u3001\u74f6(\u5b50)\u3001\u6876(\u5b50)\u3001\u807d\u3001\u7f50(\u5b50)\u3001\u76c6(\u5b50)\u3001\u934b(\u5b50)\u3001 \u7c60(\u5b50)\u3001\u76e4(\u5b50)\u3001\u7897\u3001\u676f(\u5b50)\u3001\u52fa(\u5b50)\u3001\u5319(\u6e6f\u5319)\u3001\u7b52(\u5b50)\u3001\u64d4(\u5b50)\u3001\u7c6e\u7b50\u3001 \u6753 (\u5b50)\u3001\u8336\u5319\u3001\u58fa\u3001\u76c5\u3001\u7b50\u3001\u74e2\u3001\u936c\u3001\u7f38 def: container={\u76d2,\u5323,...} Nfg-> \u6a19\u6e96\u91cf\u8a5e-\u8868\u9577\u5ea6\u7684\uff0c\u5982\uff1a\u516c\u5398\u3001\u516c\u5206\u3001\u516c\u5bf8\u3001\u516c\u5c3a\u3001\u516c\u4e08\u3001\u516c\u5f15\u3001\u516c\u91cc\u3001\u5e02\u5c3a\u3001\u71df \u9020 \u5c3a\u3001\u53f0\u5c3a\u3001\u540b(inch)\u3001\u544e(feet)\u3001\u78bc(yard)\u3001\u54e9(mile)\u3001 (\u6d77)\u6d6c\u3001\u5eb9\u3001\u565a\u3001 \u5c3a\u3001\u91cc\u3001\u91d0\u3001\u5bf8\u3001\u4e08\u3001\u7c73\u3001\u5398\u3001\u5398\u7c73\u3001\u6d77 \u54e9\u3001\u82f1\u5c3a\u3001\u82f1\u91cc\u3001\u82f1\u544e\u3001\u82f1\u5bf8\u3001 \u7c73\u7a81\u3001\u7c73\u5c3a\u3001\u5fae\u7c73\u3001\u6beb\u7c73\u3001 \u82f1\u540b\u3001\u82f1\u54e9\u3001\u5149\u5e74\u3002 def: length={\u516c\u5206,...} \u8868\u9762\u7a4d\u7684\uff0c\u5982\uff1a\u516c\u755d\u3001\u516c\u9803\u3001\u5e02\u755d\u3001\u71df\u9020\u755d\u3001\u576a\u3001\u755d\u3001\u5206\u3001\u7532\u3001\u9803\u3001\u5e73\u65b9 \u516c\u91cc\u3001\u5e73\u65b9\u516c\u5c3a\u3001\u5e73\u65b9\u516c\u5206\u3001\u5e73\u65b9\u5c3a\u3001\u5e73\u65b9\u82f1\u54e9\u3001\u82f1\u755d\u3002def: size={\u516c\u755d,...} \u8868\u91cd\u91cf\u7684\uff0c\u5982\uff1a\u516c\u514b\u3001\u516c\u65a4\u3001\u516c\u5678\u3001\u5e02\u65a4\u3001\u53f0\u5169\u3001\u53f0\u65a4(\u65e5\u65a4)\u3001\u76ce\u53f8(\u65af)\u3001 \u78c5\u3001\u516c\u64d4\u3001\u516c\u8861\u3001\u516c\u5169\u3001\u514b\u62c9\u3001\u65a4\u3001\u5169\u3001\u9322\u3001\u5678\u3001\u514b\u3001\u82f1\u78c5\u3001\u82f1\u5169\u3001\u516c\u9322\u3001 \u6beb\u514b\u3001\u6beb\u5206\u3001\u4edf\u514b\u3001\u516c\u6beb\u3002def: weight={\u516c\u514b,...} \u8868\u5bb9\u91cf\u7684\uff0c\u5982\uff1a\u516c\u64ae\u3001\u516c\u5347(\u5e02\u5347)\u3001\u71df\u9020\u5347\u3001\u53f0\u5347(\u65e5\u5347)\u3001\u76ce\u53f8\u3001\u54c1\u812b(pint)\u3001 \u52a0\u4f96(gallon)\u3001\u84b2\u5f0f\u8033(bushel)\u3001\u516c\u6597\u3001\u516c\u77f3\u3001\u516c\u79c9\u3001\u516c\u5408\u3001\u516c\u52fa\u3001\u6597\u3001\u6beb \u5347\u3001\u5938\u3001\u5938\u7279\u3001\u5938\u723e\u3001\u7acb\u65b9\u7c73\u3001\u7acb\u65b9\u5398\u7c73\u3001\u7acb\u65b9\u516c\u5206\u3001\u7acb\u65b9\u516c\u5bf8\u3001\u7acb\u65b9\u516c \u5c3a\u3001\u7acb\u5206\u516c\u91cc\u3001\u7acb\u65b9\u82f1\u5c3a\u3001\u77f3\u3001\u659b\u3001\u897f\u897f\u3002def: volume={\u516c\u64ae,\u516c\u5347,...