[Upcoming Journal Publication]: Automatic Detection of Arabic Non-Anaphoric Pronouns for Improving Anaphora Resolution

2010-08-27T13:34:00.000-07:00

The manuscript with the above title has been accepted for publication in the ACM the Transactions on Asian Language Information Processing.
Abstract:

Anaphora resolution is one of the most difficult tasks in NLP. The ability to identify non-referential pronouns before attempting an anaphora resolution task would be significant, since the system would not have to attempt resolving such pronouns and hence end up with fewer errors. In addition, the number of non-referential pronouns has been found to be non-trivial in many domains. The task of detecting non-referential pronouns could also be incorporated into a part-of-speech tagger or a parser, or treated as an initial step in semantic interpretation. In this paper, I describe a machine learning method for identifying non-referential pronouns in an annotated sub-segment of the Penn Arabic Treebank using three different feature settings. I achieve an accuracy of 97.22% with 52 different features extracted from a small window size of -5/+5 tokens surrounding each potentially non-referential pronoun.

code_swarm - Python

2010-08-01T17:45:00.000-07:00

Lost in Translation: Cross-Language Machine Translation

2009-01-20T10:27:00.000-08:00

See how machine translation (ML) as done by babel

Original English Text:
What happens when an English phrase is translated (by computer) back and forth between 5 different languages? The authors of the Systran language translation software probably never intended this application of their program. As of September 2007, translation software is almost good enough to turn grammatically correct, slang-free text from one language into grammatically incorrect, barely readable approximations in another. But the software is not equipped for 10 consecutive translations of the same piece of text. The resulting half-English, half-foreign, and totally non sequitur response bears almost no resemblance to the original. Remember the old game of "Telephone"? Something is lost, and sometimes something is gained.

Translated to French:
Que se produit quand une expression anglaise est traduite (par ordinateur) dans les deux sens entre 5 langages différents ? Les auteurs du logiciel de traduction des langues de Systran n'ont probablement jamais destiné cette application de leur programme. En date du septembre 2007, il est presque assez bon transformer logiciel de traduction texte grammaticalement correct et sans argot d'un langage en approximations grammaticalement incorrectes et à peine lisibles dans des autres. Mais le logiciel n'est pas équipé pour 10 traductions consécutives de la même partie de texte. Résulter moitié-Anglais, moitié-étranger, et totalement non réponse de sequitur ne soutient presque aucune ressemblance à l'original. Rappelez-vous le vieux jeu du " ; Telephone" ; ? Quelque chose est détruite, et parfois quelque chose est gagnée.

Translated back to English:
What occurs when an English expression is translated (by computer) in the two directions between 5 different languages? The authors of the software of translation of the languages of Systran probably never intended this application of their program. On September 2007, it is almost rather good to transform software grammatically correct text translation and without slang of a language into grammatically incorrect and hardly readable approximations in others. But the software is not equipped for 10 consecutive translations with the same part of text. To result half-English, half-foreigner, and completely non-response of sequitur does not support almost any resemblance with the original. You point out the old play of the " Telephone" ? Something is destroyed, and sometimes something is gained.

Translated to German:
Was tritt auf, wenn ein englischer Ausdruck (durch Computer) in den zwei Richtungen zwischen 5 verschiedene Sprachen übersetzt wird? Die Autoren der Software der Übersetzung von den Sprachen von Systran beabsichtigten vermutlich nie diese Anwendung ihres Programms. Am September 2007 ist es fast ziemlich gut, Textübersetzung der Software grammatisch und ohne Jargon einer Sprache in die grammatisch falschen und kaum lesbaren Näherungswerte in anderen umzuwandeln korrekte. Aber die Software wird nicht für 10 nachfolgende Übersetzungen mit dem gleichen Teil des Textes ausgerüstet. Um zu resultieren Hälfte-Englisch, unterstützt Hälfteausländer und vollständig Nichtbeantwortung von sequitur fast keine Ähnlichkeit mit der Vorlage. Sie unterstreichen das alte Spiel des " Telefone" ? Etwas wird zerstört und manchmal wird etwas gewonnen.

