{ "paper_id": "C10-1038", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T12:54:58.543951Z" }, "title": "Enriching Dictionaries with Images from the Internet -Targeting Wikipedia and a Japanese Semantic Lexicon: Lexeed", "authors": [ { "first": "Sanae", "middle": [], "last": "Fujita", "suffix": "", "affiliation": { "laboratory": "NTT Communication Science Lab", "institution": "", "location": {} }, "email": "" }, { "first": "Masaaki", "middle": [], "last": "Nagata", "suffix": "", "affiliation": { "laboratory": "NTT Communication Science Lab", "institution": "", "location": {} }, "email": "nagata.masaaki@lab.ntt.co.jp" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "We propose a simple but effective method for enriching dictionary definitions with images based on image searches. Various query expansion methods using synonyms/hypernyms (or related words) are evaluated. We demonstrate that our method is effective in obtaining highprecision images that complement dictionary entries, even for words with abstract or multiple meanings.", "pdf_parse": { "paper_id": "C10-1038", "_pdf_hash": "", "abstract": [ { "text": "We propose a simple but effective method for enriching dictionary definitions with images based on image searches. Various query expansion methods using synonyms/hypernyms (or related words) are evaluated. We demonstrate that our method is effective in obtaining highprecision images that complement dictionary entries, even for words with abstract or multiple meanings.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "The Internet is an immense resource for images. If we can form connections between these images and dictionary definitions, we can create rich dictionary resources with multimedia information. Such dictionaries have the potential to provide educational (Popescu et al., 2006) , crosslangauge information retrieval (Hayashi et al., 2009) or assistive communication tools especially for children, language learners, speakers of different languages, and people with disabilities such as dyslexia (Mihalcea and Leong, 2008; Goldberg et al., 2009) .", "cite_spans": [ { "start": 253, "end": 275, "text": "(Popescu et al., 2006)", "ref_id": "BIBREF16" }, { "start": 314, "end": 336, "text": "(Hayashi et al., 2009)", "ref_id": "BIBREF10" }, { "start": 493, "end": 519, "text": "(Mihalcea and Leong, 2008;", "ref_id": "BIBREF14" }, { "start": 520, "end": 542, "text": "Goldberg et al., 2009)", "ref_id": "BIBREF9" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Additionally, a database of typical images connected to meanings has the potential to fill the gaps between images and meanings (semantic gap). There are many studies which aim to cross the semantic gap (Ide and Yanai, 2009; Smeulders et al., 2000; Barnard et al., 2003) from the point of view of image recognition. However the semantic classes of target images are limited (e.g. Caltech-101, 256 1 ). Yansong and Lapata (2008) tried to construct image databases annotated with keywords from Web news images with their captions and articles, though the semantic coverage is unknown. In this paper, we aim to supply several suitable images for dictionary definitions. We propose a simple but effective method based on an Internet image search.", "cite_spans": [ { "start": 203, "end": 224, "text": "(Ide and Yanai, 2009;", "ref_id": "BIBREF11" }, { "start": 225, "end": 248, "text": "Smeulders et al., 2000;", "ref_id": "BIBREF17" }, { "start": 249, "end": 270, "text": "Barnard et al., 2003)", "ref_id": "BIBREF0" }, { "start": 402, "end": 427, "text": "Yansong and Lapata (2008)", "ref_id": "BIBREF20" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "There have been several studies related to supplying images for a dictionary or thesaurus. Bond et al. (2009) applied images obtained from the Open Clip Art Library (OCAL) to Japanese Word-Net. 2 They obtained candidate images by comparing the hierarchical structures of OCAL and Word-Net, and then judged whether or not the image was suitable for the synset by hand. OCAL benefits from being in the public domain; however, it cannot cover a wide variety of meanings because of the limited number of available images. Fujii and Ishikawa (2005) collected images and text from the Internet by querying lemma, and linked them to an open encyclopedia, CY-CLONE. 3 They guessed the meaning of the images by disambiguating the surrounding text. This is a straightforward approach, but it is difficult to use it to collect images with minor meanings, because in most cases the Internet search querying lemma only provides images related to the most common meaning. For example, lemma y arch may mean ''architecture'' or ''home run'' in Japanese, but a lemma search provided no image of the latter at least in the top 500.", "cite_spans": [ { "start": 91, "end": 109, "text": "Bond et al. (2009)", "ref_id": "BIBREF1" }, { "start": 518, "end": 543, "text": "Fujii and Ishikawa (2005)", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "There are some resources which link images to target synsets selected from WordNet (Fellbaum, 1998) . For example, PicNet (Borman et al., 2005) , ImageNet (Deng et al., 2009) and image ontology (Popescu et al., 2006 (Popescu et al., , 2007 Zinger et al., 2006) collect candidate images from the Internet. PicNet and ImageNet ask Web users to judge their suitability, and Zinger et al. (2006) ; Popescu et al. (2007) automatically filtered out unsuitable images using visual characteristics. These approaches can", "cite_spans": [ { "start": 83, "end": 99, "text": "(Fellbaum, 1998)", "ref_id": null }, { "start": 122, "end": 143, "text": "(Borman et al., 2005)", "ref_id": "BIBREF3" }, { "start": 155, "end": 174, "text": "(Deng et al., 2009)", "ref_id": "BIBREF5" }, { "start": 194, "end": 215, "text": "(Popescu et al., 2006", "ref_id": "BIBREF16" }, { "start": 216, "end": 239, "text": "(Popescu et al., , 2007", "ref_id": "BIBREF15" }, { "start": 240, "end": 260, "text": "Zinger et al., 2006)", "ref_id": "BIBREF21" }, { "start": 371, "end": 391, "text": "Zinger et al. (2006)", "ref_id": "BIBREF21" }, { "start": 394, "end": 415, "text": "Popescu et al. (2007)", "ref_id": "BIBREF15" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\uf8ee \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8f0 INDEX y arch (POS: noun) SENSE 1 \uf8ee \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8f0 DEFINITION \u00ea 1 k 1 G 4 1 D 0 8 m 1 W8 9 6G 3 m \u00b6 1 T\u00f9 2", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Buildings with bow-shaped top. Or its architectural style. EXAMPLE", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "G 2 \u00cb 1 H \u00cd = G y 1 @ wo 4 ? d", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "That bridge has 2 arches.