} \u8868\u6642\u9593\u7684\uff0c\u5982\uff1a\u5fae\u79d2\u3001\u91d0\u79d2\u3001\u79d2\u3001\u79d2\u9418\u3001\u5206\u3001\u5206\u9418\u3001\u523b\u3001\u523b\u9418\u3001\u9ede\u3001\u9ede\u9418\u3001 \u6642\u3001\u5c0f\u6642\u3001\u66f4\u3001\u591c\u3001\u65ec\u3001\u7d00(\u8f2a, 12 \u5e74) \u3001\u4e16\u7d00\u3001\u5929(\u65e5)\u3001\u661f\u671f(\u79ae\u62dc\u3001\u9031\u3001 \u5468) \u3001\u6708\u3001\u6708\u4efd\u3001\u5b63\u3001\u5e74(\u8f09\u3001\u6b72) \u3001\u5e74\u4efd\u3001\u665a\u3001\u5bbf\u3001\u3002def:temporal={\u5fae \u79d2,\u6708\u2026}, \u9031\u5e74\u3001\u5468\u6b72 def:duration={\u5e74} \u8868\u9322\u5e63\u7684\uff0c\u5982\uff1a\u5206\u3001\u89d2(\u6bdb)\u3001\u5143(\u5713)\u3001\u584a\u3001\u5169\u3001\u5148\u4ee4\u3001\u76e7\u6bd4\u3001\u6cd5\u90ce(\u6717)\u3001\u8fa8", "type_str": "figure" }, "FIGREF3": { "num": null, "uris": null, "text": "e.g.All Nfi, def:frequency={}, while part-of-speech of head word is Verb, i.e. E-HowNet representation of head word is {event|\u4e8b\u4ef6} and it's subclass. Except POS V_2 and VG. All Nfi,def:{null}, while part-of-speech of head word is Noun, i.e. E-HowNet of head word is {object|\u7269\u9ad4} and it's subclass. e.g.\u90e8, \u80a1\u2026,Nfh,def:location={ }, if part-of-speech of head word is Na or previous word is \u9019\"this\" or \u90a3\"that\" or \u6bcf\"every\", do not use this definition. \u90e8,\u80a1\u2026,Nfa,def:{null}, otherwise use this definition. e.g. \u76e4 ,Nfe,def:container={plate| \u76e4 },while head word is food, i.e. E-HowNet representation of head word is {edible|\u98df\u7269} and it's subclass. \u76e4,Nfb,def:{null},otherwise use this one. e.g. \u5206 ,Nfg, def:role={ \u5206 }, while head word is \u9322 \"money\", i.e. E-HowNet representation of head word is {money|\u8ca8\u5e63} and it's subclass. \u5206,Nfg, def:size={ \u5206 }, while head word is \u5730 \"land\", i.e. E-HowNet representation of head word is {land|\u9678\u5730} and it's subclass. \u5206,Nfa, def:{null}, while part-of-speech of head word is Na or Nv. For example: \u4e00\u5206\u8015\u8018\uff1b\u5341\u5206\u529b\u6c23\uff1b\u4e94\u5206\u719f. e.g.\u9ede,Nfh;Nfd,def:{null}, while part-of-speech of head word is Nab. If part-of-speech of head word is V, Naa or Nad, do not use this definition. collocation-based rule e.g.