Translated back to English:
What occurs, if an English printout (by computers) is translated in the two directions between 5 different languages? The authors of the software of the translation of the languages of Systran probably never intended this application of its program. On September 2007 it is almost rather good to convert text translation of the software grammatically and without jargon of a language into the grammatically false and hardly readable approximate values into others correct. But the software is not equipped for 10 following translations with the same part of the text. Over to result half English, supports half foreigners and completely nonresponse of sequitur almost no similarity with the collecting main. They underline the old play " Telefone" ? Sometimes something is destroyed and won somewhat.

Translated to Italian:
Che cosa accade, se uno stampato inglese (dai computer) è tradotto in due direzioni fra 5 linguaggi differenti? Gli autori del software della traduzione dei linguaggi di Systran non hanno inteso probabilmente mai questa applicazione del suo programma. Settembre 2007 è quasi piuttosto buono da convertire la traduzione del testo del software grammaticale e senza gergo di un linguaggio in valori approssimativi grammaticale falsi ed appena leggibili in altre corrette. Ma il software non è fornito per 10 traduzioni seguenti della stessa parte del testo. Sopra per risultare stranieri mezzi sostegni e a metà inglesi e completamente mancanza di risposta di sequitur quasi nessuna similarità con il main di raccolta. Sottolineano il vecchio " del gioco; Telefone" ? A volte qualcosa si distrugge e vinto piuttosto.

Translated back to English:
What happens, if an English print (from the computers) is translate in two directions between 5 different languages? The authors of the software of the translation of the languages of Systran have not intended probably never this application of its program. September 2007 is almost rather good to convert the translation of the text of the grammaticale software and without jargon of a language in false and hardly leggibili values approximate grammaticale in other correcting. But the software is not supplied for 10 following translations of the same part of the text. Over in order completely turning out foreign means supports and to English halves and lack of answer of sequitur almost no similarity with the main of collection. They emphasize the old one " of the game; Telefone" Sometimes something is destroyed and gained rather.

Translated to Portuguese:
Que acontece, se uma cópia inglesa (dos computadores) é traduz em dois sentidos entre 5 línguas diferentes? Os autores do software da tradução das línguas de Systran não pretenderam provavelmente nunca esta aplicação de seu programa. O setembro 2007 é quase um pouco bom converter a tradução do texto do software do grammaticale e sem jargão de uma língua no grammaticale aproximado falso e mal do leggibili dos valores na outra correcção. Mas o software não é fornecido para 10 traduções de seguimento da mesma parte do texto. Sobre em ordem que despeja completamente apoios extrangeiros dos meios e às metades e à falta inglesas da resposta do sequitur quase nenhuma similaridade com o cano principal da coleção. Emfatizam o um " velho do jogo; Telefone" Às vezes algo é destruído e ganhado um pouco.

Translated back to English:
What happens, if an English copy (of the computers) is translates two-way between 5 different languages? The authors of the software of the translation of the languages of Systran had not intended probably never this application of its program. September 2007 is almost a little good badly to convert the translation of the text of the software of grammaticale and without jargon of a language into grammaticale approached false and of leggibili of the values in the other correction. But software is not supplied 10 translations of pursuing of the same part of the text. On in sequence that it completely pours extrangeiros supports of the ways and to the English halves and the lack of the response it sequitur almost no similarity with the main pipe of the collection. Emfatizam the old one " of the game; Telefone" To the times something is destroyed and gained a little.

