| { |
| "paper_id": "M98-1016", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T03:16:02.546455Z" |
| }, |
| "title": "DESCRIPTION OF THE KENT RIDGE DIGITAL LABS SYSTEM USED FOR MUC-7", |
| "authors": [ |
| { |
| "first": "Shihong", |
| "middle": [], |
| "last": "Yu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kent Ridge Digital Labs", |
| "location": { |
| "addrLine": "21 Heng Mui Keng", |
| "postCode": "119613", |
| "settlement": "Terrace", |
| "country": "Singapore" |
| } |
| }, |
| "email": "shyu@krdl.org.sg" |
| }, |
| { |
| "first": "Shuanhu", |
| "middle": [], |
| "last": "Bai", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kent Ridge Digital Labs", |
| "location": { |
| "addrLine": "21 Heng Mui Keng", |
| "postCode": "119613", |
| "settlement": "Terrace", |
| "country": "Singapore" |
| } |
| }, |
| "email": "bai@krdl.org.sg" |
| }, |
| { |
| "first": "Paul", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "Kent Ridge Digital Labs", |
| "location": { |
| "addrLine": "21 Heng Mui Keng", |
| "postCode": "119613", |
| "settlement": "Terrace", |
| "country": "Singapore" |
| } |
| }, |
| "email": "paulwu@krdl.org.sg" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "We aim to build a single simple framework for tasks in text information extraction, for which, to a certain extent, the required information can be resolved locally. Our system is statistics-based. As usual, language model is built from training corpus. This is the so-called learning process. Much e ort has been spent to absorb domain knowledge in the language model in a systematic and generic way, because the system is designed not for one particular task, but for general local information extraction. For the information extraction part tagging, the system consists of the following modules: be-own-by-Richard PERSON Branson PERSON ,-chairman-of-Virgin ORG Atlantic ORG Airways ORG ;-Grouping all adjacent words with tag PERSON gives a person name, grouping those with tag ORG gives an organization name, etc. The problem becomes, for any given sequence of words w = w 1 w 2 : : : w n , nding the tags t = t 1 t 2 : : : t n correspondingly. Note that there are di erent ways of assigning tags. For the above example, tags can also be: Example 1: The-British-balloon- ,-called-the-Virgin-Global-Challenger- ,-is-to-beown-by-Richard PERSON-start Branson PERSON-end ,-chairman-of-Virgin ORG-start Atlantic ORG-continue Airways ORG-end ;-This way, extra information such as common surnames, rst names, organization endings Corp., Inc. etc and so on can beobtained. It is observed that di erent tags for a same task make di erence. We feel that choosing an appropriate tag set is a problem worthy o f careful investigation. Intuitively, a tag set for a particular task must be: su cient, meaning that the information extracted must be su cient for the task; and e cient, meaning that there should be no redundant and nonrelevant information.", |
| "pdf_parse": { |
| "paper_id": "M98-1016", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "We aim to build a single simple framework for tasks in text information extraction, for which, to a certain extent, the required information can be resolved locally. Our system is statistics-based. As usual, language model is built from training corpus. This is the so-called learning process. Much e ort has been spent to absorb domain knowledge in the language model in a systematic and generic way, because the system is designed not for one particular task, but for general local information extraction. For the information extraction part tagging, the system consists of the following modules: be-own-by-Richard PERSON Branson PERSON ,-chairman-of-Virgin ORG Atlantic ORG Airways ORG ;-Grouping all adjacent words with tag PERSON gives a person name, grouping those with tag ORG gives an organization name, etc. The problem becomes, for any given sequence of words w = w 1 w 2 : : : w n , nding the tags t = t 1 t 2 : : : t n correspondingly. Note that there are di erent ways of assigning tags. For the above example, tags can also be: Example 1: The-British-balloon- ,-called-the-Virgin-Global-Challenger- ,-is-to-beown-by-Richard PERSON-start Branson PERSON-end ,-chairman-of-Virgin ORG-start Atlantic ORG-continue Airways ORG-end ;-This way, extra information such as common surnames, rst names, organization endings Corp., Inc. etc and so on can beobtained. It is observed that di erent tags for a same task make di erence. We feel that choosing an appropriate tag set is a problem worthy o f careful investigation. Intuitively, a tag set for a particular task must be: su cient, meaning that the information extracted must be su cient for the task; and e cient, meaning that there should be no redundant and nonrelevant information.