\u5206,Nfh,def:role={score|\u5206\u6578:quantity={D4,D15}}, while the sentence also contains the words \u8003 \"give an exam\" (E-HowNet representation is {exam|\u8003\u8a66}) or \u5f97 \"get\" (E-HowNet representation is {obtain|\u5f97\u5230}) or \u5931\"lose\" (E-HowNet representation is {lose|\u5931\u53bb}), then use this definition.", "type_str": "figure" }, "TABREF1": { "html": null, "type_str": "table", "text": "\u56e0\u70ba because' are function words. The difference of their representation is that function words start with a relation but content words have under-specified relations. If a content word plays a dependent daughter of a head concept, the relation between the head concept and this content word will be established after parsing process. Suppose that the following dependent structure and semantic relations are derived after parsing the sentence", "num": null, "content": "
sense representations are exemplified below:
(6) The definition of measure words in E-HowNet
a) Measure words with content sense
d. e. f. g. represent and to identify morphological structures of DM compounds, but do not derive the IN1 -> { NO1*, NO3*} ; IN2 -> NO2* ; IN3 -> {IN1,IN2} {\u591a,\u9918,\u4f86,\u5e7e} ({\u842c,\u5104,\u5146}) ; Neu -> {IN1,IN2,IN3 } ; senses of complex DM compounds. 2.2 Lexical Sense Representation in E-HowNet Core senses of natural language are compositions of relations and entities. Lexical senses are processing units for sense composition. Conventional linguistic theories classify words into content words and function words. Content words denote entities and function words without too much content sense mainly serve grammatical function which links relations between entities/events. In E-HowNet, the senses of function words are represented by semantic roles/relations [11]. For example, 'because' is a function word. Its E-HowNet definition is shown in (1). (1) because|\u56e0\u70ba def: reason={}; which means reason(x)={y} where x is the dependent head and y is the dependent daughter of '\u56e0\u70ba'. In following sentence (2), we'll show how the lexical concepts are combined into the sense representation of the sentence. (2) Because of raining, clothes are all wet. \u56e0\u70ba\u4e0b\u96e8\uff0c\u8863\u670d\u90fd\u6fd5\u4e86 In the above sentence, '\u6fd5 wet', '\u8863\u670d clothes' and '\u4e0b\u96e8 rain' are content words while '\u90fd (3) S(reason:VP(Head:Cb:\u56e0\u70ba|dummy:VA:\u4e0b\u96e8)|theme:NP(Head:Na:\u8863\u670d) | quantity: Da:\u90fd | Head:Vh:\u6fd5|particle:Ta:\uf9ba)\u3002 After feature-unification process, the following semantic composition result (4) is derived. The sense representations of dependent daughters became the feature attributes of the sentential head 'wet|\u6fd5'. (4) def:{wet|\u6fd5: aspect={Vachieve|\u9054\u6210}, manner={complete|\u6574}, reason={rain|\u4e0b\u96e8}} In (3), function word '\u56e0\u70ba because' links the relation of 'reason' between