| { |
| "paper_id": "2009", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T09:40:18.604065Z" |
| }, |
| "title": "Development of a morphological analyser for Bengali", |
| "authors": [ |
| { |
| "first": "Abu", |
| "middle": [], |
| "last": "Zaher", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Univ. of Eng. and Tech", |
| "location": { |
| "postCode": "1000", |
| "settlement": "Dhaka", |
| "country": "Bangladesh" |
| } |
| }, |
| "email": "zaher14@gmail.com" |
| }, |
| { |
| "first": "Md", |
| "middle": [], |
| "last": "Faridee", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Univ. of Eng. and Tech", |
| "location": { |
| "postCode": "1000", |
| "settlement": "Dhaka", |
| "country": "Bangladesh" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Francis", |
| "middle": [ |
| "M" |
| ], |
| "last": "Tyers", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "ftyers@dlsi.ua.es" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "This article describes the development of an open-source morphological analyser for Bengali Language using nitestate technology. First we discuss the challenges of creating a morphological analyser for a highly inectional language like Bengali and then propose a solution to that using lttoolbox, an open-source nite-state toolkit. We then evaluate the performance of our developed system and propose ways of improving it further.", |
| "pdf_parse": { |
| "paper_id": "2009", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "This article describes the development of an open-source morphological analyser for Bengali Language using nitestate technology. First we discuss the challenges of creating a morphological analyser for a highly inectional language like Bengali and then propose a solution to that using lttoolbox, an open-source nite-state toolkit. We then evaluate the performance of our developed system and propose ways of improving it further.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Bengali exhibits diglossia, one of the two major forms is called Shadhubhasha and the other is called Chalitabhasha or SCB (Standard Colloquial Bengali). The main difference between these two forms are seen in verbal inections and higher usage of tatshama words in Shadhubhasha. Though a lot of Bengali literature and formal or legal documents have been historically written in Shadhubhasha, in modern times, SCB has almost replaced it in day-to-day usage; therefore, we chose to create our morphological analyser based on SCB.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Bengali is a classier language (Samit Bhattacharya and Basu, 2005) , meaning the verb does not change its form based on the gender and number of the subject or object, rather on the tense, aspect, modality and person. Our work found Bengali having three tenses, present, past and future; two moods, indicative and imperative; four aspects, simple, continuous, perfect and habitual, which differs a little from the grammar books. We identied 59 separate inections for each of the verbs. A sample sufx list for the Bengali verb \u0995\u09b0", |
| "cite_spans": [ |
| { |
| "start": 31, |
| "end": 66, |
| "text": "(Samit Bhattacharya and Basu, 2005)", |
| "ref_id": "BIBREF13" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "[kar] `do' is shown in table 1. It should be noted that the sufx added to the verb root is highly dependent on its syllabic structure leading to six bahttp://lrc.cornell.edu/asian/courses/ bengali NLK transliteration scheme, details at http: //en.wikipedia.org/wiki/Bengali_script# Romanization_Reference.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "J.A. P\u00e9rez-Ortiz, F. S\u00e1nchez-Mart\u00ednez, F.M. Tyers (eds.)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Proceedings of the First International Workshop on Free/Open-Source Rule-Based Machine Translation, p. 43-50 Alacant, Spain, November 2009 sic inection rule for verbs (the actual number of inectional paradigms for the verbs are however much more than that).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "For creating this morphological analyser/generator we took a slightly different approach from more traditional Bengali grammars. For example, there are seven grammatical cases in Bengali traditional grammars (Chowdhury et al., 2000) , but morphologically there are only four cases.", |
| "cite_spans": [ |
| { |
| "start": 208, |
| "end": 232, |
| "text": "(Chowdhury et al., 2000)", |
| "ref_id": "BIBREF1" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\" Animacy also plays a major role in the inectional pattern for nouns. We identied that there are four We have several sub-categories like anthroponym, toponym, cognomen, organization etc. for them.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "A small number of Adjectives inect on degree of comparison. They can also inect on gender, but these can be considered sankskritism rather than general phenomena (Islam et al., 2007) (Ortiz-Rojas et al., 2005) . The data for the analyser/generator is contained in an XML le, the Bengali monolingual dictionary (monodix). This dictionary has two principal parts, the rst one is the pardef section, which contains the inection rules for a particular type of word, the other is the entry section, which contains the list of words stating which pardef they belong to. A part of pardef denition for a verb is given in gure 1. \u25cc\u09be [\u0101] n/a n/a n/a n/a n/a n/a Inf.", |