Translated to Spanish:
¿Qué sucede, si es una copia inglesa (de los ordenadores) traduce bidireccional entre 5 diversos lenguajes? Los autores del software de la traducción de los lenguajes de SYSTRAN probablemente nunca no habían pensado esta aplicación de su programa. El septiembre de 2007 casi es un poco bueno gravemente convertir la traducción del texto del software del grammaticale y sin jerga de un lenguaje en falso acercada grammaticale y del leggibili de los valores en la otra corrección. Pero el software no se suministra 10 traducciones de prosecución de la misma parte del texto. En en orden eso vierte totalmente las ayudas de los extrangeiros de las maneras y a las mitades inglesas y a la falta de la respuesta él sequitur casi ninguna semejanza con el tubo principal de la colección. Emfatizam el viejo un " del juego; Telefone" A los tiempos algo se destruye y se gana un poco.

Translated back to English:
What happens, if it is an English copy (of the computers) translates bidirectional between 5 diverse languages? The authors of the software of the translation of the SYSTRAN languages probably never had thought this application of their program. The September of 2007 almost is a little good seriously to turn the translation of the text of the software of grammaticale and without slang of a language into false approached grammaticale and of leggibili of the values into the other correction. But software does not provide 10 translations of continuation of the same part of the text. In sequence that it spills totally the aid of the extrangeiros of the ways and to English halves and the lack of the answer he sequitur almost no similarity with the main tube of the collection. Emfatizam old " of the game; Telefone" To the times something is destroyed and a little gains.

Jurafsky & Martin's Speech and Language Processing

2009-01-20T09:37:00.000-08:00

Currently, I am taking a "Computation and Linguistic" analysis class here at IU Linguistics with Prof. Sandra Kuebler. we are using Jurafsky & Martin's (2009) Speech and Language Processing--2nd edition. The book seems really useful, and, from the very first sentence, authors motivate their readers. That sentence runs as follows: "This is an exciting time to be working in speech and language processing". In their introductory chapter--chapter 1--the authors start by spelling out the various terms used to refer to the field of "natural language processing" (NLP), mentioning terms like "speech and language processing", "human language technology", "computational linguistics", and "speech recognition and synthesis". People working within this field of study would, I believe, agree that not all these terms refer to strictly the same thing. But I think everybody will agree that there is a great deal of overlap. For me, this multidisciplinary nature of the field is among the first reasons that make NLP exciting.

Authors then move on to give examples of the subfileds/tasks that make use of NLP like "conversational agent", "dialogue systems", "machine translation", and "question answering". Then the authors briefly mention levels of linguistic knowledge (e.g., phonetics, phonology, morphology, semantics,pragmatics, and discourse) needed for speech and language processing, stating that it is this sort of linguistic knowledge that distinguishes language processing applications from other data processing.

Authors then use the concept of ambiguity to package the mission statement of NLP. They mention that most NLP tasks can be seen as attempts of resolving ambiguity at one or more of the linguistic levels. They also provide example of what ambiguities need to be solved. These incorporate talking about "part of speech tagging", "word sense disambiguation", "lexical and syntactic disambiguation", "probabilistic parsing", and "speech act interpretation".

Models and algorisms are also presented as ways of resolving ambiguities. The main models and algorisms within the field are mentioned, with a promise of more on them later. Models and algorisms are classified into three types: State machines, formal rule systems, and models based on logic. Examples of each model are provided.

The authors then by briefly discussing the early hope, or perhaps for some people promise, of compuational linguistics: Computers will be able to process natural language as intelligently as we humans do. They talk about Alan Turing's (1950) paper and the Turing test, and then about ELIZA, the early NLP system (see Weizenbaum, 1966).

The chapter ends with a "state of the art" interesting section, where current trends in computational linguistics are situated within a chronologically-organized, but also thematized, historical context.

Muhammad abdul-Mageed's Computational Linguistics

[Upcoming Journal Publication]: Automatic Detection of Arabic Non-Anaphoric Pronouns for Improving Anaphora Resolution

code_swarm - Python

Lost in Translation: Cross-Language Machine Translation

Jurafsky & Martin's Speech and Language Processing