head concept '\u6fd5 wet' and '\u4e0b\u96e8 rain'. The result of composition is expressed as reason(wet|\u6fd5)={rain|\u4e0b\u96e8}, since for simplicity the dependent head of a relation is normally omitted. Therefore reason(wet|\u6fd5)={rain|\u4e0b\u96e8} is expressed as reason={rain|\u4e0b\u96e8}; theme(wet|\u6fd5)={clothing|\u8863 \u7269} is expressed as theme={clothing|\u8863\u7269} and so on. 2.3 The sense representation for determinatives and measures in E-HowNet The sense of a DM compound is determined by its morphemes and the set of component morphemes are determinatives and measures which are exhaustively listable. Therefore in order to apply semantic composition mechanism to derive the senses of DM compounds, we need to establish the sense representations for all morphemes of determinatives and measures (5) The definition of determinatives in E-HowNet this \u9019 def: quantifier={definite|\u5b9a\u6307} first \u9996 def: ordinal={1} one \u4e00 def: quantity={1} bowl \u7897 def: container={bowl|\u7897} meter \u7c73 def: length={meter|\u516c\u5c3a} month \u6708 def: time={month|\u6708} b) Measure words without content sense \u672c copy def:{null} \u9593 room def:{null} all', '\uf9ba Le' and '(2). theme={clothing|\u8863\u7269}, \u6a23 kind def:{null}
" }, "TABREF2": { "html": null, "type_str": "table", "text": "However there are some complications need to be resolved. First of all we have to clarify the dependent relation between the determinative and the measure of a DM in order to make a right feature unification process. In principle, a dependent head will take semantic representation of its dependent daughters as its features. Usually determinatives are modifiers of measures, such as \u9019\u7897, \u4e00\u7897, \u9019\u4e00\u7897. For instance, the example (9) has the dependent", "num": null, "content": "
quantifier addresse.g. \u9019\u3001\u90a3\u3001\u6b64\u3001\u8a72\u3001\u672c\u3001\u8cb4\u3001\u655d\u3001\u5176\u3001\u67d0\u3001\u8af8 e.g. \u570b\u3001\u7701\u3001\u5dde\u3001\u7e23\u3001\u9109\u3001\u6751\u3001\u93ae\u3001\u9130\u3001\u91cc\u3001\u90e1\u3001\u5340\u3001
ordinale.g. \u7b2c\u3001\u9996 \u7ad9\u3001\u5df7\u3001\u5f04\u3001\u6bb5\u3001\u865f\u3001\u6a13\u3001\u8856\u3001\u5e02\u3001\u6d32\u3001\u5730\u3001\u8857
qualification placee.g. \u4e0a\u3001\u4e0b\u3001\u524d\u3001\u5f8c\u3001\u982d\u3001\u672b\u3001\u6b21\u3001\u9996\u3001\u5176\u4ed6\u3001\u5176\u9918\u3001 e.g \u90e8\u3001\u53f8\u3001\u8ab2\u3001\u9662\u3001\u79d1\u3001\u7cfb\u3001\u7d1a\u3001\u80a1\u3001\u5ba4\u3001\u5ef3
duration\u5225\u3001\u65c1\u3001\u4ed6\u3001\u53e6\u3001\u53e6\u5916\u3001\u5404 e.g \u9663(\u5b50)\u3001\u6703\u3001\u6703\u5152\u3001\u4e0b\u5b50