| "cite_spans": [ |
| { |
| "start": 162, |
| "end": 182, |
| "text": "(Islam et al., 2007)", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 183, |
| "end": 209, |
| "text": "(Ortiz-Rojas et al., 2005)", |
| "ref_id": null |
| }, |
| { |
| "start": 624, |
| "end": 627, |
| "text": "[\u0101]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09c7\u09a4 [t\u0113] n/a n/a n/a n/a n/a n/a Gen.", |
| "cite_spans": [ |
| { |
| "start": 3, |
| "end": 7, |
| "text": "[t\u0113]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u25cc\u09be\u09b0 [\u0101r] n/a n/a n/a n/a n/a n/a Pres. Spl. n/a ", |
| "cite_spans": [ |
| { |
| "start": 4, |
| "end": 8, |
| "text": "[\u0101r]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09bf\u25cc [i] \u09c7\u25cc\u09a8 [\u0113n] \u00f8 ! \u09bf\u25cc\u09b8 [is] \u09c7\u25cc\u09a8 [\u0113n] \u09c7\u25cc [\u0113] Pres. Cont. n/a \u09bf\u099b [chi] \u09c7\u099b\u09a8 [ch\u0113n] \u099b [", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09c7\u25cc [\u0113]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "n/a n/a n/a n/a n/a n/a Part. Cond.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09c7\u09b2 [l\u0113]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "n/a n/a n/a n/a n/a n/a Imp. Pres. n/a n/a \u25cc\u09c1 \u09a8 [un] \u00f8 \u00f8 n/a n/a Imp. Fut. n/a n/a", |
| "cite_spans": [ |
| { |
| "start": 48, |
| "end": 52, |
| "text": "[un]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09c7\u25cc\u09a8 [\u0113n] \u09c7\u25cc\u09be [\u014d] \u09bf\u25cc\u09b8 [is]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "n/a n/a ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u09c7\u09b2\u09be\u09df [gul\u014dy] /\u099f \u09c1 \u0995\u09c1 \u09c7\u09a4 [tukut\u0113]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "n/a n/a n/a n/a n/a n/a <pardef n=\"\u09ab/\u09c7\u25cc\u09b0__vblex\"> ... ... ... <e> <p> <l>\u09bf\u25cc\u09b0\u09bf\u099b\u09b2</l> <r>\u09c7\u25cc\u09b0<s n=\"vblex\"/><s n=\"past\"/><s n=\"cnt\"/><s n=\"p3\"/><s n=\"infml\"/></r> </p> <par n=\"enclitic\"/> </e> <e> <p> <l>\u09bf\u25cc\u09c7\u09b0</l> <r>\u09c7\u25cc\u09b0<s n=\"vblex\"/><s n=\"pcnd\"/></r> </p> <par n=\"enclitic\"/> </e> ... ... ... </pardef> Figure 1 : Part of paradigm denition for \u09ab/\u09c7\u25cc\u09b0__vblex <pardef n=\"enclitic\"> <!--pass-through --> <e> <p> <l /> <r /> </p> </e> <!--Enclitic \u0987, (only) --> <e> <p> <l>\u0987</l> <r><j />\u0987<s n=\"adv\" /></r> </p> </e> <!--Enclitic \u0993, (also) --> <e> <p> <l>\u0993</l> <r><j />\u0993<s n=\"adv\" /></r> </p> </e> </pardef> [th\u0101kta] ). In the following lines we concatenate the root or the umlauts with the sufxes. It is to be noted that the actual code for these two procedure involves some complex regular expression matching (e.g. detection of hashant and chandrabindu, unicode normalisation, handling irregular verbs), something which was removed from the pseudo-code for simplicity and clarity.", |
| "cite_spans": [ |
| { |
| "start": 598, |
| "end": 606, |
| "text": "[th\u0101kta]", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 299, |
| "end": 307, |
| "text": "Figure 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "A preliminary version of the verb conjugator can be found online ! which generates all the forms for a particular verb. The current morphological analyser can also be accessed online.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "As One possibility is porting this analyser (the python scripts) to a more robust architecture such as foma (Huld\u00e9n, 2009) , an open-source implementation of the Xerox nite-state tools (Beesley and Karttunen, 2003) . It should be noted that there has been similar attempts to create nite-state technology based morphological analyser for Bengali with PC-KIMMO (Dasgupta and Khan, 2004) and JKimmo (Islam and Khan, 2006 ", |
| "cite_spans": [ |
| { |
| "start": 108, |
| "end": 122, |
| "text": "(Huld\u00e9n, 2009)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 185, |
| "end": 214, |
| "text": "(Beesley and Karttunen, 2003)", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 360, |
| "end": 385, |
| "text": "(Dasgupta and Khan, 2004)", |
| "ref_id": null |
| }, |
| { |
| "start": 397, |
| "end": 418, |
| "text": "(Islam and Khan, 2006", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\" 4 Evaluation", |
| "sec_num": null |
| }, |
| { |
| "text": "[k\u0101z n\u0101 kar\u0101] n/a n/a n/a n/a n/a n/a Inf.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u0995\u09be\u099c \u09a8\u09be \u0995\u09b0\u09be", |
| "sec_num": null |
| }, |
| { |
