DOCSLIB.ORG

MWU-Aware Part-Of-Speech Tagging with a CRF Model and Lexical Resources Mathieu Constant, Anthony Sigogne

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
MWU-Aware Part-Of-Speech Tagging with a CRF Model and Lexical Resources Mathieu Constant, Anthony Sigogne MWU-aware Part-of-Speech Tagging with a CRF model and lexical resources Mathieu Constant, Anthony Sigogne To cite this version: Mathieu Constant, Anthony Sigogne. MWU-aware Part-of-Speech Tagging with a CRF model and lexical resources. ACL Workshop on Multiword Expressions: from Parsing and Generation to the Real World (MWE’11), 2011, Portland, Oregon, United States. pp.49-56. hal-00621585 HAL Id: hal-00621585 https://hal-upec-upem.archives-ouvertes.fr/hal-00621585 Submitted on 11 Sep 2013 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. MWU-aware Part-of-Speech Tagging with a CRF model and lexical resources Matthieu Constant Anthony Sigogne Universit´eParis-Est, LIGM Universit´eParis-Est, LIGM 5, bd Descartes - Champs/Marne 5, bd Descartes - Champs/Marne 77454 Marne-la-Vall´ee cedex 2, France 77454 Marne-la-Vall´ee cedex 2, France [email protected] [email protected] Abstract ing semantic processing. Indeed, taggers are gen- erally evaluated on perfectly tokenized texts where This paper describes a new part-of-speech tag- multiword units (MWU) have already been identi- ger including multiword unit (MWU) identifi- fied. cation. It is based on a Conditional Random Our paper presents a MWU-aware POS tagger Field model integrating language-independent (i.e. a POS tagger including MWU recognition1). features, as well as features computed from external lexical resources. It was imple- It is based on a Conditional Random Field (CRF) mented in a finite-state framework composed model that integrates features computed from large- of a preliminary finite-state lexical analysis coverage morphosyntactic lexicons and fine-grained and a CRF decoding using weighted finite- MWU resources. We implemented it in a finite-state state transducer composition. We showed that framework composed of a finite-state lexical ana- our tagger reaches state-of-the-art results for lyzer and a CRF-decoder using weighted transducer French in the standard evaluation conditions composition. (i.e. each multiword unit is already merged in a single token). The evaluation of the tagger In section 2, we will first describe statistical tag- integrating MWU recognition clearly shows ging based on CRF. Then, in section 3, we will the interest of incorporating features based on show how to adapt the tagging models in order to MWU resources. also identify multiword unit. Next, section 4 will present the finite-state framework used to implement the tagger. Section 5 will focus on the description of 1 Introduction our working corpus and the set of lexical resources Part-of-speech (POS) tagging reaches excellent used. In section 6, we then evaluate our tagger on results thanks to powerful discriminative multi- French. feature models such as Conditional Random Fields 2 Statistical POS tagging with Linear (Lafferty et al., 2001), Support Vector Machine Chain Conditional Random Fields (Gim´enez and M´arquez, 2004), Maximum Entropy (Ratnaparkhi, 1996). Some studies like (Denis and Linear chain Conditional Ramdom Fields (CRF) are Sagot, 2009) have shown that featuring these models discriminative probabilistic models introduced by by means of external morphosyntactic resources still (Lafferty et al., 2001) for sequential labelling. Given improves accuracy. Nevertheless, current taggers an input sequence x = (x1, x2, ..., xN ) and an out- rarely take multiword units such as compound words 1 into account, whereas they form very frequent lexi- This strategy somewhat resembles the popular approach of cal units with strong syntactic and semantic particu- joint word segmentation and part-of-speech tagging for Chi- nese, e.g. (Zhang and Clark, 2008). Moreover, other similar larities (Sag et al., 2001; Copestake et al., 2002) and experiments on the same task for French are reported in (Con- their identification is crucial for applications requir- stant et al., 2011). put sequence of labels y = (y1, y2, ..., yN ), the In practice, we can divide features fk in two model is defined as follows: families: while unigram features (uk) do not de- N K pend on the preceding tag, i.e. fk(t, yt, yt−1, x) = 1 uk(t, yt, x), bigram features (bk) depend on both P (y|x) = exp λ f (t, x, yt, yt ) Z(x) Y X k k −1 t=1 k=1 current and preceding tags, i.e. fk(t, yt, yt−1, x) = bk(t, yt, yt−1, x). In