Agglutinative languages

In a strict sense, verb movement, or verb raising, is the phenomenon by which the verb leaves the verb phrase and raises to functional projections above it to value the features of those projections. While this phenomenon is productive in Romance, it is not productive for instance in languages like English, in which either there is no verb movement at all or only a very restricted raising of the main verb. Under this general umbrella “verb movement” or “verb raising,” however, the literature includes not only the raising of the main verb in its distinct forms—finite, infinitive, gerund, participle (active past participle, passive past participle, etc.)—a process here referred to as verb raising in a strict sense, but also the raising of other V-related categories—for example, auxiliaries and functional verbs from distinct semantic classes (modal, aspectual, etc.)—a process here called verb raising in a broad sense. Since Pollock’s (1989) seminal work, generative grammar has adopted specific diagnostic tools to detect whether distinct verb forms undergo raising. In this scenario, adverbs play an important role. Occupying fixed positions within the clause, adverbs can serve as diagnostics for verb movement, both in its broad and strict senses. Thus, when one considers fine-grained representations of clausal structure, the different classes of adverbs—each occupying distinct positions along the functional spine—can help diagnose the varying heights a given verb form may obligatorily (or optionally) reach or be unable to reach. Fundamental to this type of analysis is the assumption of an adverb hierarchy, such as Cinque’s (1999) clausal hierarchy, where distinct classes of adverbs are systematically ordered.

Not only can these different classes of adverbs detect the raising or not of a given verb form, but they can also be of help in the study of intra- and inter-linguistic variation. When it comes to intra-linguistic variation, one can examine whether different verb forms reach varying heights within the functional sequence. The results of this intra-linguistic endeavor can be used to feed an inter-linguistic variation study by comparing the different heights a given verb form can reach in distinct languages. Caution is nonetheless necessary—when taking adverbs as diagnostic tools—insofar as independent reasons (precisely having nothing to do with verb raising) may well be responsible for distinct judgments.

Stop words, though often considered insignificant in text processing, play a crucial role in Natural Language Processing (NLP) operations resembling text classification, sentiment analysis, and information retrieval.With respect to the Malayalam language, a Dravidian language spoken predominantly in the Indian state of Kerala, the detection and categorization of stop words present unique challenges due to its rich morphology and syntactic structure. This study suggests a newly created hybrid deep learning method for identifying and classifying stop words in Malayalam text. This study offers a hybrid approach to stopwords recognition that combines Sandhi rule-based algorithms with manually selected methods.

Purpose: Frequently utilized terms with little semantic value are filtered out of text processing tasks as stop words. Making an inventory of stop phrases specifically designed for the Malayalam language is an essential first step in enhancing the efficacy and accuracy of numerous uses to feed natural language processing (NLP), such as text summarization, sentiment analysis, and information retrieval. This article proposes the systematic procedure of developing a comprehensive and trustworthy a compilation of stop phrases in Malayalam, a Dravidian language especially spoken in the Indian state of Kerala. A dual mechanism is employed by the methodology to compile a collection of Malayalam words that stop. Initially, the list of well-known English stopwords is compiled and converted into meaningful Malayalam terms (group 1). Secondly, distinct Malayalam stopwords are drawn from a variety of sources (group 2). Ultimately, a significantly large set of Malayalam stopwords is produced by combining the corresponding Malayalam stopwords from groups 1 and 2. Objectives: Developing a comprehensive and trustworthy a compilation of stop phrases in Malayalam, a Dravidian language especially spoken in the Indian state of Kerala. Methodology/Design/Approach: A dual mechanism is employed by the methodology to compile a collection of Malayalam words that stop. Initially, the list of well-known English stopwords is compiled and converted into meaningful Malayalam terms (group 1). Secondly, distinct Malayalam stopwords are drawn from a variety of sources (group 2).