| "text": "[k\u0101z n\u0101 kart\u0113] n/a n/a n/a n/a n/a n/a Gen.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u0995\u09be\u099c \u09a8\u09be \u0995\u09b0\u09c7\u09a4", |
| "sec_num": null |
| }, |
| { |
| "text": "[k\u0101z n\u0101 kar\u0101r] n/a n/a n/a n/a n/a n/a ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u0995\u09be\u099c \u09a8\u09be \u0995\u09b0\u09be\u09b0", |
| "sec_num": null |
| }, |
| { |
| "text": "[k\u0101z n\u0101 kar\u0113] n/a n/a n/a n/a n/a n/a Part. Cond.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u0995\u09be\u099c \u09a8\u09be \u0995\u09c7\u09b0", |
| "sec_num": null |
| }, |
| { |
| "text": "[k\u0101z n\u0101 karl\u0113] n/a n/a n/a n/a n/a n/a Imp. Pres. n/a n/a \u0995\u09be\u099c \u0995\u09b0\u09c7\u09ac\u09a8 \u09a8\u09be ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "\u0995\u09be\u099c \u09a8\u09be \u0995\u09b0\u09c7\u09b2", |
| "sec_num": null |
| }, |
| { |
| "text": "' http://en.wikipedia.org/wiki/Zipf's_ law http://crblp.bracu.ac.bd http://www.prothom-alo.com http://wiki.apertium.org/wiki/ Speling_format", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Finite State Morphology", |
| "authors": [ |
| { |
| "first": "K", |
| "middle": [ |
| "R" |
| ], |
| "last": "Beesley", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Karttunen", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Beesley, K. R. and Karttunen, L. (2003). Finite State Morphology. CSLI Publications.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "\u09ac\u09be\u0982\u09b2\u09be \u09ad\u09be\u09b7\u09be\u09b0 \u09ac \u09be\u0995\u09b0\u09a3 (A grammar of Bengali Language). National Curriculam and Textbook Board (NCTB)", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [ |
| "M" |
| ], |
| "last": "Chowdhury", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [ |
| "M H" |
| ], |
| "last": "Chowdhury", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "I" |
| ], |
| "last": "Khalil", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [ |
| "K D" |
| ], |
| "last": "Muhammad", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Lahiri", |
| "suffix": "" |
| } |
| ], |
| "year": 2000, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Chowdhury, S. M., Chowdhury, S. M. H., Khalil, M. I., Muhammad, D. K. D., and Lahiri, S. (2000). \u09ac\u09be\u0982\u09b2\u09be \u09ad\u09be\u09b7\u09be\u09b0 \u09ac \u09be\u0995\u09b0\u09a3 (A grammar of Bengali Language). National Curriculam and Textbook Board (NCTB).", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Proceedings of the 7th International Conference on Computer and Information Technology (ICCIT2004", |
| "authors": [], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "KIMMO. Proceedings of the 7th International Conference on Computer and Information Technology (ICCIT2004, Dhaka, Bangladesh).", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Foma: a nite-state compiler and library", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Huld\u00e9n", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Huld\u00e9n, M. (2009). Foma: a nite-state compiler and library. EACL 2009, pages 2932.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Jkimmo: A multilingual computational morphology framework for pc-kimmo", |
| "authors": [], |
| "year": null, |
| "venue": "Proc. of #", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jkimmo: A multilingual computational mor- phology framework for pc-kimmo. Proc. of # http://ankur.org.bd/ $ http://socghop.appspot.com/student_ project/show/google/gsoc2009/apertium/ t124021625239", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "A light weight stemmer for Bengali and its use in a spelling checker", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [ |
| "Z" |
| ], |
| "last": "Islam", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "N" |
| ], |
| "last": "Uddin", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Khan", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Proceedings of the 1st International Conference on Digital Communications and Computer Applications (DCCA2007", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Islam, M. Z., Uddin, M. N., and Khan, M. (2007). A light weight stemmer for Bengali and its use in a spelling checker. Proceedings of the 1st International Conference on Digital Communications and Computer Applications (DCCA2007, Irdbid, Jordan).", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "The World's Major Languages", |
| "authors": [ |
| { |
| "first": "M", |
| "middle": [ |
| "H" |
| ], |
| "last": "Klaiman", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Routledge", |
| "volume": "23", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Klaiman, M. H. (2009). The World's Major Lan- guages, volume 23. Routledge, 2nd edition.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "mizaci\u00f3n eciente de transductores de letras a partir de diccionarios con paradigmas. Procesamiento del lenguaje natural", |