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "HYPERNYM m 1 building, T\u00f9 2 style SEM. CLASS 865:house (main building) (\u2282 2:concrete ), 2435:pattern, method (\u2282 1000:abstract )", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\uf8f9 \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fb \uf8ee \uf8f0 IMAGE \uf8f9 \uf8fb SENSE 3 \uf8ee \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8ef \uf8f0 DEFINITION 1 @ \u00a7\u00c2D 1 \u00b2\u2022\u00c0\u00ca 1 A home run in baseball. EXAMPLE \u00a7 1 % \u00c0{ 4 \u00ca 2 D U 3 G y 3 k i< 4 8", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "A batter blasted the ball over the right-field wall.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "HYPERNYM \u00a7\u00c2D 1 honruida SYNONYM \u00b2\u2022\u00c0\u00ca 1 home run, DOMAIN 1 baseball SEM. CLASS 1680:sport (\u2282 1000:abstract ) \uf8f9 \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fb \uf8ee \uf8ef \uf8f0 IMAGE \uf8f9 \uf8fa \uf8fb \uf8f9 \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fa \uf8fb Figure 1: Simplified Entry for Lexeed & Hinoki:y arch", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "collect a large number of highly accurate images. However, target synsets are limited at present, and the coverage of polysemous words is unknown. We present a comparison with ImageNet and image ontology (Popescu et al., 2006) in \u00a7 3. In this paper, to cover a broad range of meanings, we use an Internet search. In advance, we expand the number of queries per meaning using information extracted from definition sentences. In \u00a7 3, we investigate the usability and effectiveness of several types of information targeting two different types of dictionaries, a Japanese Semantic Lexicon: Lexeed and a Web Dictionary: Japanese Wikipedia 4 ( \u00a7 2). We show that our method is simple but effective. We also analyze senses that are difficult to portray using images.", "cite_spans": [ { "start": 204, "end": 226, "text": "(Popescu et al., 2006)", "ref_id": "BIBREF16" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "We use Lexeed, a Japanese Semantic Lexicon (Kasahara et al., 2004) as a target dictionary (see Figure 1 ). Lexeed includes the 29,000 most familiar words in Japanese, split into 48,000 senses. Each entry contains the word itself and its part of speech (POS) along with definition and example sentences and links to the Goi-Taikei (GT) Japanese Ontology (Ikehara et al., 1997) . In addition, we extracted related words such as hypernyms, synonyms, and domains, from the defini-4 http://ja.wikipedia.org/ ", "cite_spans": [ { "start": 43, "end": 66, "text": "(Kasahara et al., 2004)", "ref_id": "BIBREF13" }, { "start": 353, "end": 375, "text": "(Ikehara et al., 1997)", "ref_id": "BIBREF12" } ], "ref_spans": [ { "start": 95, "end": 103, "text": "Figure 1", "ref_id": null } ], "eq_spans": [], "section": "Resources 2.1 Japanese Semantic Lexicon: Lexeed", "sec_num": "2" }, { "text": "We used Wikipedia's disambiguation pages, 5 as a target dictionary (see Figure 2) . A disambiguation page lists articles (eg. ''European Union'', ''Ehime University'') associated with the same lemma (eg. \"EU\"). Our goal is to provide images for each article listed. As shown in Figure 2 , they include various writing styles. Table 1 shows the sizes of Lexeed and Wikipedia's disambiguation pages, and the shared entries. Shared entries are rare, and account for less than On the other hand,", "cite_spans": [], "ref_spans": [ { "start": 72, "end": 81, "text": "Figure 2)", "ref_id": null }, { "start": 278, "end": 286, "text": "Figure 2", "ref_id": null }, { "start": 326, "end": 333, "text": "Table 1", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Web Dictionary :Japanese Wikipedia", "sec_num": "2.2" }, { "text": "Original (in Japanese) 1 '''EU''' 2 * [[AJ\u00e5]] 3 * [[Europa Universalis]]\u00c1 -[[\u00a8\u00c0 {\u00ca\u00c0z]]G[[\u00d7 \u00b6 \u00b6\u00bc\u00c3 \u00be\u00ca\u2022]] 4 * [[W\u00c1d\u00d3]](Ehime University) -[[W\u00c1 z]][[\u00c3\u00bf]]Dd\u00f1 \u00a7G[[ d\u00d3]] 5 '''Eu''' 6 * [[\u00bd}\u00c4\u00ab}\u2022]]Gaed 7 * [[\u00bd\u00ac\u00a2y\u2022]] -\u00f7\"\u00d5H 8 '''eu''' 9 * [[.eu]] -AJ\u00e5G[[ 9 \u00b8{\u00ca]] 10 * [[ \u00a7\u00bd]]G[[ISO 639|ISO 639-1 \u00bd ]] Gloss 1 '''EU''' 2 * [[", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Comparison of Lexeed and Wikipedia", "sec_num": "2.3" }, { "text": "in Wikipedia, y arch has only one sense, ''architecture'' corresponding to Lexeed's y 1 arch, and has no disambiguation page.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Comparison of Lexeed and Wikipedia", "sec_num": "2.3" }, { "text": "As mentioned above, Lexeed and Wikipedia have very different types of entries and senses. This research aims to investigate the possibility of supplying appropriate images for such different senses, and a method for obtaining better images.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Comparison of Lexeed and Wikipedia", "sec_num": "2.3" }, { "text": "In this paper, we propose a simple method for supplying appropriate images for each dictionary sense of a word. We collect candidate images from the Internet by using a querying image search. To obtain images even for minor senses, we expand the query by appending queries ex-tracted from definitions for each sense.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiment to Supply Images for Word Senses", "sec_num": "3" }, { "text": "In this paper, we investigated two main types of expansion, that is, the appending of mainly synonyms (SYN), and related words including hypernyms (LNK). For information retrieval, query expansion using synonyms has been adopted in several studies (Voorhees, 1994; Fang and Zhai, 2006; Unno et al., 2008) . Our LNK is similar to methods used in Deng et al. (2009) , but we note that their goal is not to give images to polysemous words (which is our intention). Popescu et al. (2006) also used synonyms (all terms in a synset) and hypernyms (immediate supertype in WordNet), but they did not investigate the effectiveness of each expansion and they forcus only on selected object synsets.", "cite_spans": [ { "start": 248, "end": 264, "text": "(Voorhees, 1994;", "ref_id": "BIBREF19" }, { "start": 265, "end": 285, "text": "Fang and Zhai, 2006;", "ref_id": "BIBREF6" }, { "start": 286, "end": 304, "text": "Unno et al., 2008)", "ref_id": "BIBREF18" }, { "start": 345, "end": 363, "text": "Deng et al. (2009)", "ref_id": "BIBREF5" }, { "start": 462, "end": 483, "text": "Popescu et al. (2006)", "ref_id": "BIBREF16" } ], "ref_spans": [], "eq_spans": [], "section": "Experiment to Supply Images for Word Senses", "sec_num": "3" }, { "text": "We collected five candidate images for each sense from the Internet by querying an image search engine. 