One of the subtasks of the Natural Language Processing (NLP) application, part of
speech tagger, is essential for other NLP applications. It includes giving each word
an  appropriate  POS  tag  that  defines  how  it  is  used  in  a  sentence.  This  thesis
developed  a  part-of-speech  tagger  for  the  Kambata  language  using  a  hybrid
technique  that  combines  rule-based  and  HMM  approaches.  The  transformationbased learner, a rule-based tagger, tags the words based on transformations or rules
that  are  directly  determined  from  the  training  corpus  without  the  assistance  of
humans or experts in the field. The HMM tagger assigns a word tag based on the
word sequence's most likely path. The hybrid approach of Kambata Language part
of speech taggers developed in this thesis use Brill's tagger as a corrector based on
a  predetermined  threshold  value  and  HMM  tagger  as  initial  annotators.  The
experiment  and  implementation  were  carried  out  using  NLTK  3.0.2  and  Python
3.4.3.  We  implemented  the  bootstrapping  method  to  reduce  the  amount  of  data
needed and the expense of data preparation. 2000 sentences were gathered from the
Kambata  Tembaro  Zone  Educational  Office  to  train  and  test  the  model.  For  the
experimental analysis, 80% of the data was used for training and 20% for testing.
With the identical training and testing sets, the three taggers HMM, rule-based, and
hybrid  tagger  performed  well,  with  accuracy  rates  of  90.9%,  92.4%,  and  94.3%,
respectively.  In  conclusion,  the  hybrid  tagger's  accuracy  shows  a
clear improvement  in  performance  over  single  taggers.  Future  works  were
recommended  to  use  morphologically  segmented  words  and  training  data  with
wide coverage/domain areas to improve tagger performance.

One of the subtasks of the Natural Language Processing (NLP) application, part of
speech tagger, is essential for other NLP applications. It includes giving each word
an  appropriate  POS  tag  that  defines  how  it  is  used  in  a  sentence.  This  thesis
developed  a  part-of-speech  tagger  for  the  Kambata  language  using  a  hybrid
technique  that  combines  rule-based  and  HMM  approaches.  The  transformationbased learner, a rule-based tagger, tags the words based on transformations or rules
that  are  directly  determined  from  the  training  corpus  without  the  assistance  of
humans or experts in the field. The HMM tagger assigns a word tag based on the
word sequence's most likely path. The hybrid approach of Kambata Language part
of speech taggers developed in this thesis use Brill's tagger as a corrector based on
a  predetermined  threshold  value  and  HMM  tagger  as  initial  annotators.  The
experiment  and  implementation  were  carried  out  using  NLTK  3.0.2  and  Python
3.4.3.  We  implemented  the  bootstrapping  method  to  reduce  the  amount  of  data
needed and the expense of data preparation. 2000 sentences were gathered from the
Kambata  Tembaro  Zone  Educational  Office  to  train  and  test  the  model.  For  the
experimental analysis, 80% of the data was used for training and 20% for testing.
With the identical training and testing sets, the three taggers HMM, rule-based, and
hybrid  tagger  performed  well,  with  accuracy  rates  of  90.9%,  92.4%,  and  94.3%,
respectively.  In  conclusion,  the  hybrid  tagger's  accuracy  shows  a
clear improvement  in  performance  over  single  taggers.  Future  works  were
recommended  to  use  morphologically  segmented  words  and  training  data  with
wide coverage/domain areas to improve tagger performance.

The study of undocumented languages often reveals two important aspects of languages: that unrelated languages can and do share linguistic features, and on the other hand, they exhibit peculiarities that broaden our knowledge of how humans perceive the world and express such perceptions through language. The Bakairi language is highly agglutinative and because of this agglutinative nature, it incorporates morphemes often represented as independent words in other languages. In forming imperative and prohibitive paradigms, the Bakairi language incorporates notions of repetitiveness or iterativity, plurality, a movement toward the speaker, the reversal, and the totality of the action. This study describes this agglutinative nature in the imperative and prohibitive morphologies of the Bakairi verbs.

Korean predicative verb forms obligatorily denote the three categories speech level, mood and sentence type which are not handled by most of the automatic word form recognition systems for this language. These categories are marked by special endings. This paper examines predicative verb forms concentrating on the lexical description of these endings' in the framework of Left-Associative Grammar (LAG). Additionally this paper suggests a system to analyse verb forms in these aspects. The results of this study have been implemented using Malaga2 and integrated into an automatic word form recognition system for Korean called KMM (Korean Malaga Morphology).

