| { | |
| "paper_id": "I05-1029", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:25:20.923873Z" | |
| }, | |
| "title": "Analogy as Functional Recategorization: Abstraction with HowNet Semantics", | |
| "authors": [ | |
| { | |
| "first": "Tony", | |
| "middle": [], | |
| "last": "Veale", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University College Dublin", | |
| "location": { | |
| "postCode": "D4", | |
| "settlement": "Belfield, Dublin", | |
| "country": "Ireland" | |
| } | |
| }, | |
| "email": "tony.veale@ucd.ie" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "One generally accepted hallmark of creative thinking is an ability to look beyond conventional labels and recategorize a concept based on its behaviour and functional potential. So while taxonomies are useful in any domain of reasoning, they typically represent the conventional label set that creative thinking attempts to look beyond. So if a linguistic taxonomy like WordNet [1] is to be useful in driving linguistic creativity, it must support some basis for recategorization, to allow an agent to reorganize its category structures in a way that unlocks the functional potential of objects, or that recognizes similarity between literally dissimilar ideas. In this paper we consider how recategorization can be used to generate analogies using the HowNet [2] ontology, a lexical resource like WordNet that in addition to being bilingual (Chinese/English) also provides explicit semantic definitions for each of the terms that it defines.", | |
| "pdf_parse": { | |
| "paper_id": "I05-1029", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "One generally accepted hallmark of creative thinking is an ability to look beyond conventional labels and recategorize a concept based on its behaviour and functional potential. So while taxonomies are useful in any domain of reasoning, they typically represent the conventional label set that creative thinking attempts to look beyond. So if a linguistic taxonomy like WordNet [1] is to be useful in driving linguistic creativity, it must support some basis for recategorization, to allow an agent to reorganize its category structures in a way that unlocks the functional potential of objects, or that recognizes similarity between literally dissimilar ideas. In this paper we consider how recategorization can be used to generate analogies using the HowNet [2] ontology, a lexical resource like WordNet that in addition to being bilingual (Chinese/English) also provides explicit semantic definitions for each of the terms that it defines.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Analogy is a knowledge-hungry process that exploits a conceptual system's ability to perform controlled generalization in one domain and re-specialization into another. The result is a taxonomic leap within an ontology that transfers semantic content from one term onto another. While all taxonomies allow vertical movement, a system must fully understand the effects of generalization on a given concept before any analogy or metaphor can be considered either deliberate or meaningful. So to properly support analogy, a taxonomy must provide a basis of abstracting not just to conventional categories, like Person, Animal or Tool, but to categories representing the specific causal behaviour of concepts such as think-agent, pain-experiencer, cutting-instrument, and so on. Thus, a surgeon can be meaningfully described as a repairman since both occupations have the function of restoring an object to an earlier and better state; a footballer can be meaningfully described as a gladiator or a warrior since each exhibits competitive behaviour; and a scalpel can be compared to a sabre, a sword or a cleaver since each has a cutting behaviour; and so on.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Theories of metaphor and analogy are typically based either on structure-mapping [3, 4] or on abstraction e.g., [5, 6, 7, 8, 9, 10] ). While the former is most associated with analogy, the latter has been a near-constant in the computational treatment of metaphor. Structure-mapping assumes that the causal behaviour of a concept is expressed in an explicit, graph-theoretic form so that unifying sub-graph isomorphisms can be found between different representations. In contrast, abstraction theories assume that analogous concepts, even when far removed in ontological terms, will nonetheless share a common hypernym that captures their causal similarity. Thus, we should expect an analogous pairing like surgeon and butcher to have different immediate hypernyms but to ultimately share an abstraction like cutting-agent (see [8, 9] ).", | |
| "cite_spans": [ | |
| { | |
| "start": 81, | |
| "end": 84, | |
| "text": "[3,", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 85, | |
| "end": 87, | |
| "text": "4]", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 112, | |
| "end": 115, | |
| "text": "[5,", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 116, | |
| "end": 118, | |
| "text": "6,", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 119, | |
| "end": 121, | |
| "text": "7,", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 122, | |
| "end": 124, | |
| "text": "8,", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 125, | |
| "end": 127, | |
| "text": "9,", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 128, | |
| "end": 131, | |
| "text": "10]", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 828, | |