our practical case, bigrams where Z(x) is a normalization factor depending exlusively depends on the two tags, i.e. they are in- on x. It is based on K features each of them be- dependent from the input sequence and the current ing defined by a binary function fk depending on position like in the Hidden Markov Model (HMM)3. the current position t in x, the current label yt, Unigram features can be sub-divided into internal the preceding one yt−1 and the whole input se- and contextual ones. Internal features provide solely quence x. The feature is activated if a given con- characteristics of the current token w0: lexical form figuration between t, yt, yt−1 and x is satisfied (i.e. (i.e. its character sequence), lowercase form, suf- fk(t, yt, yt−1, x) = 1). Each feature fk is associated fice, prefix, ambiguity classes in the external lexi- with a weight λk. The weights are the parameters cons, whether it contains a hyphen, a digit, whether of the model. They are estimated during the train- it is capitalized, all capitalized, multiword. Contex- ing process by maximizing the conditional loglikeli- tual features indicate characteristics of the surround- hood on a set of examples already labeled (training ings of the current token: token unigrams at relative data). The decoding procedure consists in labelling positions -2,-1,+1 and +2 (w−2, w−1, w+1,w+2); to- a new input sequence with respect to the model, by ken bigrams w−1w0, w0w+1 and w−1w+1; ambi- maximizing P (y|x) (or minimizing −logP (y|x)). guity classes at relative positions -2,-1,+1 and +2 There exist dynamic programming procedures such (AC−2, AC−1, AC+1,AC+2). The different feature as Viterbi algorithm in order to efficiently explore all templates used in our tagger are given in table 2. labelling possibilities. Features are defined by combining different prop- Internal unigram features w0 = X &t0 = T erties of the tokens in the input sequence and the la- Lowercase form of w0 = L &t0 = T bels at the current position and the preceding one. Prefix of w0 = P with |P | < 5 &t0 = T Properties of tokens can be either binary or tex- Suffix of w0 = S with |S| < 5 &t0 = T w0 contains a hyphen &t0 = T tual: e.g. token contains a digit, token is capital- w0 contains a digit &t0 = T ized (binary property), form of the token, suffix of w0 is capitalized &t0 = T w0 is all capital &t0 = T 4 size 2 of the token (textual property). Most tag- w0 is capitalized and BOS &t0 = T gers exclusively use language-independent proper- w0 is multiword &t0 = T Lexicon tags AC0 of w0 = A & w0 is multiword &t0 = T ties – e.g. (Ratnaparkhi, 1996; Toutanova et al., Contextual unigram features 2003; Gim´enez and M´arquez, 2004; Tsuruoka et wi = X, i ∈ {−2, −1, 1, 2} &t0 = T al., 2009). It is also possible to integrate language- wiwj = XY , (j, k) ∈ {(−1, 0), (0, 1), (−1, 1)} &t0 = T ACi = A & wi is multiword, i ∈ {−2, −1, 1, 2} &t0 = T dependant properties computed from an external Bigram features ′ broad-coverage morphosyntactic lexicon, that are t−1 = T &t0 = T POS tags found in the lexicon for the given token Table 1: Feature templates (e.g. (Denis and Sagot, 2009)). It is of great interest to deal with unknown words2 as most of them are covered by the lexicon, and to somewhat filter the list of candidate tags for each token. We therefore 3 MWU-aware POS tagging added to our system a language-dependent property: MWU-aware POS tagging consists in identifying a token is associated with the concatenation of its and labelling lexical units including multiword ones. possible tags in an external lexicon, i.e. the am- bibuity class of the token (AC). 3Hidden Markov Models of order n use strong indepen- dance assumptions: a word only depends on its corresponding 2Unknown words are words that did not occur in the training tag, and a tag only depends on its n previous tags. In our case, data. n=1. It is somewhat similar to segmentation tasks like unit: the part-of-speech of the lexical multiword unit chunking or Named Entity Recognition, that iden- (POS), its internal structure (STRUCT), its semantic tify the limits of chunk or Named Entity segments feature (SEM) and its relative position in the IOB and classify these segments. By using an IOB5 scheme (POSITION). Table 2 shows the encoding scheme (Ramshaw and Marcus, 1995), this task is of these properties in an example. The property ex- then equivalent to labelling simple tokens. Each to- traction is performed by a longest-match context- ken is labeled by a tag in the form X+B or X+I, free lookup in the resources.