| "authors": [], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "mizaci\u00f3n eciente de transductores de letras a partir de diccionarios con paradigmas. Proce- samiento del lenguaje natural, (35):5157.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Inectional morphology synthesis for bengali noun, pronoun and verb systems", |
| "authors": [ |
| { |
| "first": "Samit", |
| "middle": [], |
| "last": "Bhattacharya", |
| "suffix": "" |
| }, |
| { |
| "first": "Monojit", |
| "middle": [], |
| "last": "Choudhury", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [ |
| "S" |
| ], |
| "last": "Basu", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proc. of the National Conference on Computer Processing of Bangla (NCCPB 05)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Samit Bhattacharya, Monojit Choudhury, S. S. and Basu, A. (2005). Inectional morphology synthesis for bengali noun, pronoun and verb systems. In Proc. of the National Conference on Computer Processing of Bangla (NCCPB 05), Dhaka, Bangladesh.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "type_str": "figure", |
| "num": null, |
| "text": "also referred as Bangla by its native speakers) is an Indo-Aryan language which is mainly spoken in the region of eastern South Asia (also known as Bengal) comprising Bangladesh, the Indian State of West Bengal, southern Assam and part of Tripura. A literally rich language, its evolution can be traced back to Magadhi Prakrit and Sanskrit languages. There are an estimated 230 million speakers of the language world wide, making it sixth among the most spoken languages of the world. Building a morphological analyser for Bengali using nite-state technology requires taking into account the highly inectional properties of the language and the diverse vocabulary. This diversity can be attributed to the inuence of different cultures and languages ranging from European languages like English, French, Dutch, Portuguese to Middle Eastern languages like Arabic, Persian as well as its own kin Hindi, Urdu, and Sanskrit language, resulting in a rich set of inections. Negatives and adverbs sometimes also take the form of enclitic, being another reason for higher the inection.", |
| "uris": null |
| }, |
| "FIGREF1": { |
| "type_str": "figure", |
| "num": null, |
| "text": "A paradigm denition for Bengali enclitics of this approach is we cannot directly use existing dictionary from other resources, but the benets outweighs this in most cases.As mentioned earlier, initially we took ad-vantage of the open-source English to Bengali machine translation tool called Anubadok. Anubadok served as a starting point for creating the rules section for this project, but we eventually realised that more extensive rules would be needed to build a high quality morphological analyser and generator. Although Anubadok is a functional English to Bengali machine translation system, morphological analysis was never its main focus. So we took the most basic approach, that is consulting the grammar books. Most of the inection rules come from Klaiman (2009), Chowdhury et al. (2000) and Ray et al. (1966) as well as Wikipedia. $ Most of the part-of-speech tagged data came from Anubadok, but Anubadok uses the Penn Treebank tagset % , so a converter was needed to convert the tags to our proposed tagset which is a superset & of Apertium's normal tagset. Also, each of our each lexical category carries more lexical", |
| "uris": null |
| }, |
| "FIGREF2": { |
| "type_str": "figure", |
| "num": null, |
| "text": "calculating recall and precision we also took two sets of data. First one is the list of most frequently used 1000 words (in their surface form) from CRBPL (Prothom-alo corpus). The second one is randomly selected 1000 word text excerpt from the web-crawler data of Prothomalo. The formulas for calculating recall and pre-", |
| "uris": null |
| }, |
| "FIGREF3": { |
| "type_str": "figure", |
| "num": null, |
| "text": "Pseudo code for the procedure PROCESS-VERB", |
| "uris": null |
| }, |
| "TABREF0": { |
| "type_str": "table", |
| "content": "<table/>", |
| "text": "levels of animacy encountered in Bengali text, inanimate, animate, human and elite. The animacy level of the nouns govern which case and plurality sufx are added to them. For example, in nominative case, to construct plural form the sufx \u09b0\u09be [r\u0101] and \u0997\u09a3 [gan] are added after human, elite nouns respectively, but \u09c7\u09b2\u09be [gul\u014d] is added after inanimate and animate nouns. The details are given in table 2. The inection rules for gender of nouns vary radically, they are actually more derivational than inectional. Since right now we only deal with inectional morphology, we decided to treat genders as separate words. Inectionally similar to nouns, the pronouns do not have have genders. Proper nouns are harder to deal with in Bengali, there is no special way to readily identify proper nouns, unlike in English.", |
| "html": null, |
| "num": null |
| }, |
| "TABREF1": { |
| "type_str": "table", |