| "end": 831, | |
| "text": "[8,", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 832, | |
| "end": 834, | |
| "text": "9]", | |
| "ref_id": "BIBREF8" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "However, the idea that a standard ontology will actually provide a hypernym like cutting-agent seems convenient almost to the point of incredulity. The problem is, of course, that as much as we want our ontologies to anticipate future analogies and metaphors with these pro-active categorizations, most ontologies simply do not possess terms as prescient as these. This is the question we address in this paper: if we assume that our ontologies lack these structures, can we nonetheless enable them to be added via automatic means? We argue that we can, by generalizing not on the basis of a concept's taxonomic position but on the basis of the specific relations that define its causal behaviour.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Clearly then, this approach to analogy requires a resource that is rich in causal relations. We find this richness in HowNet [2, 11] , a bilingual lexical ontology for Chinese and English that employs an explicit propositional semantics to define each of its lexical concepts.", | |
| "cite_spans": [ | |
| { | |
| "start": 125, | |
| "end": 128, | |
| "text": "[2,", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 129, | |
| "end": 132, | |
| "text": "11]", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "With this goal in mind, the paper observes the following structure: in section two we offer a concise survey of the considerable research that has, in the past, been dedicated to abstraction theories of analogy and metaphor. In section three we then compare and contrast WordNet [1] and HowNet as candidate resources for the current abstraction approach to analogical reasoning. In section four, having established an argument as to why HowNet is to be preferred, we indicate how HowNet's semantic definitions can be transformed in the service of analogical recategorization. The performance and competence of this recategorization ability is then evaluated in section five. Speculation about further possible contributions of HowNet to analogical research is reserved for the closing remarks of section six.", | |
| "cite_spans": [ | |
| { | |
| "start": 279, | |
| "end": 282, | |
| "text": "[1]", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "That analogy and metaphor operate across multiple levels of conceptual abstraction has been well known since classical times. Aristotle first provided a compelling taxonomic account of both in his Poetics (see [5] , for a translation), and computationalists have been fascinated by this perspective ever since. While the core idea has survived relatively unchanged, one must discriminate theories that apparently presume a static type-hierarchy to be sufficient for all abstraction purposes (e.g., [6] ), from theories that posit the need for a dynamic type hierarchy (e.g., [7, 8] ). One must also differentiate theories that have actually been implemented (e.g., [6, 8, 9] ) from those that are either notional or that seem to court computational intractability (e.g., [5, 6] ). Perhaps most meaningfully, one must differentiate theories and implementations that assume hand-crafted, purpose-built ontologies (e.g., [6] ) from those that exploit an existing large-scale resource like WordNet (e.g., [8, 9] ). In the former, one has the flexibility to support as many functional abstractions like cutting-agent as are believed necessary, but at the cost of appearing to anticipate future analogies by hand-crafting them into the system. Fig. 1 . Analysis of the WordNet gloss for {Athena} suggests that the word-form \"wisdom\" has analogical potential, since it is alignable with another use in {Ganesh}. This leads to the construction of the dynamic sense {Wisdom_deity} which can be used to make analogical leaps between these concepts.", | |
| "cite_spans": [ | |
| { | |
| "start": 210, | |
| "end": 213, | |
| "text": "[5]", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 498, | |
| "end": 501, | |
| "text": "[6]", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 575, | |
| "end": 578, | |
| "text": "[7,", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 579, | |
| "end": 581, | |
| "text": "8]", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 665, | |
| "end": 668, | |
| "text": "[6,", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 669, | |
| "end": 671, | |
| "text": "8,", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 672, | |
| "end": 674, | |
| "text": "9]", | |
| "ref_id": "BIBREF8" | |
| }, | |
| { | |
| "start": 771, | |
| "end": 774, | |
| "text": "[5,", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 775, | |
| "end": 777, | |
| "text": "6]", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 918, | |
| "end": 921, | |
| "text": "[6]", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 1001, | |
| "end": 1004, | |
| "text": "[8,", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 1005, | |
| "end": 1007, | |
| "text": "9]", | |
| "ref_id": "BIBREF8" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 1238, | |
| "end": 1244, | |