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Sentence segmentor and tokenizer. This module accepts a stream of characters as input, and transforms it into a sequence of sentences and tokens. The way of tokenization can vary with di erent tasks and domains. For example, most English text is tokenized in the same way, while tokenization in Chinese itself is a research topic. Text analyzer. This module provides analysis necessary for the particular task, be it semantic, syntactic, orthographic, etc. This same analyzer is also applied in the learning process. Hypothesis generator. The possibilities for each word token are determined. Rules can be captured by letting one word have one choice, as is the case in the recognition of time, date, money and percentage terms for the Chinese Named Entity NE task. These are identi ed by pattern matching rules. Disambiguation module. This is essentially implementation of Viterbi algorithm.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "All the above modules will be described in detail in the following sections.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "First of all, a brief of the modeling of the problem is in order. Each word in text is assigned a tag, information can then be obtained from tags of all words. For example, for the English NE task, Example 1: The -British -balloon -, -called -the -Virgin -Global -Challenger -, -is -to -", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "TEXT INFORMATION EXTRACTION TO TAGGING", |
| "sec_num": null |
| }, |
| { |
| "text": "Careful consideration has been given to study how to absorb domain knowledge in language models in a generic and systematic way. The basic idea is, as much as possible relevant and signi cant information to the task contained in the original corpus should retain in back-o corpora where back-o features are stored, so that correct decisions can be made from the statistics generated from the back-o corpora when they can not be done from the statistics from the original training corpus.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "LEARNING PROCESS: INFORMATION DISTILLATION OF TRAINING CORPUS Learning Process in General", |
| "sec_num": null |
| }, |
| { |
| "text": "The original training corpus is in the form of word tag, statistics about words and tags including local contextual information can be obtained. Each w ord in the corpus is given a back-o feature by the principle that the back-o features of all words should extract the most information from the corpus relevant to the particular task. The information loss is compensated by gain of generosity. A back-o corpus in the form of back-o feature tag is then generated, and statistics can be obtained in the same manner. The original corpus is processed this way for a certain number of times. Every time, a less descriptive back-o corpus which gains more in generosity is generated, and thus the corresponding statistics.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "LEARNING PROCESS: INFORMATION DISTILLATION OF TRAINING CORPUS Learning Process in General", |
| "sec_num": null |
| }, |
| { |
| "text": "For example, semantic classes can be used as back-o features for all the words in Example 1, which gives the back-o corpus of the following form: ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "LEARNING PROCESS: INFORMATION DISTILLATION OF TRAINING CORPUS Learning Process in General", |
| "sec_num": null |
| }, |
| { |
| "text": "We have a text corpus of about 500,000 words from People Daily and Xinhua News Agency, all of which w ere manually checked for both word segmentation and part of speech tagging. In addition, we have a lexicon of 89,777 words, in which 5351 words are labeled as geographic names, 304 words are people's name and 183 are organization names. 1167 words consist of more than 4 characters. The longest word meaning Great Britain and North Ireland United Kingdom\" contains 13 characters. About 50,000 di erent w ords appeared in the 500,000 words corpus. We also have three entity name lists: people name list 67,616 entries, location name list 6,451 entries and organization name list 6190 entries. Observation: Problems and Solutions 1. Intuitively, case information of proper names in English writing system provides good indication about locations and boundaries of entity names. There are successful systems 2 which are built upon this intuition. Unfortunately, the uniformity of character string in Chinese writing system does not contain such information.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for Chinese NE Training Corpus and Supporting Resources", |
| "sec_num": null |
| }, |
| { |