| "content": "<table><tr><td/><td>tium, hence the analyser data conforms to Aper-</td></tr><tr><td/><td>tium's dix format</td></tr><tr><td/><td>-</td></tr><tr><td colspan=\"2\">ect on degree.</td></tr><tr><td/><td>Like most other languages, there are small</td></tr><tr><td colspan=\"2\">number of exceptions to all the general set of</td></tr><tr><td colspan=\"2\">rules, and these words are treated separately.</td></tr><tr><td>2</td><td>Design</td></tr><tr><td colspan=\"2\">The morphological analyser/generator was devel-</td></tr><tr><td colspan=\"2\">oped as a part of a new language pair for Aper-</td></tr><tr><td/><td>! Although no sufx is added, the pronunciation here is</td></tr><tr><td colspan=\"2\">different [kar\u0101]</td></tr><tr><td colspan=\"2\">\" http://en.wikipedia.org/wiki/Bengali_ grammar#Case</td></tr></table>", |
| "text": "", |
| "html": null, |
| "num": null |
| }, |
| "TABREF2": { |
| "type_str": "table", |
| "content": "<table><tr><td>When the transducer encounters the surface form`\u09bf\u09ab\u09b0\u09bf\u099b\u09b2' [phirchila], it outputs lemma`\u09c7\u09ab\u09b0' [ph\u0113r] along with tag vblex, past, cnt, p3, infml. Thus the morphological analyser realises 3rd person, informal, past continuous form of verb`\u09c7\u09ab\u09b0' from`\u09bf\u09ab\u09b0\u09bf\u099b\u09b2'. The benet of using Apertium (particularly lt-toolbox, the toolset responsible for managing the monodix) is its robust architecture. The primary advantage is that the analyser can also double as a morphological generator. On the other hand, given an XML based monodix it can create a com-piled dictionary which is generally faster than a normal text or database based dictionary, optimal for running small memory footprint devices. Bengali is distinct in nature as certain clitics are used to convey several adverbial meaning. For Second Third and Impersonal Familiar Informal Polite Familiar example, \u0995\u09b0 [kar] First Polite Ger.</td></tr></table>", |
| "text": "\u0986\u09aa\u09bf\u09a8 [\u0101pni] -you, but \u0986\u09aa\u09bf\u09a8\u0993 [\u0101pnio]you too. Here the clitic \u0993 is used to convey the meaning of adverb`also'. So we create a new pardef for this. A sample pardef for enclitic can be seen in gure 2. As can be seen in table 1, the lemma form that we chose for the verb is a bit different from traditional dictionary formats where the verbs are generally represented in their gerund form. We chose to use the present indenite, second person, informal inected form as the lemma. The justication for this decision lies in the fact that this inection is the shortest (in length) for every verb. Also, according to Sanskrit grammar, this form is analogous to the original stem (i.e. dhatu) after which afxes are added to create the real verb. Anubadok, # an open-source English to Bengali machine translation system from which a lot of linguistic data was derived, also uses this format for verb lemma. A potential disadvantage", |
| "html": null, |
| "num": null |
| }, |
| "TABREF5": { |
| "type_str": "table", |
| "content": "<table><tr><td/><td colspan=\"2\">Inanimate</td><td/><td>Animate</td><td colspan=\"2\">Human</td><td/><td>Elite</td></tr><tr><td/><td>Singular</td><td>Plural</td><td>Singular</td><td>Plural</td><td>Singular</td><td>Plural</td><td>Singular</td><td>Plural</td></tr><tr><td>Nominative</td><td>\u00f8/\u099f\u09be [t .\u0101 ] /\u0996\u09be\u09a8\u09be [kh\u0101n\u0101]</td><td>\u09c7\u09b2\u09be [gul\u014d] /\u099f \u09c1 \u0995\u09c1 [t . uku]</td><td>\u00f8/\u099f\u09be [t .\u0101 ]</td><td>\u09c7\u09b2\u09be [gul\u014d]</td><td>\u00f8/\u099f\u09be [t .\u0101 ]</td><td>\u09b0\u09be [r\u0101]</td><td>\u00f8</td><td>\u09b0\u09be [r\u0101] /\u0997\u09a3 [gan . a] /\u09ac\u09c3 [br . inda]</td></tr><tr><td>Objective Genitive</td><td>/\u0996\u09be\u09bf\u09a8 [kh\u0101ni] \u00f8/\u099f\u09be [t .\u0101 ] /\u0996\u09be\u09a8\u09be [kh\u0101n\u0101] /\u0996\u09be\u09bf\u09a8 [kh\u0101ni] \u09b0 [ra] /\u09c7\u25cc\u09b0 [\u0113r] /\u09c7\u09df\u09b0 [y\u0113r] /\u099f\u09be\u09b0 [t .\u0101 r]</td><td>\u09c7\u09b2\u09be [gul\u014d] /\u099f \u09c1 \u0995\u09c1 [t . uku] \u09c7\u09b2\u09be\u09b0 [gul\u014dr] /\u099f \u09c1 \u0995\u09c1 \u09b0 [t . ukur]</td><td>\u09c7\u0995 [k\u0113] /\u099f\u09be\u09c7\u0995 [t .\u0101 k\u0113] \u09b0 [ra] /\u09c7\u25cc\u09b0 [\u0113r] /\u09c7\u09df\u09b0 [y\u0113r]</td><td>\u09c7\u09b2\u09be\u09c7\u0995 [gul\u014dk\u0113] \u09c7\u09b2\u09be\u09b0 [gul\u014dr]</td><td>\u09c7\u0995 [k\u0113] /\u099f\u09be\u09c7\u0995 [t .\u0101 k\u0113] \u09b0 [ra] /\u09c7\u25cc\u09b0 [\u0113r] /\u09c7\u09df\u09b0 [y\u0113r]</td><td>\u09c7\u09a6\u09b0\u09c7\u0995 [d\u0113rk\u0113] \u09c7\u09a6\u09b0 [d\u0113r]</td><td>\u09c7\u0995 [k\u0113] \u09b0 [ra] /\u09c7\u25cc\u09b0 [\u0113r] /\u09c7\u09df\u09b0 [y\u0113r]</td><td>\u09c7\u0995 [k\u0113] /\u0997\u09a3\u09c7\u0995 [gan . k\u0113] /\u09ac\u09c3 \u09c7\u0995 [br . indak\u0113] \u09c7\u09a6\u09b0 [d\u0113r] /\u0997\u09c7\u09a3\u09b0 [gan .\u0113 r] /\u09ac\u09c3 \u09c7 \u09b0 [br . ind\u0113r]</td></tr><tr><td/><td>/\u0996\u09be\u09a8\u09be\u09b0 [kh\u0101n\u0101r]</td><td/><td>/\u099f\u09be\u09b0 [t .\u0101 r]</td><td/><td>/\u099f\u09be\u09b0 [t .\u0101 r]</td><td/><td/><td/></tr><tr><td>Locative</td><td>/\u0996\u09be\u09bf\u09a8\u09b0 [kh\u0101nir] \u09c7\u09df [y\u0113] /\u09df [y]</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>/\u09c7\u09a4 [t\u0113] /\u099f\u09be\u09df [t\u0101y]</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>/\u0996\u09be\u09a8\u09be\u09df [kh\u0101n\u0101y]</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>/\u0996\u09be\u09bf\u09a8\u09c7\u09a4 [kh\u0101nit\u0113]</td><td/><td/><td/><td/><td/><td/><td/></tr></table>", |