This study investigates the computational challenges and opportunities presented by the agglutinative structures in Urdu, a language characterized by its complex system of morphemebased word formation. Agglutinative languages, including Urdu, pose significant difficulties in natural language processing (NLP) due to the intricate ways in which morphemes each carrying distinct grammatical or semantic meanings are combined to form words. Despite its linguistic richness and central role among South Asian languages, Urdu has been relatively underrepresented in global computational research, leading to a lack of robust NLP tools tailored to its unique morphological features. This gap highlights the need for extensive linguistic resources, including annotated corpora and models that can specifically address the complexities of Urdu's agglutinative morphology, which remain largely unexplored. Using the Emille Urdu Corpus, this research systematically analyzes the frequency and distribution of agglutinative structures in Urdu. A Python-based annotation process was employed to tag prefixes and suffixes, facilitating a more granular understanding of Urdu morphology. The study highlights key patterns, such as the prevalent use of prefixes like ‫"-نا"‬ (nā-) and ‫"-بد"‬ (bad-) to form words with negative connotations and the transformation of adjectives and verbs into nouns through suffixes like "-‫"گی‬ (gī) and ‫"ی-"‬ (ī). Furthermore, the research explores the limitations of traditional rule-based models in handling Urdu's morphological complexity and advocates for the adoption of machine learning and deep learning techniques. These modern approaches, particularly Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), show promise in accurately modeling Urdu's agglutinative morphology, though they require extensive linguistic data and computational resources. The findings underscore the need for comprehensive linguistic resources and advanced computational models to enhance Urdu NLP. By addressing these challenges, the study aims to contribute to the development of more effective and scalable NLP tools, thereby improving access to Urdu-language content in digital platforms and advancing the broader field of computational linguistics for agglutinative languages.

Рассматриваются максимальные внешнеплоские графы с двумя симплициальными вершинами, уложенные на равносторонней треугольной решётке (РМВП-графы). Слабо двойственные графы РМВП-графов изоморфны и подобны молекулярным графам изомеров сопряжённых полиеновых углеводородов (СПУ). Поэтому РМВПграфы удобно применять для прогнозирования физико-химических свойств СПУ. Предложен новый подход к поиску многочисленных корреляций структура свойство между топологическими индексами (ТИ) РМВП-графов из некоторого заданного множества и физико-химическими параметрами изомеров (СПУ). Сущность такого подхода заключается в предварительном отборе ТИ, пригодных для прогнозирования свойств неразветвлённых изомеров СПУ, с последующим поиском корреляций структура свойство уже между отобранными ТИ и физико-химическими параметрами изомеров СПУ. Пригодность предложенного подхода была доказана с помощью расчётного эксперимента. В рамках этого подхода исследовались ТИ РМВП-графов порядков n = 8, 10, . . . , 28, слабые двойственные графы которых изоморфны и геометрически подобны молекулярным графам модели СПУ, принятой в работе S.

言語能力と言語運用の相互関係の考察に、口答会話試験を用いてその言語運用を評価する場合がある。発話行為を導く場面を設定し、対話を促す文法項目を前もってその試験の受験者に伝達して行うというものが多い。ただ、発話者が内化した言語知識(ラング)を用いて、恣意的な判断で発信するとされる発話活動(パロール)を、画一的な模擬会話での会話交流だけで評価していいものかどうかが常に疑問に残る。バフチンの唱える「対話性」に関する考察が、口答会話試験における評価方法の妥当性に一つの理論的根拠を提供している。本稿では、第二言語としての日本語中級レベルの敬語表現の口答会話試験を例解にして、その「対話的」な談話交流を考察するものである

This paper presents a newly funded international project for machine translation and automated analysis of ancient cuneiform 1 languages where NLP specialists and Assyriologists collaborate to create an information retrieval system for Sumerian. 2 This research is conceived in response to the need to translate large numbers of administrative texts that are only available in transcription, in order to make them accessible to a wider audience. The methodology includes creation of a specialized NLP pipeline and also the use of linguistic linked open data to increase access to the results. The project Machine Translation and Automated Analysis of Cuneiform Languages (MTAAC) 3 fo-1 The Cuneiform script was invented in Ancient Iraq more than 5000 years ago. Signs were drawn, and later impressed, onto a tablet-shaped fresh lump of clay using a reed stylus. This script was in use for 4000 years to record texts in different languages such as Sumerian, Akkadian and Elamite. See figure in section 1b for an example. 2 We would like to thank the reviewers, and Robert K. Englund and Heather D. Baker, for their insightful comments and suggestions.

In this paper, we re-evaluate morph (data-driven subword) and word lexical models used for large vocabulary continuous speech recognition of agglutinative languages. Since such speech recognition systems are applied mostly for information retrieval purposes we ...