Load more

Recommended publications
  • A Way with Words: Recent Advances in Lexical Theory and Analysis: a Festschrift for Patrick Hanks
    A Way with Words: Recent Advances in Lexical Theory and Analysis: A Festschrift for Patrick Hanks Gilles-Maurice de Schryver (editor) (Ghent University and University of the Western Cape) Kampala: Menha Publishers, 2010, vii+375 pp; ISBN 978-9970-10-101-6, e59.95 Reviewed by Paul Cook University of Toronto In his introduction to this collection of articles dedicated to Patrick Hanks, de Schryver presents a quote from Atkins referring to Hanks as “the ideal lexicographer’s lexicogra- pher.” Indeed, Hanks has had a formidable career in lexicography, including playing an editorial role in the production of four major English dictionaries. But Hanks’s achievements reach far beyond lexicography; in particular, Hanks has made numerous contributions to computational linguistics. Hanks is co-author of the tremendously influential paper “Word association norms, mutual information, and lexicography” (Church and Hanks 1989) and maintains close ties to our field. Furthermore, Hanks has advanced the understanding of the relationship between words and their meanings in text through his theory of norms and exploitations (Hanks forthcoming). The range of Hanks’s interests is reflected in the authors and topics of the articles in this Festschrift; this review concentrates more on those articles that are likely to be of most interest to computational linguists, and does not discuss some articles that appeal primarily to a lexicographical audience. Following the introduction to Hanks and his career by de Schryver, the collection is split into three parts: Theoretical Aspects and Background, Computing Lexical Relations, and Lexical Analysis and Dictionary Writing. 1. Part I: Theoretical Aspects and Background Part I begins with an unfinished article by the late John Sinclair, in which he begins to put forward an argument that multi-word units of meaning should be given the same status in dictionaries as ordinary headwords.
    [Show full text]
  • Finetuning Pre-Trained Language Models for Sentiment Classification of COVID19 Tweets
    Technological University Dublin ARROW@TU Dublin Dissertations School of Computer Sciences 2020 Finetuning Pre-Trained Language Models for Sentiment Classification of COVID19 Tweets Arjun Dussa Technological University Dublin Follow this and additional works at: https://arrow.tudublin.ie/scschcomdis Part of the Computer Engineering Commons Recommended Citation Dussa, A. (2020) Finetuning Pre-trained language models for sentiment classification of COVID19 tweets,Dissertation, Technological University Dublin. doi:10.21427/fhx8-vk25 This Dissertation is brought to you for free and open access by the School of Computer Sciences at ARROW@TU Dublin. It has been accepted for inclusion in Dissertations by an authorized administrator of ARROW@TU Dublin. For more information, please contact [email protected], [email protected]. This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License Finetuning Pre-trained language models for sentiment classification of COVID19 tweets Arjun Dussa A dissertation submitted in partial fulfilment of the requirements of Technological University Dublin for the degree of M.Sc. in Computer Science (Data Analytics) September 2020 Declaration I certify that this dissertation which I now submit for examination for the award of MSc in Computing (Data Analytics), is entirely my own work and has not been taken from the work of others save and to the extent that such work has been cited and acknowledged within the test of my work. This dissertation was prepared according to the regulations for postgraduate study of the Technological University Dublin and has not been submitted in whole or part for an award in any other Institute or University.