9 Then we manually evaluated the suitability of the image for explaining the target sense. The evaluator determined whether or not the image was appropriate (T), acceptable (M), or inappropriate (F). The evaluator also noted the reasons for F. Figure 3 shows an example for 8WF' onion.", "cite_spans": [], "ref_spans": [ { "start": 348, "end": 356, "text": "Figure 3", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "As shown in Figure 3 , the evaluator determined T, M or F for each candidate image.", "cite_spans": [], "ref_spans": [ { "start": 12, "end": 20, "text": "Figure 3", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "(1) For an image that is related but that does not explain the sense, the evaluation is F. For example, for 8WF' onion, the images of onion dishes such as (2) in Figure 3 are F. On the other hand, the images that show onions themselves such as (1), (4) and 5 One point of judgment, specifically between T and M, is whether the image is typical or not. With 8WF' onion, most typical images are similar to", "cite_spans": [], "ref_spans": [ { "start": 162, "end": 170, "text": "Figure 3", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "(2) (3) (4) (5) T (Appropriate) F (Inappropriate) M (Acceptable) T (Appropriate) T (Appropriate)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "(1), (4) and (5). The image (3) may not be typical but is helpful for understanding, and (2) may lead to a misunderstanding if this is the only image shown to the dictionary user. This is why (3) is judged to be M and (2) is judged to be F.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "We evaluated 200 target senses for Lexeed, and 100 for Wikipedia. 10", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experimental and Evaluation Method", "sec_num": "3.1" }, { "text": "In this paper, we expand queries using the Hinoki Ontology , which includes related words extracted from the definition sentences. Table 2 shows the data for the Hinoki Ontology.", "cite_spans": [], "ref_spans": [ { "start": 131, "end": 138, "text": "Table 2", "ref_id": "TABREF2" } ], "eq_spans": [], "section": "Experiment: Lexeed", "sec_num": "3.2" }, { "text": "For SYN, we expand queries using synonyms, abbreviations, other names in Table 2 , and vari- 10 We performed an image search in September 2009 for Lexeed, and in December 2009 for Wikipedia.", "cite_spans": [ { "start": 93, "end": 95, "text": "10", "ref_id": null } ], "ref_spans": [ { "start": 73, "end": 80, "text": "Table 2", "ref_id": "TABREF2" } ], "eq_spans": [], "section": "Experiment: Lexeed", "sec_num": "3.2" }, { "text": "ant spellings found in the dictionary. On the other hand, for LNK, we use all the remaining relations, namely hypernyms, domains, etc. Additionally, we use only normal spellings with no expansion, when the target words are monosemous (MONO). One exception should be noted. When the normal spelling employs hiragana (Japanese syllabary characters), we expand it using a variant spelling. For example, AlU dragonfly is expanded by the variant spelling \u00c0\u00a8dragonfly.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiment: Lexeed", "sec_num": "3.2" }, { "text": "To investigate the trends and difficulties based on various conditions, we split the Lexeed senses into four types, namely, concrete and monosemous (MC), or polysemous (PC), not concrete and monosemous (MA), or polysemous (PA). We selected 50 target senses for evaluation randomly for each type. The target senses were randomly selected without distinguishing them in terms of their POS.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiment: Lexeed", "sec_num": "3.2" }, { "text": "Note that we regard the sense as being something concrete that is linked to GT's semantic classes subsumed by 2:concrete , such as 8WF' onion (\u2282 677:crop/harvest/farm products \u2282 2:concrete ). Table 3 shows the ratio of T (appropriate), M (acceptable) and F (inappropriate) images for the target sense. We calculated the ratio using all five candidate images, for example, in Figure 3 , the ratio of appropriate images is 60 % (three of five).", "cite_spans": [], "ref_spans": [ { "start": 192, "end": 199, "text": "Table 3", "ref_id": "TABREF5" }, { "start": 375, "end": 383, "text": "Figure 3", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "Experiment: Lexeed", "sec_num": "3.2" }, { "text": "In Table 3 , the baseline shows a case where the query only involves the lemma (normal spelling). As shown in Table 3 , SYN has higher precision than LNK. This means that SYN can focus on the appropriate sense. With polysemous words (PC, PA), expansion works more effectively, and helps to supply appropriate images for each sense. However, with MC, both LNK and SYN have less precision. This is because the target senses of MC are majorities, so expansion is adversely affected. Although MONO alone has good precision, because hiragana is often used as readings and has high ambiguity, appending the variant spelling helps us to focus on the appropriate sense.", "cite_spans": [], "ref_spans": [ { "start": 3, "end": 10, "text": "Table 3", "ref_id": "TABREF5" }, { "start": 110, "end": 117, "text": "Table 3", "ref_id": "TABREF5" } ], "eq_spans": [], "section": "Results and Discussion: Lexeed", "sec_num": "3.3" }, { "text": "Here, we focus on LNK of PC, and then analyze the reasons for F ( Table 5) . In Table 5 , in 24.3% of cases it is \"difficult to portray the sense using images\" (The numbers of senses for which it is \"difficult to portray the sense using images\" are, 3 of MC, 9 of PC, 10 of MA, and 16 of PA. We investigate such senses in more detail in \u00a7 3.4.).", "cite_spans": [], "ref_spans": [ { "start": 66, "end": 74, "text": "Table 5)", "ref_id": "TABREF7" }, { "start": 80, "end": 87, "text": "Table 5", "ref_id": "TABREF7" } ], "eq_spans": [], "section": "Results and Discussion: Lexeed", "sec_num": "3.3" }, { "text": "For such senses, no method can provide suitable images, as might be expected. Therefore, we exclude targets where it is \"difficult to portray the sense using images\", then we recalculated the ratio of appropriate images. Table 4 shows the capability of our proposed method for senses that can be explored using images. This leads to 66.3 % precision (15.3% improvement) even for most difficult target type, PA.", "cite_spans": [], "ref_spans": [ { "start": 221, "end": 228, "text": "Table 4", "ref_id": "TABREF6" } ], "eq_spans": [], "section": "Results and Discussion: Lexeed", "sec_num": "3.3" }, { "text": "Again, when we look at Table 5 , reasons 2-5 (33.3 %) will be improved. In particular, \"hypernym leads to ambiguity\" makes up more than 10%. Hypernyms sometimes work well, but sometimes they lead to other words included in the hypernyms. For example, appending the hypernym \u00d9 foods to 0 boiled-dried fish leads to images of \"foods made with boiled-dried fish\". This is why SYN obtained better results than LNK. Then, with \"expanded by minor sense\" and when the original sense is dominant majority, expansion reduced the precision. Therefore, we should expand using only words with major senses.", "cite_spans": [], "ref_spans": [ { "start": 23, "end": 30, "text": "Table 5", "ref_id": "TABREF7" } ], "eq_spans": [], "section": "Results and Discussion: Lexeed", "sec_num": "3.3" }, { "text": "As described above, the target senses are randomly selected without being distinguished by their POS, because we also want to investigate the features of senses that can be shown by images. Table 6 shows the ratio of senses judged as \"difficult to portray the sense using images\" (labeled as \"Not Shown\") for each POS. As