| "text": "Fig. 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Abstraction Theories of Analogy", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "This current work follows the latter course. We intend to automatically construct a new taxonomy of analogically-useful abstractions like cutting-agent, by analysing the semantic content of the definitions assigned to each word-sense in HowNet. Past work (e.g., [8] ) has attempted this automatic construction of analogically-friendly taxonomies from WordNet, resulting in an approach that involves as much informationextraction from free text as it does semantic inference. This is because WordNet's glosses, unlike the semantic definitions of HowNet, are free-form sentences designed for human, rather than machine, consumption. For instance, Figure 1 above illustrates how features can be lifted from WordNet glosses to create new intermediate taxonyms, or dynamic types, from which subsequent abstraction-based analogies can be generated.", | |
| "cite_spans": [ | |
| { | |
| "start": 262, | |
| "end": 265, | |
| "text": "[8]", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 645, | |
| "end": 653, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Abstraction Theories of Analogy", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "The explicitly-structured semantic forms that one finds in HowNet definitions will clearly make this lifting of features more logical and less heuristic. In general, this makes HowNet an ideal knowledge-source for a computational model of metaphor and analogy (e.g., see [10] for a topical perspective).", | |
| "cite_spans": [ | |
| { | |
| "start": 271, | |
| "end": 275, | |
| "text": "[10]", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstraction Theories of Analogy", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Generalization can be considered \"controlled\" if, when moving to a higher level of abstraction in a taxonomy, a conceptual system is able to precisely quantify that meaning which is lost. In this sense at least, most large-scale taxonomies do not provide a significant degree of control. Perhaps nowhere is this observation more keenly felt than in weak lexical ontologies like Princeton WordNet (PWN). In PWN [1] , generalization of a concept/synset does not generally yield a functional or behavioural abstraction of the original concept. This is so because WordNet's taxonomy is designed not to capture common causality, function and behaviour, but to show how existing lexemes relate to each other. For example, the common abstraction that unites {surgeon, sawbones} and {tree_surgeon} is not a concept that captures a shared sense of repair, improvement or care, but {person, human}. To be fair, much the same must be said of other taxonomies, even that of HowNet [2, 11] , a Chinese/English semantic dictionary, and Cyc [12] . However, as we shall demonstrate, HowNet contains the necessary basis for such abstractions in its relational semantic definitions.", | |
| "cite_spans": [ | |
| { | |
| "start": 410, | |
| "end": 413, | |
| "text": "[1]", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 969, | |
| "end": 972, | |
| "text": "[2,", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 973, | |
| "end": 976, | |
| "text": "11]", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 1026, | |
| "end": 1030, | |
| "text": "[12]", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Comparing WordNet and HowNet", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "PWN and HowNet have each been designed according a different theory of semantic organization. PWN is differential is nature: rather than attempting to express the meaning of a word explicitly, PWN instead differentiates words with different meanings by placing them in different synsets, and further differentiates synsets from one another by assigning them to different positions in its ontology. In contrast, HowNet is constructive in nature, exploiting sememes from a less discriminating taxonomy than PWN's to compose a semantic representation of meaning for each word sense.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Comparing WordNet and HowNet", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Nonetheless, HowNet compensates strongly with its constructive semantics. For example, HowNet assigns the concept surgeon|\u533b \u751f the following definition:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Comparing WordNet and HowNet", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "{human|\u4eba:HostOf={Occupation|\u804c \u4f4d },domain={medical|\u533b }, {doctor|\u533b \u6cbb :", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Comparing WordNet and HowNet", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "agent={~}}} which can be glossed thus: \"a surgeon is a human with an occupation in the medical domain who acts as the agent of a doctoring activity.\" The {~} serves as a selfreference here, to mark the location of the concept being defined in the given semantic structure. The oblique reference offered by the tilde construct serves to make the definition more generic (thereby facilitating analogy), so that many different concepts can conceivably employ the same definition. Thus, HowNet uses the above definition not only for surgeon, but for medical workers in general, from orderlies to nurses to internists and neurologists.