本研究は、シンハラ語母語話者が、日本語への文学翻訳を通して産
出した日本語を分析し、母語の影響や類似表現の比較などを通して
両言語の相違を示すことで、翻訳上の問題の克服に寄与することを
目指す。その際、シンハラ語原文の構造分析を行い、両言語の間に
存在する言語類型論的な異なりや、文法上の特徴、語彙の多義性な
どを、また、談話分析の観点から使用文脈を検討し、対照言語学的
観点から記述した。それにより、母語話者の協働による翻訳作業を
通して浮き彫りにされた翻訳上の問題、すなわち、非母語話者の産
出する目標言語の不自然さについての説明を施すことが可能となっ
た。

This paper constitutes an approximation to the automatic treatment ofthe Basque morphology. After considering several alternatives, the Two Level Formalism is adopted and adapted to Basque. On the other han~a morphosyntactic analysis is adopted which is superior to the above mentioned formalism. On the basis of the results of the analysis proposed here, some theoretical problems that the automatization oflanguage, in this case ofBasque, raises are also considered: morpheme partitioning vs. piling, ambiguity, lexical delimitation, ellipsis within words, nominalization and categorial changes.

This paper presents a newly funded international project for machine translation and automated analysis of ancient cuneiform 1 languages where NLP specialists and Assyriologists collaborate to create an information retrieval system for Sumerian. 2 This research is conceived in response to the need to translate large numbers of administrative texts that are only available in transcription, in order to make them accessible to a wider audience. The methodology includes creation of a specialized NLP pipeline and also the use of linguistic linked open data to increase access to the results. 1 Context The project Machine Translation and Automated Analysis of Cuneiform Languages (MTAAC) 3 fo-1 The Cuneiform script was invented in Ancient Iraq more than 5000 years ago. Signs were drawn, and later impressed, onto a tablet-shaped fresh lump of clay using a reed stylus. This script was in use for 4000 years to record texts in different languages such as Sumerian, Akkadian and Elamite. See figure 1 in section 1b for an example. 2 We would like to thank the reviewers, and Robert K. Englund and Heather D. Baker, for their insightful comments and suggestions. 3 The project is generously funded by the Deutsche Forschungsgemeinschaft, the Social Sciences and Humanities Research Council, and the National Endowment for the Humanities through the TAP Digging into Data Challenge. See the project website at https://cdli-gh.github.io/mtaac.

This paper presents a newly funded international project for machine translation and automated analysis of ancient cuneiform 1 languages where NLP specialists and Assyriologists collaborate to create an information retrieval system for Sumerian. 2 This research is conceived in response to the need to translate large numbers of administrative texts that are only available in transcription, in order to make them accessible to a wider audience. The methodology includes creation of a specialized NLP pipeline and also the use of linguistic linked open data to increase access to the results. 1 Context The project Machine Translation and Automated Analysis of Cuneiform Languages (MTAAC) 3 fo-1 The Cuneiform script was invented in Ancient Iraq more than 5000 years ago. Signs were drawn, and later impressed, onto a tablet-shaped fresh lump of clay using a reed stylus. This script was in use for 4000 years to record texts in different languages such as Sumerian, Akkadian and Elamite. See figure 1 in section 1b for an example. 2 We would like to thank the reviewers, and Robert K. Englund and Heather D. Baker, for their insightful comments and suggestions. 3 The project is generously funded by the Deutsche Forschungsgemeinschaft, the Social Sciences and Humanities Research Council, and the National Endowment for the Humanities through the TAP Digging into Data Challenge. See the project website at https://cdli-gh.github.io/mtaac.

日本語初級文法項目「～んです」は、早い段階で学習するものであるにもかかわらず、学習者には実際の会話で容易に取り入れられないものの一つである。この文型には何らかの「前提」となる発話条件が想定されている場合が多く、文脈や状況によってその「発話意図」がさまざまに想起され、「～んです」の用法が多種多様に異なってしまうからである。本学留学生別科の学習者を対象に行った調査でも、前提条件として設定した場面から、「～んです」との繋がりを理解し、適切に使える学生が少なかった。そこで、本稿では、この多岐にわたる用法を持つ「～んです」の使い方を網羅的に一度に教えるのでなく、前提条件との繋がりから、話し手と聞き手の共有情報を明らかにし、そこから想起される「発話意図」のわかりやすいものから小出しにして教えるという教授法を提案するものである