regards POS, the majority of selected senses are nouns, followed by verbal nouns and verbs. We expected that the majority of nouns and verbal nouns whould be \"Shown\", but did not expect that a majority of verb is also \"Shown\". Other POSs are too rare to judge, although they tend to fall in the \"Not Shown\" category. Furthermore, in Table 7 , for nouns and verbal nouns, we show the ratio of senses for each type (\"Concrete\" or \"not Concrete\") judged in terms of \"difficult to portray the sense using images\". We classified the senses into \"Concrete\" or \"not Concrete\" based on GT's semantic classes, as described in \u00a7 3.2. As shown in Table 7 , 90.5 % of \"Concrete\" nouns are judged as \"Shown\", and only 9.5 % of senses are judged as \"Not Shown\" 11 . However 68.8 % of \"not Concrete\" nouns are also judged as \"Shown\".", "cite_spans": [], "ref_spans": [ { "start": 190, "end": 197, "text": "Table 6", "ref_id": "TABREF3" }, { "start": 655, "end": 662, "text": "Table 7", "ref_id": "TABREF4" }, { "start": 958, "end": 965, "text": "Table 7", "ref_id": "TABREF4" } ], "eq_spans": [], "section": "Discussion: Senses can/cannot be shown by images", "sec_num": "3.4" }, { "text": "Therefore, both POS and type (\"Concrete\" or \"not Concrete\") are helpful, but not perfect features as regards knowing the sense is \"difficult to portray the sense using images\". In future work we will undertake further analysis to determine the critical features.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion: Senses can/cannot be shown by images", "sec_num": "3.4" }, { "text": "For LNK we use the Wikipedia hyperlinks (shown as [[ ] With SYN, we use synonyms extracted with heuristics. Table 8 shows the main rules that we used to extract synonyms. We extracted synonyms for 98.0 % of 197,912 senses.", "cite_spans": [ { "start": 50, "end": 54, "text": "[[ ]", "ref_id": null } ], "ref_spans": [ { "start": 108, "end": 115, "text": "Table 8", "ref_id": "TABREF8" } ], "eq_spans": [], "section": "Experiment: Wikipedia", "sec_num": "3.5" }, { "text": "Then we randomly selected 50 target senses for evaluation from lemmas shared/unshared by Lexeed.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Experiment: Wikipedia", "sec_num": "3.5" }, { "text": "We do not show the baseline in Table 9 , but it is always below 10%. For all target senses, expansion provides more suitable images. Because there are so many senses in Wikipedia, no target sense is in the majority. As shown in Table 9 , there are few differences between SYN and LNK, because most of the synonyms used for SYN are also links. However, SYN has slightly superior precision as regards T (Appropriate), which means the process of extracting synonyms helped to reject links that were poorly with the target senses.", "cite_spans": [], "ref_spans": [ { "start": 31, "end": 38, "text": "Table 9", "ref_id": "TABREF9" }, { "start": 228, "end": 235, "text": "Table 9", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Results and Discussion: Wikipedia", "sec_num": "3.6" }, { "text": "Also in Lexeed, expansion using synonyms (SYN) had higher precision than hypernyms (LNK). Because we do not know the total number of suitable images for the target senses on the Internet, we cannot estimate the recall with this evaluation method. However, we speculate that hypernyms 11 For example, \u00d3 conference ( \u2282 373:organization, etc. \u2282 2:concrete ), )bhc parental surrogate ( \u2282 342:agent/representative \u2282 2:concrete ), and so on. provide higher recall. Deng et al. (2009) undertook expansion using hypernyms and this may be an appropriate way to obtain many more images for each sense. However, because our aim is employ several suitable images for each sense, high precision is preferable to high recall.", "cite_spans": [ { "start": 284, "end": 286, "text": "11", "ref_id": null }, { "start": 459, "end": 477, "text": "Deng et al. (2009)", "ref_id": "BIBREF5" } ], "ref_spans": [], "eq_spans": [], "section": "Results and Discussion: Wikipedia", "sec_num": "3.6" }, { "text": "Now, we focus on LNK shared by Lexeed, and then we analyze the reasons for F (Table 10) . In contrast to Lexeed, no sense is classified as \"difficult to portray the sense using images\". However, there are many senses where it is difficult to decide what kind of images \"explain the target sense\".", "cite_spans": [], "ref_spans": [ { "start": 77, "end": 87, "text": "(Table 10)", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Results and Discussion: Wikipedia", "sec_num": "3.6" }, { "text": "For example, in Table 10 , with \"maybe T (Appropriate)\", the target sense was a personal name and the image was his/her representative work. In this paper, for personal names, only the images of the person are judged to be T, despite the fact that supplying images of representative work for novelists or artists may be suitable.", "cite_spans": [], "ref_spans": [ { "start": 16, "end": 24, "text": "Table 10", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Results and Discussion: Wikipedia", "sec_num": "3.6" }, { "text": "In this study, we obtained five images per sense, but only one image was sufficient for some senses, for example, an image of an album cover for the name of an album. In contrast, several different types of images are needed for some senses. For example, for the name of a city, images of maps, landscapes, city offices, symbols of the city, etc. are all suitable. Therefore, it may be better to estimate a rough class first, such as the name of an album, artist and place, and then obtain preassigned types of images.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Results and Discussion: Wikipedia", "sec_num": "3.6" }, { "text": "The goal of this work was to supply several suitable images for dictionary definitions. The target dictionaries were Lexeed and Wikipedia, which have very different characteristics. To cover a wide range of senses, we collected candidate images from the Internet by querying an image search engine. Then, to obtain suitable and different images for each sense, we expanded the queries by appending related words extracted from the definition sentences. In this paper, we tried two types of expansion, one mainly using synonyms (SYN), and one mainly using hypernyms or related links (LNK).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusions", "sec_num": "4" }, { "text": "The results show that SYN provided better precision than LNK, especially for Lexeed. Also, query expansion provided a substantial improvement for polysemous words. Our proposed method is simple but effective for our purpose, that is supplying suitable and different images for each sense.