| "text": "One should look for such analogous indicative characteristics which may be unique in Chinese language.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for Chinese NE Training Corpus and Supporting Resources", |
| "sec_num": null |
| }, |
| { |
| "text": "2. Word in Chinese is a vague concept and there is no clear de nition for it. There are boundary ambiguities between words in texts for even human being understanding, and inevitably machine processing. Tokenization, or word segmentation is still a problem in Chinese NLP. W ord boundary ambiguities exist not only between commonly used words which are not in entity names, but also between commonly used words and entity names. 3. Besides the uniformity appearance of characters, proper names in Chinese can consist of commonly used words. As a matter of fact, almost all Chinese characters can be a commonly used words themselves, including those in entity names such as people's names, location names, etc. Therefore, unlike English, the problem of Chinese entity recognition should not be isolated from the problem of tokenization, or word segmentation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for Chinese NE Training Corpus and Supporting Resources", |
| "sec_num": null |
| }, |
| { |
| "text": "One level of back-o features, which are also called word classes, are obtained by the following way:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Building Language Models", |
| "sec_num": null |
| }, |
| { |
| "text": "We extend the idea in the new word detection engine of the integrated model of Chinese word segmentor and part of speech tagger 1 . The idea is to extend the scope of an interested word class of new word, the proper names, into named entities by looking into broader range of constituents. Under this framework, we believe contextual statistics plays important rules in deciding word boundary and predicting the categories of named entities, while local statistics, or information resides within words or entities, can provide evidence for suggesting the appearance of named entity and deciding the validity of these entities. We need to make full use of both contextual and local statistics to recognize these named entities, thus contextual language model and entity models are created. The basic process to build the model is like this:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Building Language Models", |
| "sec_num": null |
| }, |
| { |
| "text": "1. Change the tag set of the part-of-speech tagger by splitting the tag NOUN into more detailed tags related to the particular task, which include the symbolic notions of person, location, organization, date, time, money and percentage. 2. Replace the tag NOUN in the training corpus with the above extended new tags.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Building Language Models", |
| "sec_num": null |
| }, |
| { |
| "text": "Only ambiguous words are manually checked. 3. Build contextual language model with the training corpus with the new tag set. 4. Build entity models from the entity name lists. Each e n tity has its own model.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Building Language Models", |
| "sec_num": null |
| }, |
| { |
| "text": "Training Corpus and Supporting Resources SGML marked up for NE task only Brown corpus and corpus from Wall Street Journal. In total the size of words is 7.2MB, words with SGML-markup is 9.5MB. Supporting resources include the location list, country list, corporation reference list and the people's surname list provided by MUC. Only the single-word entries in these lists are in actual use.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "Observation: Problems and Solutions Case information, or more generally, orthographic information, gives good evidence of names, as was observed in 2 . Although things get muddled up when one really gets deep into it: e.g. rst words of sentences, words which do not have all normal lower case form e.g. I\", or words whose cases are changed due to other reasons such as formatting e.g. titles, being artifacts, etc. Nevertheless, this is an very important information for identifying entity names. Prepositions are also helpful, so are common su xes and pre xes of the entities, such as Corp., Mr., and so on. In general, all such useful information should be somehow sorted out. Word classes tailored for this particular purpose will be ideal. Building Language Models There are two levels of back-o features represented by w ord classes. For the following words, the two back-o features are the same:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "Hand-crafted special words for NE task. Each possesses a di erent word class represented by word itself. These special words include