quantitye.g. \u4e00\u3001\u4e8c\u3001\u842c\u3001\u96d9\u3001\u6bcf\u3001\u4efb\u4f55\u3001\u4e00\u3001\u5168\u3001\u6eff\u3001\u6574\u3001\u4e00
\u5207\u3001\u82e5\u5e72\u3001\u6709\u7684\u3001\u4e00\u4e9b\u3001\u90e8\u4efd\u3001\u6709\u4e9b\u3001\u8a31\u591a\u3001\u5f88\u591a\u3001\u597d
\u591a\u3001\u597d\u5e7e\u3001\u597d\u4e9b\u3001\u5c11\u8a31\u3001\u591a\u3001\u8a31\u8a31\u591a\u591a\u3001\u5e7e\u8a31\u3001\u591a\u6578\u3001
\u5c11\u6578\u3001\u5927\u591a\u6578\u3001\u6cf0\u534a\u3001\u4e0d\u5c11\u3001\u500b\u628a\u3001\u534a\u6578\u3001\u8af8\u591a
Formal={.Ques.}e.g. \u4f55\u3001\u5565\u3001\u4ec0\u9ebc
Quantity={over,e.g. \u9918\u3001\u8a31\u3001\u8db3\u3001\u4e4b\u591a\u3001\u51fa\u982d\u3001\u597d\u5e7e\u3001\u958b\u5916\u3001\u6574\u3001\u6b63
approximate, exact} relations of
position NP(quantifier:DM(quantifier:Neu:\u4e00|container:Nfa:\u7897)|Head:Nab:\u9eb5) e.g. \u684c\u5b50\u3001\u9662\u5b50\u3001\u5730\u3001\u5c4b\u5b50\u3001\u6c60\u3001\u8154\u3001\u5bb6\u5b50
containere.g. \u76d2(\u5b50)\u3001\u5323(\u5b50)\u3001\u7bb1(\u5b50)\u3001\u6ac3\u5b50\u3001\u6ae5(\u5b50)\u3001\u7c43(\u5b50)\u3001\u7c0d(\u5b50)\u3001
\u7210\u5b50\u3001\u5305(\u5152)\u3001\u888b(\u5152)\u3001\u6c60\u5b50\u3001\u74f6(\u5b50)\u3001\u6876(\u5b50)\u3001\u807d\u3001\u7f50(\u5b50)\u3001
\u76c6(\u5b50)\u3001\u934b(\u5b50)\u3001\u7c60(\u5b50)\u3001\u76e4(\u5b50)\u3001\u7897\u3001\u676f(\u5b50)\u3001\u52fa(\u5b50)\u3001
\u5319(\u6e6f\u5319)\u3001\u7b52(\u5b50)\u3001\u64d4(\u5b50)\u3001\u7c6e\u7b50\u3001\u6753(\u5b50)\u3001\u8336\u5319\u3001\u58fa\u3001\u76c5\u3001
\u7b50\u3001\u74e2\u3001\u936c\u3001\u7f38
lengthe.g. \u516c\u5398\u3001\u516c\u5206\u3001\u516c\u5bf8\u3001\u516c\u5c3a\u3001\u516c\u4e08\u3001\u516c\u5f15\u3001\u516c\u91cc\u3001\u5e02
\u5c3a\u3001\u71df\u9020 \u5c3a\u3001\u53f0\u5c3a\u3001\u540b(inch)\u3001\u544e(feet)\u3001\u78bc(yard)\u3001\u54e9
(mile)\u3001 (\u6d77)\u6d6c\u3001\u5eb9\u3001\u565a\u3001\u5c3a\u3001\u91cc\u3001\u91d0\u3001\u5bf8\u3001\u4e08\u3001\u7c73\u3001
\u5398\u3001\u5398\u7c73\u3001\u6d77 \u54e9\u3001\u82f1\u5c3a\u3001\u82f1\u91cc\u3001\u82f1\u544e\u3001\u82f1\u5bf8\u3001\u7c73\u7a81\u3001
\u7c73\u5c3a\u3001\u5fae\u7c73\u3001\u6beb\u7c73\u3001 \u82f1\u540b\u3001\u82f1\u54e9\u3001\u5149\u5e74
sizee.g. \u516c\u755d\u3001\u516c\u9803\u3001\u5e02\u755d\u3001\u71df\u9020\u755d\u3001\u576a\u3001\u755d\u3001\u5206\u3001\u7532\u3001\u9803\u3001
\u5e73\u65b9\u516c\u91cc\u3001\u5e73\u65b9\u516c\u5c3a\u3001\u5e73\u65b9\u516c\u5206\u3001\u5e73\u65b9\u5c3a\u3001\u5e73\u65b9\u82f1\u54e9\u3001
\u82f1\u755d