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusions", "sec_num": "4" }, { "text": "In future work we intend to analyze senses that are difficult/easy to portray using images in more detail, using not only semantic charactaristics but also visual features (Csurka et al., 2004) . We also intend to improve the expansion method. One way to achieve this is to filter out expansions with minor senses. As for Wikipedia, we should approximate the class first, such as the name of an album, artist and place, then obtain preassigned types of images.", "cite_spans": [ { "start": 172, "end": 193, "text": "(Csurka et al., 2004)", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "Conclusions", "sec_num": "4" }, { "text": "http://www.vision.caltech.edu/Image Datasets/Caltech101, 256/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "http://nlpwww.nict.go.jp/wn-ja/ 3 http://cyclone.cl.cs.titech.ac.jp/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Version 20091011.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Shared lemmas are 6I buckwheat noodle, {\u00c2 cycle, \u00ac\u00c4} owl, etc. 7 Lemmas only in Wikipedia are {\u00ae Aesop, \u00aa Biot/Veoh, Gi fall name, etc. 8 Lemmas only in Lexeed are \u00b6 pay later, \u00bd\u00b9\u00c0 humorous, e> selection, etc.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "We used Google AJAX images API, http://code.google.com/intl/ja/apis/ajaxsearch/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Matching Words and Pictures", "authors": [ { "first": "Kobus", "middle": [], "last": "Barnard", "suffix": "" }, { "first": "Pinar", "middle": [], "last": "Duygulu", "suffix": "" }, { "first": "David", "middle": [], "last": "Nando De Freitas", "suffix": "" }, { "first": "David", "middle": [], "last": "Forsyth", "suffix": "" }, { "first": "Michael", "middle": [ "I" ], "last": "Blei", "suffix": "" }, { "first": "", "middle": [], "last": "Jordan", "suffix": "" } ], "year": 2003, "venue": "Journal of Machine Learning Research", "volume": "3", "issue": "", "pages": "1107--1135", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kobus Barnard, Pinar Duygulu, Nando de Freitas, David Forsyth, David Blei, and Michael I. Jordan. 2003. Match- ing Words and Pictures. Journal of Machine Learning Research, Vol. 3, pp. 1107-1135.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Enhancing the Japanese WordNet", "authors": [ { "first": "Francis", "middle": [], "last": "Bond", "suffix": "" }, { "first": "Hitoshi", "middle": [], "last": "Isahara", "suffix": "" }, { "first": "Sanae", "middle": [], "last": "Fujita", "suffix": "" }, { "first": "Kiyotaka", "middle": [], "last": "Uchimoto", "suffix": "" }, { "first": "Takayuki", "middle": [], "last": "Kuribayashi", "suffix": "" }, { "first": "Kyoko", "middle": [], "last": "Kanzaki", "suffix": "" } ], "year": 2009, "venue": "The 7th Workshop on Asian Language Resources, in conjunction with ACL-IJCNLP-2009", "volume": "", "issue": "", "pages": "1--8", "other_ids": {}, "num": null, "urls": [], "raw_text": "Francis Bond, Hitoshi Isahara, Sanae Fujita, Kiyotaka Uchi- moto, Takayuki Kuribayashi, and Kyoko Kanzaki. 2009. Enhancing the Japanese WordNet. In The 7th Workshop on Asian Language Resources, in conjunction with ACL- IJCNLP-2009, pp. 1-8.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Acquiring an Ontology for a Fundamental Vocabulary", "authors": [ { "first": "Francis", "middle": [], "last": "Bond", "suffix": "" }, { "first": "Eric", "middle": [], "last": "Nichols", "suffix": "" }, { "first": "Sanae", "middle": [], "last": "Fujita", "suffix": "" }, { "first": "Takaaki", "middle": [], "last": "Tanaka", "suffix": "" } ], "year": 2004, "venue": "Proceedings of the 20th International Conference on Computational Linguistics: COLING-2004", "volume": "", "issue": "", "pages": "1319--1325", "other_ids": {}, "num": null, "urls": [], "raw_text": "Francis Bond, Eric Nichols, Sanae Fujita, and Takaaki Tanaka. 2004. Acquiring an Ontology for a Fundamental Vocabulary. In Proceedings of the 20th International Con- ference on Computational Linguistics: COLING-2004, pp. 1319-1325.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Pic-Net: Pictorial Representations for Illustrated Semantic Networks", "authors": [ { "first": "Andy", "middle": [], "last": "Borman", "suffix": "" }, { "first": "Rada", "middle": [], "last": "Mihalcea", "suffix": "" }, { "first": "Paul", "middle": [], "last": "Tarau", "suffix": "" } ], "year": 2005, "venue": "Proceedings of the AAAI Spring Symposium on Knowledge Collection from Volunteer Contributors", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Andy Borman, Rada Mihalcea, and Paul Tarau. 2005. Pic- Net: Pictorial Representations for Illustrated Semantic Networks. In Proceedings of the AAAI Spring Symposium on Knowledge Collection from Volunteer Contributors.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Visual categorization with bags of keypoints", "authors": [ { "first": "Gabriela", "middle": [], "last": "Csurka", "suffix": "" }, { "first": "Cedric", "middle": [], "last": "Bray", "suffix": "" } ], "year": 2004, "venue": "ECCV International Workshop on Statistical Learning in Computer Vision", "volume": "", "issue": "", "pages": "59--74", "other_ids": {}, "num": null, "urls": [], "raw_text": "Gabriela Csurka, Cedric Bray, Chris Dance, and Lixin Fan. 2004. Visual categorization with bags of keypoints. In ECCV International Workshop on Statistical Learning in Computer Vision, pp. 59-74.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "ImageNet: A Large-Scale Hierarchical Image Database", "authors": [ { "first": "Jia", "middle": [], "last": "Deng", "suffix": "" }, { "first": "Wei", "middle": [], "last": "Dong", "suffix": "" }, { "first": "Richard", "middle": [], "last": "Socher", "suffix": "" }, { "first": "Li-Jia", "middle": [], "last": "Li", "suffix": "" }, { "first": "Kai", "middle": [], "last": "Li", "suffix": "" }, { "first": "Li", "middle": [], "last": "Fei-Fei", "suffix": "" } ], "year": 2009, "venue": "IEEE Computer Vision and Pattern Recognition (CVPR)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. ImageNet: A Large-Scale Hierarchical Image Database. In IEEE Computer Vision and Pattern Recognition (CVPR).", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Semantic term matching in axiomatic approaches to information retrieval", "authors": [ { "first": "Hui", "middle": [], "last": "Fang", "suffix": "" }, { "first": "Chengxiang", "middle": [], "last": "Zhai", "suffix": "" } ], "year": 2006, "venue": "Proceedings of the 29th Annual International ACM SIGIR Conference on Research and Development in Informaion Retrieval", "volume": "", "issue": "", "pages": "115--122", "other_ids": {}, "num": null, "urls": [], "raw_text": "Hui Fang and ChengXiang Zhai. 2006. Semantic term matching in axiomatic approaches to information re- trieval. In Proceedings of the 29th Annual International ACM SIGIR Conference on Research and Development in Informaion Retrieval, pp. 115-122. ACM.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "WordNet: An Electronic Lexical Database", "authors": [], "year": 1998, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Christine Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. MIT Press.