    [Show full text]
  • Lexical Analysis
    From words to numbers Parsing, tokenization, extract information Natural Language Processing Piotr Fulmański Lecture goals • Tokenizing your text into words and n-grams (tokens) • Compressing your token vocabulary with stemming and lemmatization • Building a vector representation of a statement Natural Language Processing in Action by Hobson Lane Cole Howard Hannes Max Hapke Manning Publications, 2019 Terminology Terminology • A phoneme is a unit of sound that distinguishes one word from another in a particular language. • In linguistics, a word of a spoken language can be defined as the smallest sequence of phonemes that can be uttered in isolation with objective or practical meaning. • The concept of "word" is usually distinguished from that of a morpheme. • Every word is composed of one or more morphemes. • A morphem is the smallest meaningful unit in a language even if it will not stand on its own. A morpheme is not necessarily the same as a word. The main difference between a morpheme and a word is that a morpheme sometimes does not stand alone, but a word, by definition, always stands alone. Terminology SEGMENTATION • Text segmentation is the process of dividing written text into meaningful units, such as words, sentences, or topics. • Word segmentation is the problem of dividing a string of written language into its component words. Text segmentation, retrieved 2020-10-20, https://en.wikipedia.org/wiki/Text_segmentation Terminology SEGMENTATION IS NOT SO EASY • Trying to resolve the question of what a word is and how to divide up text into words we can face many "exceptions": • Is “ice cream” one word or two to you? Don’t both words have entries in your mental dictionary that are separate from the compound word “ice cream”? • What about the contraction “don’t”? Should that string of characters be split into one or two “packets of meaning?” • The single statement “Don’t!” means “Don’t you do that!” or “You, do not do that!” That’s three hidden packets of meaning for a total of five tokens you’d like your machine to know about.
    [Show full text]
  • Lexical Sense Labeling and Sentiment Potential Analysis Using Corpus-Based Dependency Graph
    mathematics Article Lexical Sense Labeling and Sentiment Potential Analysis Using Corpus-Based Dependency Graph Tajana Ban Kirigin 1,* , Sanda Bujaˇci´cBabi´c 1 and Benedikt Perak 2 1 Department of Mathematics, University of Rijeka, R. Matejˇci´c2, 51000 Rijeka, Croatia; [email protected] 2 Faculty of Humanities and Social Sciences, University of Rijeka, SveuˇcilišnaAvenija 4, 51000 Rijeka, Croatia; [email protected] * Correspondence: [email protected] Abstract: This paper describes a graph method for labeling word senses and identifying lexical sentiment potential by integrating the corpus-based syntactic-semantic dependency graph layer, lexical semantic and sentiment dictionaries. The method, implemented as ConGraCNet application on different languages and corpora, projects a semantic function onto a particular syntactical de- pendency layer and constructs a seed lexeme graph with collocates of high conceptual similarity. The seed lexeme graph is clustered into subgraphs that reveal the polysemous semantic nature of a lexeme in a corpus. The construction of the WordNet hypernym graph provides a set of synset labels that generalize the senses for each lexical cluster. By integrating sentiment dictionaries, we introduce graph propagation methods for sentiment analysis. Original dictionary sentiment values are integrated into ConGraCNet lexical graph to compute sentiment values of node lexemes and lexical clusters, and identify the sentiment potential of lexemes with respect to a corpus. The method can be used to resolve sparseness of sentiment dictionaries and enrich the sentiment evaluation of Citation: Ban Kirigin, T.; lexical structures in sentiment dictionaries by revealing the relative sentiment potential of polysemous Bujaˇci´cBabi´c,S.; Perak, B. Lexical Sense Labeling and Sentiment lexemes with respect to a specific corpus.