weighte.g. \u516c\u514b\u3001\u516c\u65a4\u3001\u516c\u5678\u3001\u5e02\u65a4\u3001\u53f0\u5169\u3001\u53f0\u65a4(\u65e5\u65a4)\u3001\u76ce\u53f8
(\u65af)\u3001\u78c5\u3001\u516c\u64d4\u3001\u516c\u8861\u3001\u516c\u5169\u3001\u514b\u62c9\u3001\u65a4\u3001\u5169\u3001\u9322\u3001\u5678\u3001
\u514b\u3001\u82f1\u78c5\u3001\u82f1\u5169\u3001\u516c\u9322\u3001\u6beb\u514b\u3001\u6beb\u5206\u3001\u4edf\u514b\u3001\u516c\u6beb
volumee.g. \u516c\u64ae\u3001\u516c\u5347(\u5e02\u5347)\u3001\u71df\u9020\u5347\u3001\u53f0\u5347(\u65e5\u5347)\u3001\u76ce\u53f8\u3001\u54c1
, once its morpheme sense representations and \u812b(pint)\u3001\u52a0\u4f96(gallon)\u3001\u84b2\u5f0f\u8033(bushel)\u3001\u516c\u6597\u3001\u516c\u77f3\u3001\u516c
semantic head are known. \u79c9\u3001\u516c\u5408\u3001\u516c\u52fa\u3001\u6597\u3001\u6beb\u5347\u3001\u5938\u3001\u5938\u7279\u3001\u5938\u723e\u3001\u7acb\u65b9\u7c73\u3001
(7) one \u4e00 def:quantity={1} + bowl \u7897 def: container={bowl|\u7897} \u7acb\u65b9\u5398\u7c73\u3001\u7acb\u65b9\u516c\u5206\u3001\u7acb\u65b9\u516c\u5bf8\u3001\u7acb\u65b9\u516c\u5c3a\u3001\u7acb\u5206\u516c\u91cc\u3001
one bowl \u4e00\u7897def: container={bowl|\u7897:quantity={1}} \u7acb\u65b9\u82f1\u5c3a\u3001\u77f3\u3001\u659b\u3001\u897f\u897f
(8) this \u9019 def: quantifier={definite|\u5b9a\u6307} + \u672c copy def:{null} time e.g. \u5fae\u79d2\u3001\u91d0\u79d2\u3001\u79d2\u3001\u79d2\u9418\u3001\u5206\u3001\u5206\u9418\u3001\u523b\u3001\u523b\u9418\u3001\u9ede\u3001
this copy \u9019\u672cdef: quantifier={definite|\u5b9a\u6307} \u9ede\u9418\u3001\u6642\u3001\u5c0f\u6642\u3001\u66f4\u3001\u591c\u3001\u65ec\u3001\u7d00(\u8f2a, 12 \u5e74) \u3001\u4e16\u7d00\u3001\u5929
Table 1. Major semantic roles played by determinants and measures (\u65e5)\u3001\u661f\u671f(\u79ae\u62dc\u3001\u9031\u3001\u5468) \u3001\u6708\u3001\u6708\u4efd\u3001\u5b63\u3001\u5e74(\u8f09\u3001\u6b72) \u3001
\u9031\u5e74\u3001\u5468\u6b72\u3001\u5e74\u4efd\u3001\u665a\u3001\u5bbf\u3001\u4e16\u3001\u8f29\u3001\u8f29\u5b50\u3001\u4ee3\u3001\u5b78\u671f\u3001
Semantic RoleD/M \u5b78\u5e74\u3001\u5e74\u4ee3
" }, "TABREF4": { "html": null, "type_str": "table", "text": "The result of semantic composition for DM compounds.", "num": null, "content": "
DM Compounds E-HowNet Representation
\u4e8c\u5341\u842c\u5143def:role={money|\u8ca8\u5e63:quantity={200000}}
\u53e6\u4e00\u500bdef:qualification={other|\u53e6},quantity={1}
\u4e8c\u767e\u4e09\u5341\u516d\u5206def:role={\u5206\u6578:quantity={236}}
\u524d\u4e94\u5929def:time={day| \u65e5 :qualification={preceding| \u4e0a \u6b21 },
quantity={5}}
\u4e00\u767e\u4e00\u5341\u516d\u9ede\u4e03def:role={\u7f8e\u5143:quantity={11670000000}}
\u5104\u7f8e\u5143
" } } } }