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Image Retrieval and Disambiguation for Encyclopedic Web Search", "authors": [ { "first": "Atsushi", "middle": [], "last": "Fujii", "suffix": "" }, { "first": "Tetsuya", "middle": [], "last": "Ishikawa", "suffix": "" } ], "year": 2005, "venue": "Proceedings of the International Joint Conference on Artificial Intelligence: IJCAI-2005", "volume": "", "issue": "", "pages": "1598--1599", "other_ids": {}, "num": null, "urls": [], "raw_text": "Atsushi Fujii and Tetsuya Ishikawa. 2005. Image Retrieval and Disambiguation for Encyclopedic Web Search. In Proceedings of the International Joint Conference on Ar- tificial Intelligence: IJCAI-2005, pp. 1598-1599.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Toward Text-to-Picture Synthesis", "authors": [ { "first": "Andrew", "middle": [ "B" ], "last": "Goldberg", "suffix": "" }, { "first": "Jake", "middle": [], "last": "Rosin", "suffix": "" }, { "first": "Xiaojin", "middle": [], "last": "Zhu", "suffix": "" }, { "first": "Charles", "middle": [ "R" ], "last": "Dyer", "suffix": "" } ], "year": 2009, "venue": "NIPS 2009 Mini-Symposia on Assistive Machine Learning for People with Disabilities", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Andrew B. Goldberg, Jake Rosin, Xiaojin Zhu, and Charles R. Dyer. 2009. Toward Text-to-Picture Synthe- sis. In NIPS 2009 Mini-Symposia on Assistive Machine Learning for People with Disabilities.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Utilizing Images for Assisting Cross-language Information Retrieval on the Web", "authors": [ { "first": "Yoshihiko", "middle": [], "last": "Hayashi", "suffix": "" }, { "first": "Savas", "middle": [], "last": "Bora", "suffix": "" }, { "first": "Masaaki", "middle": [], "last": "Nagata", "suffix": "" } ], "year": 2009, "venue": "International Workshop on Web Information Retrieval Support Systems", "volume": "", "issue": "", "pages": "100--103", "other_ids": {}, "num": null, "urls": [], "raw_text": "Yoshihiko Hayashi, Savas Bora, and Masaaki Nagata. 2009. Utilizing Images for Assisting Cross-language Informa- tion Retrieval on the Web. In International Workshop on Web Information Retrieval Support Systems, pp. 100-103.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Crossing the Semantic Gap : Towards the Understanding of Image and Video Contents", "authors": [ { "first": "Ichiro", "middle": [], "last": "Ide", "suffix": "" }, { "first": "Keiji", "middle": [], "last": "Yanai", "suffix": "" } ], "year": 2009, "venue": "Journal of Japanese Society for Artificial Intelligence", "volume": "24", "issue": "5", "pages": "691--699", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ichiro Ide and Keiji Yanai. 2009. Crossing the Semantic Gap : Towards the Understanding of Image and Video Con- tents. Journal of Japanese Society for Artificial Intelli- gence, Vol. 24, No. 5, pp. 691-699. (in Japanese).", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Goi-Taikei -A Japanese Lexicon. Iwanami Shoten, Tokyo. 5 volumes/CD-ROM", "authors": [ { "first": "Satoru", "middle": [], "last": "Ikehara", "suffix": "" }, { "first": "Masahiro", "middle": [], "last": "Miyazaki", "suffix": "" }, { "first": "Satoshi", "middle": [], "last": "Shirai", "suffix": "" }, { "first": "Akio", "middle": [], "last": "Yokoo", "suffix": "" }, { "first": "Hiromi", "middle": [], "last": "Nakaiwa", "suffix": "" }, { "first": "Kentaro", "middle": [], "last": "Ogura", "suffix": "" }, { "first": "Yoshifumi", "middle": [], "last": "Ooyama", "suffix": "" }, { "first": "Yoshihiko", "middle": [], "last": "Hayashi", "suffix": "" } ], "year": 1997, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Satoru Ikehara, Masahiro Miyazaki, Satoshi Shirai, Akio Yokoo, Hiromi Nakaiwa, Kentaro Ogura, Yoshifumi Ooyama, and Yoshihiko Hayashi. 1997. Goi-Taikei -A Japanese Lexicon. Iwanami Shoten, Tokyo. 5 volumes/CD-ROM.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Construction of a Japanese Semantic Lexicon: Lexeed", "authors": [ { "first": "Kaname", "middle": [], "last": "Kasahara", "suffix": "" }, { "first": "Hiroshi", "middle": [], "last": "Sato", "suffix": "" }, { "first": "Francis", "middle": [], "last": "Bond", "suffix": "" }, { "first": "Takaaki", "middle": [], "last": "Tanaka", "suffix": "" }, { "first": "Sanae", "middle": [], "last": "Fujita", "suffix": "" }, { "first": "Tomoko", "middle": [], "last": "Kanasugi", "suffix": "" }, { "first": "Shigeaki", "middle": [], "last": "Amano", "suffix": "" } ], "year": 2004, "venue": "IEICE Technical Report: 2004-NLC-159", "volume": "", "issue": "", "pages": "75--82", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kaname Kasahara, Hiroshi Sato, Francis Bond, Takaaki Tanaka, Sanae Fujita, Tomoko Kanasugi, and Shigeaki Amano. 2004. Construction of a Japanese Semantic Lexi- con: Lexeed. In IEICE Technical Report: 2004-NLC-159, pp. 75-82. (in Japanese).", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "Toward communicating simple sentences using pictorial representations", "authors": [ { "first": "Rada", "middle": [], "last": "Mihalcea", "suffix": "" }, { "first": "", "middle": [], "last": "Chee Wee", "suffix": "" }, { "first": "", "middle": [], "last": "Leong", "suffix": "" } ], "year": 2008, "venue": "Machine Translation", "volume": "22", "issue": "3", "pages": "153--173", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rada Mihalcea and Chee Wee Leong. 2008. Toward commu- nicating simple sentences using pictorial representations. Machine Translation, Vol. 22, No. 3, pp. 153-173.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Ontology Driven Content Based Image Retrieval", "authors": [ { "first": "Adrian", "middle": [], "last": "Popescu", "suffix": "" }, { "first": "Christophe", "middle": [], "last": "Millet", "suffix": "" }, { "first": "Pierre-Alain", "middle": [], "last": "Mo\u00ebllic", "suffix": "" } ], "year": 2007, "venue": "Proceedings of the ACM International Conference on Image and Video Retrieval", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Adrian Popescu, Christophe Millet, and Pierre-Alain Mo\u00ebllic. 2007. Ontology Driven Content Based Image Retrieval. In Proceedings of the ACM International Con- ference on Image and Video Retrieval.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Automatic Construction of a Grounded Multimedia Ontology of Objects to Illustrate Concepts in a Learning Process", "authors": [ { "first": "Adrian", "middle": [], "last": "Popescu", "suffix": "" }, { "first": "Christophe", "middle": [], "last": "Millet", "suffix": "" }, { "first": "Pierre-Alain", "middle": [], "last": "Mo\u00ebllic", "suffix": "" }, { "first": "Patrick", "middle": [], "last": "H\u00e8de", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Grefenstette", "suffix": "" } ], "year": 2006, "venue": "NET-TIES 2006 Conference: Advanced Educational Technologies for a Future e-Europe", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Adrian Popescu, Christophe Millet, Pierre-Alain Mo\u00ebllic, Patrick H\u00e8de, and Gregory Grefenstette. 2006. Automatic Construction of a Grounded Multimedia Ontology of Ob- jects to Illustrate Concepts in a Learning Process. In NET- TIES 2006 Conference: Advanced Educational Technolo- gies for a Future e-Europe.