    [Show full text]
  • Automated Citation Sentiment Analysis Using High Order N-Grams: a Preliminary Investigation
    Turkish Journal of Electrical Engineering & Computer Sciences Turk J Elec Eng & Comp Sci (2018) 26: 1922 – 1932 http://journals.tubitak.gov.tr/elektrik/ © TÜBİTAK Research Article doi:10.3906/elk-1712-24 Automated citation sentiment analysis using high order n-grams: a preliminary investigation Muhammad Touseef IKRAM1;∗,, Muhammad Tanvir AFZAL1, Naveed Anwer BUTT2, 1Department of Computer Science, Capital University of Science and Technology, Islamabad, Pakistan 2Department of Computer Science, University of Gujrat, Gujrat, Pakistan Received: 03.12.2017 • Accepted/Published Online: 02.02.2018 • Final Version: 27.07.2018 Abstract: Scientific papers hold an association with previous research contributions (i.e. books, journals or conference papers, and web resources) in the form of citations. Citations are deemed as a link or relatedness of the previous work to the cited work. The nature of the cited material could be supportive (positive), contrastive (negative), or objective (neutral). Extraction of the author’s sentiment towards the cited scientific articles is an emerging research discipline due to various linguistic differences between the citation sentences and other domains of sentiment analysis. In this paper, we propose a technique for the identification of the sentiment of the citing author towards the cited paper by extracting unigram, bigram, trigram, and pentagram adjective and adverb patterns from the citation text. After POS tagging of the citation text, we use the sentence parser for the extraction of linguistic features comprising adjectives, adverbs, and n-grams from the citation text. A sentiment score is then assigned to distinguish them as positive, negative, and neutral. In addition, the proposed technique is compared with manually classified citation text and 2 commercial tools, namely SEMANTRIA and THEYSAY, to determine their applicability to the citation corpus.
    [Show full text]
  • A Framework for Representing Lexical Resources
    A framework for representing lexical resources Fabrice Issac LDI Universite´ Paris 13 Abstract demonstration of the use of lexical resources in different languages. Our goal is to propose a description model for the lexicon. We describe a software 2 Context framework for representing the lexicon and its variations called Proteus. Various Whatever the writing system of a language (logo- examples show the different possibilities graphic, syllabic or alphabetic), it seems that the offered by this tool. We conclude with word is a central concept. Nonetheless, the very a demonstration of the use of lexical re- definition of a word is subject to variation depend- sources in complex, real examples. ing on the language studied. For some Asian lan- guages such as Mandarin or Vietnamese, the no- 1 Introduction tion of word delimiter does not exist ; for others, such as French or English, the space is a good in- Natural language processing relies as well on dicator. Likewise, for some languages, preposi- methods, algorithms, or formal models as on lin- tions are included in words, while for others they guistic resources. Processing textual data involves form a separate unit. a classic sequence of steps : it begins with the nor- malisation of data and ends with their lexical, syn- Languages can be classified according to their tactic and semantic analysis. Lexical resources are morphological mechanisms and their complexity ; the first to be used and must be of excellent qual- for instance, the morphological systems of French ity. or English are relatively simple compared to that Traditionally, a lexical resource is represented of Arabic or Greek.
    [Show full text]
  • Content Extraction and Lexical Analysis from Customer-Agent Interactions
    Lexical Analysis and Content Extraction from Customer-Agent Interactions Sergiu Nisioi Anca Bucur? Liviu P. Dinu Human Language Technologies Research Center, ?Center of Excellence in Image Studies, University of Bucharest fsergiu.nisioi,[email protected], [email protected] Abstract but also to forward a solution for cleaning and ex- tracting useful content from raw text. Given the In this paper, we provide a lexical compara- large amounts of unstructured data that is being tive analysis of the vocabulary used by cus- collected as email exchanges, we believe that our tomers and agents in an Enterprise Resource proposed method can be a viable solution for con- Planning (ERP) environment and a potential solution to clean the data and extract relevant tent extraction and cleanup as a preprocessing step content for NLP. As a result, we demonstrate for indexing and search, near-duplicate detection, that the actual vocabulary for the language that accurate classification by categories, user intent prevails in the ERP conversations is highly di- extraction or automatic reply generation. vergent from the standardized dictionary and We carry our analysis for Romanian (ISO 639- further different from general language usage 1 ro) - a Romance language spoken by almost as extracted from the Common Crawl cor- 24 million people, but with a relatively limited pus. Moreover, in specific business commu- nication circumstances, where it is expected number of NLP resources. The purpose of our to observe a high usage of standardized lan- approach is twofold - to provide a comparative guage, code switching and non-standard ex- analysis between how words are used in question- pression are predominant, emphasizing once answer interactions between customers and call more the discrepancy between the day-to-day center agents (at the corpus level) and language as use of language and the standardized one.