I\", the\", past\", pound\", following\", of\", in\", May\", etc. In total there are about 100 such words; Words from the supporting resources as stated in the beginning of this section. Words from a same list possess a same word class. From the above statistics, it's interesting to notice that non-rst common words which are initial capitalized have a far more chance to be organization than person frequencies 7525 vs 195 and location frequencies 7525 vs 896. This agrees with general observations. Also interesting is that such w ords have a higher chance not to be any of the seven entities. This comes as a bit surprise. For NLP researchers, though, it may not be a surprise at all. This example also gives a sense how general observations are represented in a precise way.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "Further research is to becarried out to justify quantitively the merits of this learning process. Its full potential has yet to be exploited. So far, our experimentation has proved that:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "1. Various kinds of text analysis syntactic, semantic, orthographic, etc can be incorporated into the same framework in a precise way, which will be used in the information extraction tagging stage in the same way; 2. It provides an easy way to absorb human knowledge as well as domain knowledge, and thus customization can be done easily;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "3. It gives great exibility a s h o w to optimize the system. 1 and 2 are somehow clear from the above discussion. Details on the disambiguation module will reveal 3. English: for each word basically look for all the possibilities from the database rst. If the word is not found, look for the possibilities of its back-o features. 4. Disambiguation module. Recall that information extraction from word sequence w becomes nding the corresponding tag sequence t. In the paradigm of maximum likelihood estimation, the best set of tags t is the one such that probtjw = max t 0 probt 0 jw. This is equivalently to nd t such that probtw = max t 0 probt`w because probt 0 jw = probtw 0 =probw and probw is a constant for any given w.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Learning Process for English NE", |
| "sec_num": null |
| }, |
| { |
| "text": "The following equality i s w ell-known: probtw = probt 1 probw 1 jt 1 probt 2 jt 1 w 1 probw 2 jt 1 w 1 t 2 probt n jt 1 w 1 : : : t n , 1 w n , 1 probw n jt 1 w 1 : : : t n , 1 w n , 1 t n : 1 Computationally, it is only feasible when some actually most dependencies are dropped, for example, probt k jt 1 w 1 : : : t k , 1 w k , 1 probt k jt k,1 t k,2 ; 2 probw k jt 1 w 1 : : : t k , 1 w k , 1 t k probw k jt k t k,1 : 3 2 and 3 can bejusti ed by Hidden Markov Modeling for the generation of word sequences. As always, Viterbi algorithm is employed to compute the probability 1, given any approximations like 2 and 3. When sparse data problem is encountered, back-o and smoothing strategy can be adopted, e.g.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "probw k jt k t k,1 backof f to ! probw k jt k ;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "or for unknown words, substitute word in 4 with its back-o features, e.g.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "probw k jt k t k,1 backof f to ! probbof 1 k jt k t k,1 backof f to ! probbof 2 k jt k t k,1 : : :", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "backof f to ! probbof N k jt k t k,1", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "backof f to ! probbof 1 k jt k : : : backof f to ! probbof N k jt k ; where N is the total number of back-o features for the word. Note that no smoothing is employed in the above s c heme. From this scheme one can see that there exist various ways of back-o and smoothing. This characteristics, as well as the free choices of back-o features, is where the exibility of the system lies. Remark. In the actual system, back-o and smoothing schemes are di erent from the above. The actual schemes are not included because they are more complicated, and yet no systematic experimentation has been done to show that they are better than other options.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "DETAILS OF THE SYSTEM MODULES", |
| "sec_num": null |
| }, |
| { |
| "text": "The system currently processes one sentence at a time, and no memory is kept once the sentence is done. Furthermore, due to limitation of time, the guidelines for both Chinese and English NE are not entirely followed, as we didn't have time to read the guidelines carefully!", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "PERFORMANCE ANALYSIS", |
| "sec_num": null |
| }, |
| { |