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Content-based Image Retrieval at the End of the Early Years", "authors": [ { "first": "W", "middle": [ "M" ], "last": "Arnold", "suffix": "" }, { "first": "Marcel", "middle": [], "last": "Smeulders", "suffix": "" }, { "first": "Simone", "middle": [], "last": "Worring", "suffix": "" }, { "first": "Amarnath", "middle": [], "last": "Santini", "suffix": "" }, { "first": "Ramesh", "middle": [], "last": "Gupta", "suffix": "" }, { "first": "", "middle": [], "last": "Jain", "suffix": "" } ], "year": 2000, "venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence", "volume": "22", "issue": "12", "pages": "1349--1380", "other_ids": {}, "num": null, "urls": [], "raw_text": "Arnold W.M. Smeulders, Marcel Worring, Simone Santini, Amarnath Gupta, and Ramesh Jain. 2000. Content-based Image Retrieval at the End of the Early Years. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, Vol. 22, No. 12, pp. 1349-1380.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Information Retrieval using Automatically Extracted paraphrases", "authors": [ { "first": "Yuya", "middle": [], "last": "Unno", "suffix": "" }, { "first": "Yusuke", "middle": [], "last": "Miyao", "suffix": "" }, { "first": "Jyunichi", "middle": [], "last": "Tujii", "suffix": "" } ], "year": 2008, "venue": "Proceedings of the 14th Annual Meeting of The Association for Natural Language Processing: NLP-2008", "volume": "", "issue": "", "pages": "123--126", "other_ids": {}, "num": null, "urls": [], "raw_text": "Yuya Unno, Yusuke Miyao, and Jyunichi Tujii. 2008. In- formation Retrieval using Automatically Extracted para- phrases. In Proceedings of the 14th Annual Meeting of The Association for Natural Language Processing: NLP- 2008, pp. 123-126. (in Japanese).", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Query Expansion using Lexical-Semantic Relations", "authors": [ { "first": "Ellen", "middle": [ "M" ], "last": "Voorhees", "suffix": "" } ], "year": 1994, "venue": "Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Informaion Retrieval", "volume": "", "issue": "", "pages": "61--69", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ellen M. Voorhees. 1994. Query Expansion using Lexical- Semantic Relations. In Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Informaion Retrieval, pp. 61-69.", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "Automatic image annotation using auxiliary text information", "authors": [ { "first": "Feng", "middle": [], "last": "Yansong", "suffix": "" }, { "first": "Mirella", "middle": [], "last": "Lapata", "suffix": "" } ], "year": 2008, "venue": "Association for Computational Linguistics", "volume": "", "issue": "", "pages": "272--280", "other_ids": {}, "num": null, "urls": [], "raw_text": "Feng Yansong and Mirella Lapata. 2008. Automatic image annotation using auxiliary text information. In Proceed- ings of ACL-08: HLT, pp. 272-280. Association for Com- putational Linguistics.", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "Clustering and semantically filtering web images to create a large-scale image ontology", "authors": [ { "first": "Svitlana", "middle": [], "last": "Zinger", "suffix": "" }, { "first": "Christophe", "middle": [], "last": "Millet", "suffix": "" }, { "first": "Benoit", "middle": [], "last": "Mathieu", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Grefenstette", "suffix": "" }, { "first": "Patrick", "middle": [], "last": "H\u00e8de", "suffix": "" }, { "first": "Pierre-Alain", "middle": [], "last": "Mo\u00ebllic", "suffix": "" } ], "year": 2006, "venue": "SPIE 18th", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Svitlana Zinger, Christophe Millet, Benoit Mathieu, Gregory Grefenstette, Patrick H\u00e8de, and Pierre-Alain Mo\u00ebllic. 2006. Clustering and semantically filtering web images to create a large-scale image ontology. In SPIE 18th", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "Annual Symposium Electronic Imaging, Internet Imaging VII", "authors": [], "year": null, "venue": "", "volume": "6061", "issue": "", "pages": "89--97", "other_ids": {}, "num": null, "urls": [], "raw_text": "Annual Symposium Electronic Imaging, Internet Imaging VII, Vol. 6061, pp. 89-97.", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "uris": null, "num": null, "text": "). 2 Analyzed by Mecab, http://mecab.sourceforge.net/ tions (called Hinoki Ontology). The images in Figure 1 are samples provided using our method." }, "FIGREF1": { "type_str": "figure", "uris": null, "num": null, "text": "Examples of Candidate Images and Evaluations for 8WF' onion" }, "FIGREF2": { "type_str": "figure", "uris": null, "num": null, "text": "in Figure 3 are T. With (3) in Figure 3, the image may show the onion itself or a field of onions, therefore the evaluation is M." }, "FIGREF3": { "type_str": "figure", "uris": null, "num": null, "text": "] in Fig 2). 95.5 % of all senses include [[ ]], 85.4 % linked to an actual page, and [[ ]] appeared 0.95 times per sense. Note that we do not use time expression links such as [[2010]] and [[1990s]]." }, "TABREF0": { "html": null, "num": null, "text": "Size of Lexeed and Japanese Wikipedia (disambiguation)", "type_str": "table", "content": "
No. Entries Senses Ave. Senses/Entry Max. Senses/Entry Monosemous Ave. Words/Definition 2Lexeed Wikipedia Shared Lemma 29,272 33,299 2,228 48,009 197,912 1 19,703 1.6 5.9 8.8 57 320 148 19,080 74 2 14.4 10.7 11.0
1 From the all 215,883 lists, we extracted lists showing senses obtained by heuristics (see lines 2,3,4,6,7,9 and 10 for Figure 2
" }, "TABREF1": { "html": null, "num": null, "text": "Wikipedia has many senses, but most of them are proper nouns. For example, in Lexeed, \u00a9\u00b5AE\u00c1 sunflower is monosemous, but in Wikipedia, 67 senses are listed, including 65 proper nouns besides ''plant'' and ''sunflower oil''.", "type_str": "table", "content": "
European Union]]
3 * [[Europa Universalis]] series -a [[histori-
cal computer game]] by [[Paradox Interactive]]
4 * [[Ehime University]] -a [[National Univer-
sity]] in [[Matsuyama]],[[Ehime Prefecture]]
5 '''Eu'''
6 * [[Europium]]'s chemical element symbol
7 * [[euphonium]] -a brass instrument
8 '''eu'''
9 * [[.eu]] -[[country-code top-level domain]]
for the European Union
10 * [[ISO 639|ISO 639-1 language code]] of
[[Basque]]
" }, "TABREF2": { "html": null, "num": null, "text": "Data for Hinoki Ontology", "type_str": "table", "content": "
Type Hypernym Synonym Domain Hyponym Meronym Abbreviation Other name Other TotalNo. 47,054 69.1 y 1 arch % Lemma 14,068 20.6 y 3 arch 1,868 2.7 y 3 arch 757 1.1 7c6 1 buy and sell 7d sell Example Related Word T\u00f9 \u00b2\u2022\u00c0\u00ca homer baseball 686 1.0 + 1 lean \u00b2\u00ee fish meat 383 0.6 R 2 A(sia) yy Asia 216 0.3 F0-X 2 shave \u00ca\u00ca plug outlet 3102 4.6^X\u00cb& 1 papillote \u00b2 fish 68,134 100
" }, "TABREF3": { "html": null, "num": null, "text": "Ratio of Senses judged as \"difficult to portray the sense using images\" for each POS", "type_str": "table", "content": "
POSShownNot Shown Total
No.% No.%No.