| "text": "Sufx table for the Bengali verb \u0995\u09b0 [kar]`do'", |
| "html": null, |
| "num": null |
| }, |
| "TABREF6": { |
| "type_str": "table", |
| "content": "<table/>", |
| "text": "Sufx table for Bengali noun inections", |
| "html": null, |
| "num": null |
| }, |
| "TABREF7": { |
| "type_str": "table", |
| "content": "<table><tr><td colspan=\"2\">Part of speech Number of entries</td><td/></tr><tr><td>Noun</td><td>2,035</td><td/></tr><tr><td>Adjective</td><td>866</td><td/></tr><tr><td>Proper noun</td><td>800</td><td/></tr><tr><td>Adverb</td><td>432</td><td/></tr><tr><td>Verb</td><td>136</td><td/></tr><tr><td>Numeral Post-position</td><td>123 46</td><td/><td>The fre-</td></tr><tr><td>Determiner Pronoun Conjunction</td><td>45 32 8 4,523</td><td colspan=\"2\">quency list of the words were obtained from CR-BLP. The list was created at CRBLP as per their Prothom-alo lexicon project , the analysis done on the frequency of Bengali words by crawling</td></tr><tr><td/><td/><td colspan=\"2\">Prothom-alo, one of the most circulated Ben-</td></tr><tr><td/><td/><td colspan=\"2\">gali newspapers in Bangladesh. We were able to</td></tr><tr><td/><td/><td colspan=\"2\">get a list of most frequently used 20,000 Bengali</td></tr><tr><td/><td/><td colspan=\"2\">words from the project.</td></tr><tr><td/><td/><td>3</td><td>Development</td></tr><tr><td/><td/><td colspan=\"2\">The development of the XML dictionary was</td></tr><tr><td/><td/><td colspan=\"2\">done using open source tools like Python, PHP,</td></tr><tr><td/><td/><td colspan=\"2\">MySQL and shell-scripting. At rst we focused</td></tr><tr><td/><td/><td colspan=\"2\">on the open category words starting with verbs</td></tr><tr><td/><td/><td colspan=\"2\">and then moving onto nouns, pronouns and ad-</td></tr><tr><td/><td/><td colspan=\"2\">jectives etc. We rst populated our database with</td></tr><tr><td/><td/><td colspan=\"2\">lemmas, then made additional changes to the ta-</td></tr><tr><td/><td/><td colspan=\"2\">bles with proper animacy, gender, number tags.</td></tr><tr><td/><td/><td colspan=\"2\">Then we put down the inection rules in our</td></tr><tr><td/><td/><td colspan=\"2\">python scripts. In some cases, we took the help</td></tr><tr><td/><td/><td colspan=\"2\">of another intermediate format called Speling for-</td></tr><tr><td/><td/><td colspan=\"2\">mat for ease of use. The scripts with the in-</td></tr><tr><td/><td/><td colspan=\"2\">ection rules generate the semi-colon delimited</td></tr><tr><td/><td/><td colspan=\"2\">les which are then used to create the nal Aper-</td></tr><tr><td/><td/><td colspan=\"2\">tium dictionary. We then use lttoolbox to compile</td></tr><tr><td/><td/><td colspan=\"2\">the text dictionary to binary format. The Speling</td></tr><tr><td/><td/><td colspan=\"2\">format does not currently support enclitics, so it</td></tr><tr><td/><td/><td colspan=\"2\">cannot be used when the words might take on ad-</td></tr><tr><td/><td/><td colspan=\"2\">ditional enclitic, e.g. verbs and nouns. In these</td></tr><tr><td/><td/><td colspan=\"2\">cases, the monodix is directly generated by the</td></tr><tr><td/><td/><td colspan=\"2\">script.</td></tr><tr><td/><td/><td/><td>Figures 3 and 4 show the pseudo code for gen-</td></tr><tr><td/><td/><td colspan=\"2\">erating the inections of verbs. In line 2 of g-</td></tr><tr><td/><td/><td colspan=\"2\">ure 3 we calculate the effective length of the</td></tr><tr><td/><td/><td colspan=\"2\">verb root by removing`\u25cc\u0981 ' (chandrabindu) and</td></tr><tr><td/><td/><td colspan=\"2\">\u25cc\u09cd ' (hashant). Then depending on the length of</td></tr><tr><td/><td/><td colspan=\"2\">the verb, we choose appropriate function to gen-</td></tr><tr><td/><td/><td colspan=\"2\">erate the actual inection. Figure 4 shows how</td></tr></table>", |
| "text": "http://en.wikipedia.org/wiki/Bengali_ grammar % http://www.cis.upenn.edu/~treebank/ TagSets data such as gender, animacy of nouns, mood and aspect of tense; we had to manually tag them. We wanted to remain faithful to Zipf's Law, ' which Anubadok does not follow. So a lot of high frequency words that were not in Anubadok's dictionary had to be manually tagged.", |
| "html": null, |
| "num": null |
| }, |
| "TABREF8": { |
| "type_str": "table", |
| "content": "<table><tr><td>we are generating the inected forms. First in</td></tr><tr><td>line 1 and 2, we generate two type of umlauts for</td></tr><tr><td>this particular kind of verb root. Normally this is</td></tr><tr><td>a regressive vowel harmony, a phonological phe-</td></tr><tr><td>nomenon attributive to SCB. But for some verbs</td></tr><tr><td>like \u09af\u09be [ y\u0101], \u0986\u099b [\u0101cha], it is rather a morphological</td></tr><tr><td>phenomenon (e.g. \u09af\u09be [ y\u0101]</td></tr></table>", |