| "text": "The F-measures of formal run for Chinese and English are 86.38 and 77.74, respectively. Given the limited time less than six months and resources three persons, all half time, we are satisfactory with the performance. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "PERFORMANCE ANALYSIS", |
| "sec_num": null |
| }, |
| { |
| "text": "Our brief experimentation in Chinese and English Named Entity recognition shows that the system has great potential that deserves further investigation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "FUTURE RESEARCH DIRECTION", |
| "sec_num": null |
| }, |
| { |
| "text": "1. Modeling of the problem: currently information and knowledge is represented in the form of word tag. This may pose too much restriction. A better way of representing information and knowledge, in other words, a better modeling of the problem, should be studied.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "FUTURE RESEARCH DIRECTION", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "2. Quantitive justi cation of the learning process knowledge distillation should also be studied. The system should be able to compare di erent set of back-o features and thus the best one can be chosen. 3. The system provides great exibility a s h o w to optimize it. The optimization should be done systematicly, rather than trial by trial as is the case for the time being.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "acknowledgement", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "An Integrated M o del of Chinese Word S e gmentation and Part of Speech Tagging", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Bai", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Bai, An Integrated M o del of Chinese Word S e gmentation and Part of Speech Tagging, Advances and Applications on Computational Linguistics 1995, Tsinghua University Press.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Nymble: a High-Performance Learning Name-nder", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [ |
| "M" |
| ], |
| "last": "Bikel", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Miller", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Schwartz", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Weischedel", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D.M. Bikel, S. Miller, R. Schwartz and R. Weischedel, Nymble: a High-Performance Learning Name-nder.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "uris": null, |
| "type_str": "figure", |
| "num": null, |
| "text": "Information Distillation of Training Corpus pos1 -pos2 -... posM-1 PERSON -posM PERSON ... posN-3 ORG posN-2 ORG posN-1 ORG posn-1 -The generation of back-o corpora is described byFigure 1. The total numb e r o f b a c k-o corpora therein is a controllable parameter." |
| }, |
| "TABREF1": { |
| "content": "<table><tr><td colspan=\"2\">Level 1: the -COUN ADJ -WordClass1 -, -WordClass2 -the -WordClass3 -WordClass4 -WordClass5 -, -WordClass6 -to -WordClass7 -WordClass8 -by -WordClass9 PERSON WordClass10 PERSON , -WordClass11 -of -WordClass12 ORG Loc ORG WordClass13 slash ORG ; -</td></tr><tr><td colspan=\"2\">Level 2: the -COUN ADJ -LowerCaseWord -, -LowerCaseWord -the -CommonWor-dInitCap -CommonWordInitCap -CommonWordInitCap -, -LowerCaseWord -to -LowerCaseWord -LowerCaseWord -by -initCapNotCommonWord PERSON initCapNotCommonWord PERSON , -LowerCaseWord -of -CommonWordInit-Cap ORG Loc ORG CommonWordInitCap ORG ; -</td></tr><tr><td colspan=\"2\">Statistics such as the possibilities of CommonWordInitCap which are NOT rst words of sentences and the corresponding frequencies can be obtained from the second back-o corpus. From our corpus, these are:</td></tr><tr><td colspan=\"2\">Organization None of the named entities 8493 7525 Location 896 Person 195 Date 8 Money 2</td></tr><tr><td>word class</td><td>example</td></tr><tr><td colspan=\"2\">oneDigitNum containsDigitAndColon containsAlphaDigit allCaps capPeriod rstCommonWordInitCap rstNonCommonWordIC CommonWordInitCap initCapNotCommonWord David 1 2:34 A4 KRDL M. Department mixedCasesWord ValueJet charApos O'clock allLowerCase can compoundWord ad-hoc</td></tr></table>", |
| "text": "Hand-crafted lists of words, which include week words Monday, Tuesday, ..., month words January, February, ..., cardinal numbers one, two, 1 31, ..., ordinal numbers 1st, rst, 2nd, second, ..., etc. For the rest of words, the rst level features are word classes provided by a machine auto classi cation of words, while the second level of features include:In total, the number of orthographic features is about 30. To give a sense what information is extracted from the original training corpus, for example, the two back-o sentences for Example 1 are:", |
| "type_str": "table", |
| "html": null, |
| "num": null |
| } |
| } |
| } |
| } |