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Comparing WordNet and HowNet", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Our scheme for converting HowNet's constructive definitions into a more differential form hinges on the use of the tilde as a self-reference in relational structures. For instance, consider the semantic definition that HowNet gives to repairman|\u4fee\u7406\u5de5:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Extracting Functional Structure", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "{human|\u4eba:HostOf={Occupation|\u804c \u4f4d }, {repair|\u4fee\u7406:agent={~}}}", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Extracting Functional Structure", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Noting the position of {~} here, we can infer that a repairman is the agent of a repairing activity, or in differential terms, a repair-agent. Now, since HowNet defines re-pair|\u4fee\u4fee as a specialization of the reinstatement activity resume|\u6062\u590d, we can further establish repair-agent as a specialization of resume-agent. This double layer of abstraction establishes a new taxonomy that organizes wordconcepts according to their analogical potential, rather than their formal ontological properties. For instance, as shown in Figure 2 , resume-agent encompasses not only repair-agent, but doctor-agent, since HowNet also defines the predicate doctor|\u533b \u6cbb as a specialization of the predicate resume|\u6062\u590d .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 520, | |
| "end": 528, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Extracting Functional Structure", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "In general, given a semantic fragment F:role={~} in a HowNet definition, we create the new abstractions F-role and F'-role, where F' is the immediate hypernym of F. The role in question might be agent, instrument, location, patient, or any other role that HowNet supports. By way of example, Figure 3 ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 292, | |
| "end": 300, | |
| "text": "Figure 3", | |
| "ref_id": "FIGREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Extracting Functional Structure", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "We evaluate the analogical potential of the newly derived functional taxonomy using four criteria: topology -the branching structure of the new taxonomy dictates its ability to generate analogies; coverage -the percentage of unique HowNet definitions that can be functionally re-indexed in the new taxonomy; recall -the percentage of unique definitions for which at least one analogy can be found using the new taxonomy; and parsimony-the percentage of abstractions in the new taxonomy that can be used to generate analogies.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "\u5f62 \u5f62 \u53d8 5 Evaluating Analogical Competence", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The new functional taxonomy contains 1579 mid-level abstractions and 838 upper-level abstractions. In total, the taxonomy contains only 2219 unique abstractions, revealing that in 8% of cases, the upper-level abstraction of one concept serves as the upper-level abstraction of another. Analogies will be generated only if two or more unique concept definitions are coindexed under the same mid-level or upper-level abstraction in the new functional taxonomy. For example, knight|\u9a91 Nonetheless, we note that a certain degree of metaphoric licence has already been exercised by HowNet's designers in assigning semantic structures, so that even semantically distant concepts can still share the same mid-level abstraction. Creative analogies like \"Death is an assassin\" can, as shown in Figure 4 , be understood via a single generalization.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 784, | |
| "end": 792, | |
| "text": "Figure 4", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Topological Characteristics of the New Functional Taxonomy", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "attack-agent assassin|\u523a\u5ba2 intruder|\u4fb5\u7565\u8005 Death|\u6b7b\u795e man-eater|\u98df\u4eba\u9ca8 Fig. 4 . Semantic diversity among concepts with the same mid-level abstraction Furthermore, because HowNet contains 95,407 unique lexical concepts (excluding synonyms) but only 23,507 unique semantic definitions, these definitions must be under-specified to the extent that many are shared by non-identical concepts (e.g., cart|\u677f\u8f66 and bicycle|\u5355 \u8f66 , are simply defined as manual vehicles).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 61, | |
| "end": 67, | |
| "text": "Fig. 4", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "MakeBad-agent kill-agent", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Since this new taxonomy is derived from the use of {~} in HowNet definitions, both the coverage and recall of analogy generation crucially depend on the widespread use of this reflexive construct. However, of the 23,505 unique definitions in HowNet, just 6430 employ thus form of self-reference. The coverage of the new taxonomy is thus 27% of HowNet definitions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Analogical Coverage", | |
| "sec_num": "5.2" | |
| }, | |
| { | |
| "text": "A majority of the abstractions in the new taxonomy, 59%, serve to co-index two or more HowNet definitions. Overall, analogies are generated for 6184 unique HowNet definitions, though these individual definitions may have many different lexical realizations. The recall rate thus is 26% of HowNet's 23,507 unique definitions, or 96% of the 6430 HowNet definitions that make use of {~}. The most productive abstraction is control_agent, which serves to co-index 210 unique definitions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Analogical Recall", | |