| "text": "Number of entries in the lexicon for the main parts of speech", |
| "html": null, |
| "num": null |
| }, |
| "TABREF9": { |
| "type_str": "table", |
| "content": "<table><tr><td>shows, the number of currently tagged</td></tr><tr><td>parts of speech is 4,523. This is not a very high</td></tr><tr><td>number. However, as we shall later see, our dic-</td></tr><tr><td>tionary puts emphasis on the most frequently used</td></tr><tr><td>words, therefore achieving a higher effective cov-</td></tr><tr><td>erage.</td></tr><tr><td>Doing an evaluation poses some major chal-</td></tr><tr><td>lenges for us. As we have stated before, this</td></tr><tr><td>implementation of Bengali morphological anal-</td></tr><tr><td>yser was created keeping in mind only the SCB</td></tr><tr><td>(Standard Colloquial Bengali). However there</td></tr><tr><td>! http://xixona.dlsi.ua.es/~fran/ bengali/conj/ \" http://xixona.dlsi.ua.es/~fran/ bengali/analysis.php</td></tr></table>", |
| "text": "few digital SCB text resources available in the Internet, apart form Wikipedia and the Prothomalo website. Therefore, our evaluation was also conned to corpus developed from Wikipedia and Prothom-alo. The data was collected by running a web crawler on these two web sites.", |
| "html": null, |
| "num": null |
| }, |
| "TABREF10": { |
| "type_str": "table", |
| "content": "<table><tr><td/><td colspan=\"4\">shows the na\u00efve coverage for wikipedia</td></tr><tr><td colspan=\"5\">and prothom-alo. The statistics clearly show op-</td></tr><tr><td colspan=\"5\">timization nature of our analyser. Since the fre-</td></tr><tr><td colspan=\"5\">quency list was derived from prothom-alo, we get</td></tr><tr><td colspan=\"5\">a higher coverage (80.35%) of prothom-alo than</td></tr><tr><td colspan=\"3\">of wikipedia (68.21%).</td><td/><td/></tr><tr><td>Site</td><td>File size</td><td>Total</td><td>Recognised</td><td>Na\u00efve</td></tr><tr><td/><td>(MB)</td><td>words</td><td>words</td><td>coverage</td></tr><tr><td colspan=\"4\">Wkipedia 27.1 MB 1,730,745 1,180,542</td><td>68.21%</td></tr><tr><td>Prothom-</td><td colspan=\"3\">23.4 MB 1,572,601 1,263,661</td><td>80.35%</td></tr><tr><td>alo</td><td/><td/><td/><td/></tr></table>", |
| "text": "", |
| "html": null, |
| "num": null |
| }, |
| "TABREF11": { |
| "type_str": "table", |
| "content": "<table/>", |
| "text": "", |
| "html": null, |
| "num": null |
| }, |
| "TABREF12": { |
| "type_str": "table", |
| "content": "<table><tr><td>shows the recall and precision values</td></tr><tr><td>for the two sets of data. Precision is high in both</td></tr><tr><td>cases. Recall value falls in the latter case (random</td></tr><tr><td>excerpt), but this is still a high value.</td></tr></table>", |
| "text": "", |
| "html": null, |
| "num": null |
| }, |
| "TABREF13": { |
| "type_str": "table", |
| "content": "<table><tr><td/><td>Type</td><td>Correct Retrieved Total Precision Recall</td></tr><tr><td/><td colspan=\"2\">Top 1000 1,626 1,723 1,631 99.6% 94.37%</td></tr><tr><td colspan=\"3\">Random 1000 1,328 1,504 1,330 99.8% 88.29%</td></tr><tr><td colspan=\"3\">GEN-INFLECTION-TAKE(verb)</td></tr><tr><td colspan=\"3\">1 uml \u2190 replace`\u09c7\u25cc' with`\u09bf\u25cc' and`\u09c7\u25cc\u09be' with`\u25cc\u09c1 ' in verb</td></tr><tr><td colspan=\"3\">2 uml2 \u2190 replace`\u09c7\u25cc' with`\u25cc\u09be' in verb</td></tr><tr><td>3</td><td colspan=\"2\">I[Ger.] \u2190 verb + \u0993\u09df\u09be</td></tr><tr><td>4</td><td colspan=\"2\">I[Inf.] \u2190 uml + \u09c7\u09a4</td></tr><tr><td>5</td><td colspan=\"2\">I[Gen.] \u2190 verb + \u09ac\u09be\u09b0</td></tr><tr><td>14</td><td colspan=\"2\">I[P art.P ast] \u2190 uml + \u09c7\u09df</td></tr><tr><td>15</td><td colspan=\"2\">I[P art.Cond.] \u2190 uml + \u09c7\u09b2</td></tr><tr><td>16</td><td colspan=\"2\">I[Figure 4: Pseudo code for procedure GEN-INFLECTION-TAKE</td></tr></table>", |
| "text": "I[F ut.Spl.] \u2190 verb + \u09ac , verb + \u09c7\u09ac\u09a8 , verb + \u09c7\u09ac , uml + \u09bf\u09ac , verb+ \u09c7\u09ac\u09a8 , verb + \u09c7\u09ac 9 I[P ast.Spl.] \u2190 uml + \u09b2\u09be\u09ae , uml + \u09c7\u09b2\u09a8 , uml + \u09c7\u09b2 , uml + \u09bf\u09b2 , uml + \u09c7\u09b2\u09a8 , uml + \u09b2 10 I[P ast.Hbl.] \u2190 uml + \u09a4\u09be\u09ae , uml + \u09c7\u09a4\u09a8 , uml + \u09c7\u09a4 , uml + \u09bf\u09a4 , uml + \u09c7\u09a4\u09a8 , uml + \u09a4 11 I[P ast.Cont.] \u2190 uml + \u09bf \u09b2\u09be\u09ae , uml + \u09bf \u09c7\u09b2\u09a8 , uml + \u09bf \u09c7\u09b2 , uml + \u09bf \u09bf\u09b2 , uml + \u09bf \u09c7\u09b2\u09a8 , uml + \u09bf\u09b2 12 I[P erf.] \u2190 uml + \u09c7\u09df\u09bf\u099b , uml + \u09c7\u09df\u09c7\u099b\u09a8 , uml + \u09c7\u09df\u099b , uml + \u09c7\u09df\u09bf\u099b\u09b8 , uml + \u09c7\u09df\u09c7\u099b\u09a8 , uml + \u09c7\u09df\u09c7\u099b 13 I[P luP erf.] \u2190 uml + \u09c7\u09df\u09bf\u099b\u09b2\u09be\u09ae , uml + \u09c7\u09df\u09bf\u099b\u09c7\u09b2\u09a8 , uml + \u09c7\u09df\u09bf\u099b\u09c7\u09b2 , uml + \u09c7\u09df\u09bf\u099b\u09bf\u09b2 , uml + \u09c7\u09df\u09bf\u099b\u09c7\u09b2\u09a8 , uml + \u09c7\u09df\u09bf\u099b\u09b2 Imp.P res.] \u2190 verb + \u09a8 , uml2 + \u0993 , verb, verb+ \u09a8 , uml + \u0995 17 I[Imp.F ut.] \u2190 verb + \u09c7\u09ac\u09a8 , uml + \u0993 , uml + \u09b8 , verb+ \u09c7\u09ac\u09a8 , uml + \u09c7\u09ac", |