    [Show full text]
  • A Corpus-Based Investigation of Lexical Cohesion in En & It
    A CORPUS-BASED INVESTIGATION OF LEXICAL COHESION IN EN & IT NON-TRANSLATED TEXTS AND IN IT TRANSLATED TEXTS A thesis submitted To Kent State University in partial Fufillment of the requirements for the Degree of Doctor of Philosophy by Leonardo Giannossa August, 2012 © Copyright by Leonardo Giannossa 2012 All Rights Reserved ii Dissertation written by Leonardo Giannossa M.A., University of Bari – Italy, 2007 B.A., University of Bari, Italy, 2005 Approved by ______________________________, Chair, Doctoral Dissertation Committee Brian Baer ______________________________, Members, Doctoral Dissertation Committee Richard K. Washbourne ______________________________, Erik Angelone ______________________________, Patricia Dunmire ______________________________, Sarah Rilling Accepted by ______________________________, Interim Chair, Modern and Classical Language Studies Keiran J. Dunne ______________________________, Dean, College of Arts and Sciences Timothy Moerland iii Table of Contents LIST OF FIGURES .......................................................................................................... vii LIST OF TABLES ........................................................................................................... viii DEDICATION ................................................................................................................... xi ACKNOWLEDGEMENTS .............................................................................................. xii ABSTRACT ....................................................................................................................
    [Show full text]
  • Lexical Analysis
    Lexical Analysis Lecture 3 Instructor: Fredrik Kjolstad Slide design by Prof. Alex Aiken, with modifications 1 Outline • Informal sketch of lexical analysis – Identifies tokens in input string • Issues in lexical analysis – Lookahead – Ambiguities • Specifying lexers (aka. scanners) – By regular expressions (aka. regex) – Examples of regular expressions 2 Lexical Analysis • What do we want to do? Example: if (i == j) Z = 0; else Z = 1; • The input is just a string of characters: \tif (i == j)\n\t\tz = 0;\n\telse\n\t\tz = 1; • Goal: Partition input string into substrings – Where the substrings are called tokens 3 What’s a Token? • A syntactic category – In English: noun, verb, adjective, … – In a programming language: Identifier, Integer, Keyword, Whitespace, … 4 Tokens • A token class corresponds to a set of strings Infinite set • Examples i var1 ports – Identifier: strings of letters or foo Person digits, starting with a letter … – Integer: a non-empty string of digits – Keyword: “else” or “if” or “begin” or … – Whitespace: a non-empty sequence of blanks, newlines, and tabs 5 What are Tokens For? • Classify program substrings according to role • Lexical analysis produces a stream of tokens • … which is input to the parser • Parser relies on token distinctions – An identifier is treated differently than a keyword 6 Designing a Lexical Analyzer: Step 1 • Define a finite set of tokens – Tokens describe all items of interest • Identifiers, integers, keywords – Choice of tokens depends on • language • design of parser 7 Example • Recall \tif
    [Show full text]
  • Text-To-Speech Synthesis
    Sproat, R. & Olive, J. “Text-to-Speech Synthesis” Digital Signal Processing Handbook Ed. Vijay K. Madisetti and Douglas B. Williams Boca Raton: CRC Press LLC, 1999 c 1999byCRCPressLLC 46 Text-to-Speech Synthesis Richard Sproat 46.1 Introduction 46.2 Text Analysis and Linguistic Analysis Bell Laboratories • • • Lucent Technologies Text Preprocessing Accentuation Word Pronunciation In- tonational Phrasing • Segmental Durations • Intonation Joseph Olive 46.3 Speech Synthesis Bell Laboratories 46.4 The Future of TTS Lucent Technologies References 46.1 Introduction Text-to-speech synthesis has had a long history, one that can be traced back at least to Dudley’s “Voder”, developed at Bell Laboratories and demonstrated at the 1939 World’s Fair [1]. Practical systems for automatically generating speech parameters from a linguistic representation (such as a phoneme string) were not available until the 1960s, and systems for converting from ordinary text into speech were first completed in the 1970s, with MITalk being the best-known such system [2]. Many projects in text-to-speech conversion have been initiated in the intervening years, and papers on many of these systems have been published.1 It is tempting to think of the problem of converting written text into speech as “speech recognition in reverse”: current speech recognition systems are generally deemed successful if they can convert speech input into the sequence of words that was uttered by the speaker, so one might imagine that a text-to-speech (TTS) synthesizer would start with the words in the text, convert each word one-by-one into speech (being careful to pronounce each word correctly), and concatenate the result together.