| "sec_num": "5.3" | |
| }, | |
| { | |
| "text": "Overall, 1,315 of the 2219 nodes in the new taxonomy prove useful in co-indexing two or more unique definitions, while 904 nodes serve to index just a single definition. The parsimony of the new taxonomy is thus 59%, which reveals a reasonable, if not ideal, level of representational uniformity across HowNet's semantic definitions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Parsimony of Recall", | |
| "sec_num": "5.4" | |
| }, | |
| { | |
| "text": "While just 27% of HowNet's definitions are sufficiently structured to support analogy, we are encouraged that almost all of this generative potential can be achieved with a new functional taxonomy that is straightforward and efficient to construct. Furthermore, though 27% may seem slim, these analogically-friendly {~} structures are concentrated in the areas of the HowNet taxonomy that can most benefit from analogical re-description. As revealed in Table 1 below, some areas of HowNet are clearly more amenable to analogical reasoning than others. The analogical potential of this ontologization becomes clear when one notices that it supports the classical analogy of philosopher as midwife. Clearly, then, we have just scratched the surface of what can usefully be derived from the lexico-semantic content of HowNet. Our current investigations with HowNet suggest that the full semantic richness of Chinese orthography may yet play a considerable role in supporting creative reasoning at a linguistic level, if only because it opens a window onto a different cultural perspective on words and concepts.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 453, | |
| "end": 460, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Conclusions and Future Work", | |
| "sec_num": "6" | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "WordNet: A Lexical Database for English", | |
| "authors": [ | |
| { | |
| "first": "G", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Miller", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "Communications of the ACM", | |
| "volume": "38", | |
| "issue": "11", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Miller, G. A.: WordNet: A Lexical Database for English. Communications of the ACM, Vol. 38 No. 11 (1995)", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Knowledge Description: What, How and Who?", | |
| "authors": [ | |
| { | |
| "first": "Z", | |
| "middle": [], | |
| "last": "Dong", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "The Proceedings of the International Symposium on Electronic Dictionaries", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dong, Z.: Knowledge Description: What, How and Who? The Proceedings of the Interna- tional Symposium on Electronic Dictionaries, Tokyo, Japan (1988)", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Structure-Mapping Engine: Algorithm and Examples", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Falkenhainer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Forbus", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Gentner", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "Artificial Intelligence", | |
| "volume": "41", | |
| "issue": "", | |
| "pages": "1--63", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Falkenhainer, B.; Forbus, K.; and Gentner, D.: Structure-Mapping Engine: Algorithm and Examples. Artificial Intelligence, 41, pages 1-63 (1989)", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "The Competence of Sub-Optimal Structure Mapping on 'Hard' Analogies. The proceedings of IJCAI'97, the Int. Joint Conference on Artificial Intelligence", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Veale", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Keane", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Veale, T., Keane, M. T.: The Competence of Sub-Optimal Structure Mapping on 'Hard' Analogies. The proceedings of IJCAI'97, the Int. Joint Conference on Artificial Intelli- gence, Nagoya, Japan. Morgan Kaufman, San Mateo California (1997)", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Aristotle's Poetics", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Hutton", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1982, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hutton, J.: Aristotle's Poetics. Norton, New York (1982)", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Lexical Ambiguity Resolution: Perspectives from Psycholinguistics, Neuropsychology and Artificial Intelligence", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Fass", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Fass, D: An Account of Coherence, Semantic Relations, Metonymy, and Lexical Ambigu- ity Resolution. In: Small, S. I, Cottrell, G. W., Tanenhaus, M.K. (eds.): Lexical Ambiguity Resolution: Perspectives from Psycholinguistics, Neuropsychology and Artificial Intelli- gence. Morgan Kaufman, San Mateo California (1988)", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Knowledge Representation and Metaphor. Studies in Cognitive systems", | |
| "authors": [ | |
| { | |