Noun132 85.223 14.8155
Verbal Noun15 78.94 21.119
Verb9 81.82 18.211
Affix4 57.13 42.97
Pronoun0021002
Adjective1501502
Adverb0021002
Interjection1 100001
Conjunction0011001
Total162813819200
" }, "TABREF4": { "html": null, "num": null, "text": "", "type_str": "table", "content": "
: Ratio of Concrete/Not Concrete Senses
judged as \"difficult to portray the sense using im-
ages\": for Nouns and Verbal Nouns Type Shown Not Shown Total
No.% No.% No.
Concrete114 90.5 129.5 126
Not Concrete 33 68.8 15 31.348
Total147 84.5 27 15.5 174
" }, "TABREF5": { "html": null, "num": null, "text": "Ratio of Appropriate Images for Sense (Precision): Lexeed", "type_str": "table", "content": "
Target TypeExpanding MethodF (Inappropriate) T (Appropriate) M (Acceptable) No. % No. % No. %T+M No.%Total
Mono-semous (MC) Poly-crete semous Con-(PC)SYN LNK MONO baseline SYN LNK baseline18 82 42 46 94 111 18024.0 33.5 112 36 16.8 181 18.4 171 38.7 88 44.4 92 72.0 5348.0 45.7 72.4 68.4 36.2 36.8 21.221 51 27 33 61 47 1728.0 20.8 10.8 13.2 25.1 18.8 6.857 76.0 163 66.5 208 83.2 204 81.6 149 61.3 139 55.6 70 28.075 245 250 250 243 250 250
not Con-crete semous Mono-semous (MA) Poly-(PA)SYN LNK MONO baseline SYN LNK baseline32 138 98 112 122 150 20142.7 57.5 40.0 44.8 49.0 60.2 80.721 54 98 86 64 52 3628.0 22.5 40.0 34.4 25.7 20.9 14.522 48 49 52 63 47 1229.3 20.0 20.0 20.8 25.3 18.9 4.843 57.3 102 42.5 147 60.0 138 55.2 127 51.0 99 39.8 48 19.375 240 245 250 249 249 249
" }, "TABREF6": { "html": null, "num": null, "text": "", "type_str": "table", "content": "
: Ratio of Appropriate Images for Sense (Precision), excluding senses that are difficult to portray
using images: Lexeed
Target TypeExpanding MethodF (Inappropriate) T (Appropriate) M (Acceptable) No. % No. % No. %T+M No.%Total
Mono-Con-semous (MC) Poly-crete semous (PC)SYN LNK MONO baseline SYN LNK baseline15 71 29 35 61 84 13921.4 30.9 112 36 12.3 180 14.9 170 30.8 85 40.0 89 67.8 5351.4 48.7 76.6 72.3 42.9 42.4 25.919 47 26 30 52 37 1327.1 20.4 11.1 12.8 26.3 17.6 6.355 78.6 159 69.1 206 87.7 200 85.1 137 69.2 126 60.0 66 32.270 230 235 235 198 210 205
not Con-crete semous Mono-semous (MA) Poly-(PA)SYN LNK MONO baseline SYN LNK baseline17 101 65 72 57 81 12234.0 51.8 33.3 36 33.7 47.9 72.220 54 94 85 63 52 3640.0 27.7 48.2 42.5 37.3 30.8 21.313 40 36 43 49 36 1126.0 20.5 18.5 21.5 29 21.3 6.533 66.0 94 48.2 130 66.7 128 64.0 112 66.3 88 52.1 47 27.850 195 195 809 169 169 169
" }, "TABREF7": { "html": null, "num": null, "text": "Reasons for F: PC, LNK:Lexeed \u00c1\u00ca link (\u2282 \u00c1\u00ca links, usually means lynx)", "type_str": "table", "content": "
No. Reason 1 difficult to portray the sense using images 2 hypernym leads to ambiguity 12 10.8 0 boiled-dried fish (\u2282 \u00d9 foods) No. % Example 27 24.3 ,e me ''humble expressions used for oneself'' 3 expanded by minor sense 11 9.9 4 no expansion is better 8 7.2 \u00b8\u00c0\u00b5\u00ca cameraman (\u2282 \u00d8 staff) 5 original sense is TOO minor 6 5.4 lake (\u2282 \u00a1 lake), usually means sea 6 Other 47 42.3
Total111 100
" }, "TABREF8": { "html": null, "num": null, "text": "Rules for Extracting Synonyms for SYN: Wikipedia {\u00a3 dog One of [[Oriental Zodiac]]g[[ dog]] quotation key words, \u00cd See etc. {\u00a3 dog [[Chinese character]]'s [[radical parts]], See [[u inu-bu]]", "type_str": "table", "content": "
Example
" }, "TABREF9": { "html": null, "num": null, "text": "Ratio of Appropriate Images for Sense (Precision): Wikipedia Target Expanding F (Inappropriate) T (Appropriate) M (Acceptable)", "type_str": "table", "content": "
T+M
" }, "TABREF10": { "html": null, "num": null, "text": "Reasons for F: Shared by Lexeed, LNK: Wikipedia N! fue (reading) \u00ac Hue, city name in Vietnam (available words in def.) 8 inappropriate queries 10 10.9 \u00c3\u00bc\u00c0 regular w\u00f1\u00c93C active roster \u00b9\u00ca monkey \u00b9\u00ca\u00a8\u00ca Monkey Punch", "type_str": "table", "content": "
No. ReasonNo.%Example
LemmaLinks
7 lack of queries 14 15.2 (available words in def.) 2 hypernym lead to ambiguity 5 5.4 \u00ba\u00bc cache \u00ba\u00ca geocaching 9 maybe T (Appropriate) 5 5.4 6 Other 58 63
Total92 100
" } } } }