| "html": null, |
| "num": null |
| }, |
| "TABREF14": { |
| "type_str": "table", |
| "content": "<table><tr><td>5</td><td>Discussion</td></tr><tr><td colspan=\"2\">Currently the analyser is in preliminary stage and</td></tr><tr><td colspan=\"2\">there are lots of room for improvement. Firstly,</td></tr><tr><td colspan=\"2\">the coverage needs to be expanded. This will re-</td></tr><tr><td colspan=\"2\">quire manual tagging of many words. Secondly,</td></tr><tr><td colspan=\"2\">the verb section faces difculty in treating multi-</td></tr><tr><td colspan=\"2\">word verbs and the negative form are not well</td></tr><tr><td colspan=\"2\">recognised. This is because several forms of the</td></tr><tr><td colspan=\"2\">verb like innitives and participles demand a neg-</td></tr><tr><td colspan=\"2\">ative particle before the verb while nite forms</td></tr><tr><td colspan=\"2\">require the particle to follow the verb and in some</td></tr><tr><td colspan=\"2\">cases as enclitic. Matters get complicated in the</td></tr><tr><td colspan=\"2\">former case, multi-word verbs require the nega-</td></tr><tr><td colspan=\"2\">tive particle to be in the middle. Table 6 shows the</td></tr><tr><td colspan=\"2\">negative forms of a multi-word verb \u0995\u09be\u099c \u0995\u09b0`work'</td></tr><tr><td colspan=\"2\">clearly identifying the problem. This can be par-</td></tr><tr><td colspan=\"2\">tially taken care of by a nested paradigm, but this</td></tr><tr><td colspan=\"2\">makes the analyser slow, we need to come up with</td></tr><tr><td colspan=\"2\">a better solution to solve this.</td></tr></table>", |
| "text": "Precision and Recall", |
| "html": null, |
| "num": null |
| }, |
| "TABREF15": { |
| "type_str": "table", |
| "content": "<table><tr><td>\u0995\u09be\u099c \u0995\u09b0 [k\u0101z kar]</td><td>First person</td><td colspan=\"3\">could also be used as a stemmer for any search Second person Third person / Impersonal</td></tr><tr><td>Ger.</td><td>Polite</td><td>Familiar engine for # Informal</td><td>Polite</td><td>Familiar</td></tr><tr><td/><td/><td colspan=\"3\">To our knowledge, this is the rst open-</td></tr><tr><td/><td/><td colspan=\"3\">source attempt in creating a fully-functional wide-</td></tr><tr><td/><td/><td colspan=\"3\">coverage morphological analyser and generator</td></tr><tr><td/><td/><td colspan=\"3\">for Bengali that is publicly available to every-</td></tr><tr><td/><td/><td colspan=\"3\">one. All the tools that were used in this project</td></tr><tr><td/><td/><td colspan=\"3\">are open source and the output, both the linguistic</td></tr><tr><td/><td/><td colspan=\"3\">data and the toolset is available under the GNU</td></tr><tr><td/><td/><td>GPL license.</td><td/><td/></tr><tr><td colspan=\"2\">). However, PC-KIMMO is not Unicode</td><td/><td/><td/></tr><tr><td colspan=\"2\">compliant, it cannot be directly used for Ben-</td><td/><td/><td/></tr><tr><td colspan=\"2\">gali morphological analysis. On the other hand</td><td/><td/><td/></tr><tr><td colspan=\"2\">JKimmo requires a transliteration scheme. The</td><td/><td/><td/></tr><tr><td colspan=\"2\">solution we present here is Unicode compliant</td><td/><td/><td/></tr><tr><td colspan=\"2\">and requires no transliteration scheme. Further-</td><td/><td/><td/></tr><tr><td colspan=\"2\">more, foma is fully able to handle Unicode char-</td><td/><td/><td/></tr><tr><td colspan=\"2\">acters, thus maintaining our Unicode compliance</td><td/><td/><td/></tr><tr><td colspan=\"2\">should we choose to port to it.</td><td/><td/><td/></tr><tr><td colspan=\"2\">As mentioned earlier, this morphological anal-</td><td/><td/><td/></tr><tr><td colspan=\"2\">yser/generator was created as a part of a new lan-</td><td/><td/><td/></tr><tr><td colspan=\"2\">guage pair bn-en for Apertium. We are on our</td><td/><td/><td/></tr><tr><td colspan=\"2\">way of creating a functional English to Bengali</td><td/><td/><td/></tr><tr><td colspan=\"2\">translation system. The morphological analyser</td><td/><td/><td/></tr></table>", |
| "text": "Bengali language. It could also double as a spell checker. In fact works are in the way to create a new Bengali dictionary for Firefox and OpenOfce.Org using this morphological analyser by Ankur.", |
| "html": null, |
| "num": null |
| }, |
| "TABREF18": { |
| "type_str": "table", |
| "content": "<table/>", |
| "text": "Example of negative inections of a compound verb 9th International Conference on Computer and Information Technology (ICCIT 2006), Dhaka, Bangladesh.", |
| "html": null, |
| "num": null |
| } |
| } |
| } |
| } |