| "first": "E", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Way", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1991, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Way, E. C.: Knowledge Representation and Metaphor. Studies in Cognitive systems, Klu- wer Academic Publishers (1991)", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Dynamic Type Creation in Metaphor Interpretation and Analogical Reasoning: A Case-Study with WordNet", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Veale", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "the proceedings of ICCS2003, the 2003 International Conference on Conceptual Structures", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Veale. T.: Dynamic Type Creation in Metaphor Interpretation and Analogical Reasoning: A Case-Study with WordNet. In the proceedings of ICCS2003, the 2003 International Conference on Conceptual Structures, Dresden, Germany (2003)", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "WordNet sits the S.A.T.: A Knowledge-Based Approach to Lexical Analogy", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Veale", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "The proceedings of ECAI'2004, the 16th European Conf. on Artificial Intelligence", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Veale, T.: WordNet sits the S.A.T.: A Knowledge-Based Approach to Lexical Analogy. The proceedings of ECAI'2004, the 16th European Conf. on Artificial Intelligence. John Wiley: London (2004)", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Analogy Generation in HowNet", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Veale", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "the proceedings of IJCAI'05, the 19 th International Joint Conference on Artificial Intelligence", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Veale, T.: Analogy Generation in HowNet. In the proceedings of IJCAI'05, the 19 th Inter- national Joint Conference on Artificial Intelligence. Morgan Kaufmann: CA.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Fighting Arbitrariness in WordNet-like Lexical Databases -A Natural Language Motivated Remedy. The proceedings of GWC 2004, the 2 nd Global WordNet conference", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "H S" | |
| ], | |
| "last": "Wong", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wong, S.H.S.: Fighting Arbitrariness in WordNet-like Lexical Databases -A Natural Language Motivated Remedy. The proceedings of GWC 2004, the 2 nd Global WordNet conference. Edited by Sojka, Pala, Smrz, Fellbaum, Vossen (2004)", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Building Large Knowledge-Based Systems", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Lenat", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "R", | |
| "middle": [ | |
| "V" | |
| ], | |
| "last": "Guha", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lenat, D., Guha, R.V.: Building Large Knowledge-Based Systems. Addison Wesley (1990)", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "uris": null, | |
| "text": "Portion of a three-level functional hierarchy derived from HowNet", | |
| "type_str": "figure" | |
| }, | |
| "FIGREF1": { | |
| "num": null, | |
| "uris": null, | |
| "text": "illustrates a partial hierarchy derived from the HowNet semantics of various form-altering tools: A hierarchy of instruments derived from instances of AlterForm|", | |
| "type_str": "figure" | |
| }, | |
| "FIGREF2": { | |
| "num": null, | |
| "uris": null, | |
| "text": "and gladiator.|\u6253\u6597\u8005 are both co-indexed directly under the mid-level abstraction fight-agent. Likewise, gladiator|\u6253\u6597\u8005 is indexed under HaveContest-agent via fight-agent, while footballer|\u8db3\u7403\u8fd0 \u52a8 \u52a8 is indexed under HaveContest-agent via compete-agent. The upper-level of abstraction, represented here by HaveContest-agent, is necessary to facilitate analogy between semantically distant concepts.", | |
| "type_str": "figure" | |
| }, | |
| "FIGREF3": { | |
| "num": null, | |
| "uris": null, | |
| "text": "means \"knife\". This logographic compositionality affords a kind of semantic transparency on a scale that alphabetic writing systems (like that of English) simply can not match Thus, as \"philosopher\", can be seen via HowNet as a composition of \u54f2 \u5b66 (\"philosophy\") and \u5bb6 (\"specialist\" or \"scientist\"). In turn, philosophy|\u54f2\u5b66 is organized by HowNet as a specialization of knowledge|\u77e5\u8bc6 , as is logic| By decomposing compound terms in this way and generalizing the extracted modifiers, yet another threelevel taxonomy can be constructed. For instance, from these examples the partial taxonomy ofFig. 5can be derived. Portion of a three-level hierarchy derived from compound Chinese terms", | |
| "type_str": "figure" | |
| }, | |
| "TABREF1": { | |
| "num": null, | |
| "content": "<table><tr><td/><td colspan=\"2\">Humans</td><td>Artefacts</td><td>Animals</td><td>Overall</td></tr><tr><td>Coverage</td><td/><td>.65</td><td>.68</td><td>.42</td><td>.27</td></tr><tr><td>Recall</td><td/><td>.54</td><td>.58</td><td>.16</td><td>.26</td></tr><tr><td>Parsimony</td><td/><td>.50</td><td>.54</td><td>.22</td><td>.59</td></tr><tr><td>acters but of ideas, for</td><td>\u624b</td><td colspan=\"4\">\u624b \u672f means \"surgery\" and \u672f , meaning \"scalpel\", is a composite not just of char-\u5200</td></tr></table>", | |
| "html": null, | |
| "text": "Analogical coverage/recall for different areas of HowNetBut the analogical potential of HowNet resides not just in its explicit propositional semantics, but in its use of Chinese orthography. Consider that most Chinese entries in HowNet are multi-character terms, where each character is not so much a letter as a morpheme. . For instance,", | |
| "type_str": "table" | |
| } | |
| } | |
| } | |
| } |