    [Show full text]
  • Robust Extended Tokenization Framework for Romanian by Semantic Parallel Texts Processing
    International Journal on Natural Language Computing (IJNLC) Vol. 2, No.6, December 2013 ROBUST EXTENDED TOKENIZATION FRAMEWORK FOR ROMANIAN BY SEMANTIC PARALLEL TEXTS PROCESSING Eng. Marius Zubac 1 and Prof. PhD Eng. Vasile D ădârlat 2 1Department of Computer Science, Technical University, Cluj-Napoca, Romania 2Department of Computer Science, Technical University, Cluj-Napoca, Romania ABSTRACT Tokenization is considered a solved problem when reduced to just word borders identification, punctuation and white spaces handling. Obtaining a high quality outcome from this process is essential for subsequent NLP piped processes (POS-tagging, WSD). In this paper we claim that to obtain this quality we need to use in the tokenization disambiguation process all linguistic, morphosyntactic, and semantic-level word-related information as necessary. We also claim that semantic disambiguation performs much better in a bilingual context than in a monolingual one. Then we prove that for the disambiguation purposes the bilingual text provided by high profile on-line machine translation services performs almost to the same level with human-originated parallel texts (Gold standard). Finally we claim that the tokenization algorithm incorporated in TORO can be used as a criterion for on-line machine translation services comparative quality assessment and we provide a setup for this purpose. KEYWORDS Semantic Tokenization, Bi-lingual Corpora, Romanian WordNet, Bing, Google Translate, Comparative MT Quality Assessment 1. INTRODUCTION When faced with the task of text processing today’s systems need to implement robust procedures for input text acceptance no matter the final processing purpose ranging from information retrieval and sentiment analysis to machine translation applications. This acceptance criteria may include various filtering pre-processing of the text, however the vast majority of researchers [1], [2], [3] agree that tokenization should be the first step of the text processing pipe as shown in the Figure 1 preceding any lexical analysis step.
    [Show full text]
  • Semantic Analysis Lexical Analysis
    9. Text and Document Analytics Jacobs University Visualization and Computer Graphics Lab Unstructured Data Data Mining / Knowledge Discovery Structured Data Multimedia Free Text Hypertext HomeLoan ( Frank Rizzo bought <a href>Frank Rizzo Loanee: Frank Rizzo his home from Lake </a> Bought Lender: MWF View Real Estate in <a hef>this home</a> Agency: Lake View 1992. from <a href>Lake Amount: $200,000 He paid $200,000 View Real Estate</a> Term: 15 years under a15-year loan In <b>1992</b>. ) Loans($200K,[map],...) from MW Financial. <p>... Jacobs University Visualization and Computer Graphics Lab Data Analytics 312 Unstructured Data • There are many examples of text-based documents (all in ‘electronic’ format…) – e-mails, corporate Web pages, customer surveys, résumés, medical records, DNA sequences, technical papers, incident reports, news stories, etc. • Not enough time or patience to read – Can we extract the most vital kernels of information? • Find a way to gain knowledge (in summarised form) from all that text, without reading or examining them fully first Jacobs University Visualization and Computer Graphics Lab Data Analytics 313 Tasks Jacobs University Visualization and Computer Graphics Lab Data Analytics 314 9.1 Text Mining Jacobs University Visualization and Computer Graphics Lab Text Mining Jacobs University Visualization and Computer Graphics Lab Data Analytics 316 Text Mining • Typically falls into one of two categories: – Analysis of text: I have a bunch of text I am interested in, tell me something about it • E.g. sentiment analysis, “buzz” searches –Retrieval: There is a large corpus of text documents, and I want the one closest to a specified query • E.g.
    [Show full text]