When individuals with speaking disabilities, dysarthrics, try to communicate using speech, they often have to use speech synthesizers which require them to type word symbols or sound symbols. This input method often makes realtime operations difficult and dysarthric users fail to control the flow of conversation. In this study, we are developing a new and novel speech synthesizer where not symbol inputs but hand motions are used to generate speech. In recent years, statistical voice conversion techniques have been proposed based on space mapping between given parallel data sequences. By applying these methods, a hand space and a vowel space is mapped and a converter from hand motions to vowel transitions is developed. It has been reported that the proposed method is effective enough to generate Japanese five vowels. In this paper, we discuss expansion of this system to consonant generation and pitch control. For the former, two methods are examined: waveform concatenation and space ...

In this paper, I will present a quantitative analysis of the frequency and ranking of the following items in modem Korean: (l) prefinal and final endings (the latter are used as conclusive or non-conclusive morphemes), and (2) the grammatical patterns (including adverbial phrases that are connected with the sentential adverbs). In 1997, I conducted the first survey of agg lutinating suffixes, entitled "Quantitative Research on Verbal Morphemes in Modern Korean" (Lee 1997b). <Frequency of Prefinal and Final Verbal Endings> 001 FE127 ta t:J-: 6039 002 FE042 n L 5790 003 FE021 ko:il : 3843 004 FE093 nun ~ : 30~ 005 PFE015 ess ~ : 2454 006 FE371 e <>1 2341 etc. Since then a couple of others' studies on the frequency of suffixes have been reported based on their larger corpus but those of grammatical patterns not yet done by others but me as below in this paper. <Frequency of Grammatical Patterns> 001 Igr.-e(a) kata-<>1(o~) 7tt:J-741 003 1-ci(yo)-Al (k) 539 005 1-ci anhta(-cito anhta) 468-Al ~t:J-(-Al.£ ~t:J-) etc. 002 2-ko iss ta 004 1-e cwuta 006-e(a) ota (gr.=grade)-:il 9J. t:J-fi37-<>1 'ft:J-468-<>1(or) .2.t:J-428 The above findings are useful in the following ways: First, they are instructive in developing a rational basis for pedagogical syllabi and materials in teaching Korean as a foreign or a second language. Second, the findings can also be applied to analyzing the order of presentation of morphemes in elementary Korean materials for children in Korea. Third, the findings are relevant to machine translation because they provide infonnation on which Korean expressions are more preferable in translation.

The paper describes automatic speech recognition experiments and results on the spontaneous Hungarian MALACH speech corpus. A novel morph-based lexical modeling approach is compared to the traditional wordbased one and to another, previously best performing morph-based one in terms of word and letter error rates. The applied language and acoustic modeling techniques are also detailed. Using unsupervised speaker adaptations along with morph based lexical models 14.4%-8.1% absolute word error rate reductions have been achieved on a 2 speakers, 2 hours test set as compared to the speaker independent baseline results.

Abstract Orthographies should not only represent phonology, but meaning and grammar as well, at the morpheme, word and sentence levels. Developing an orthography does not therefore depend on phonological analysis alone, but must be based on a multi-level linguistic analysis. With respect to deciding how to divide a string of morphemes into written words, these different levels may be in conflict, with semantic, grammatical, phonological and sociolinguistic factors all jostling for supremacy. This paper examines a set of nine criteria developed by Van Dyken and Kutsch Lojenga for determining word breaks in a writing system. These criteria compete with one another for dominance in the Bantu languages described, so we focus on cases where the criteria conflict, and present ways in which these conflicts can be minimised and an optimal orthography achieved.

El presente texto pretende hacer una reflexion sobre la forma de educar de hoy los pueblos indigenas, en particular lo wayuu. Esta reflexion se fundamenta en los relatos escritos por el maestro Ramon Paz Ipuana, consignados en su obra magistral, denominada como el libro sagrado de la cultura wayuu: Mitos, leyendas y cuentos Guajiros (1972). Quien haya leido a este autor, podra detectar sin ningun problema que en este articulo estan presentes algunos cuentos de este libro. A lo largo de este texto tambien se tiene en cuenta la experiencia de otro personaje como en el caso de ATAHUALPA. Finalmente es una invitacion para que los docentes wayuu piensen en una educacion desde los pensamientos sagrados de la cultura wayuu; para esto, las comunidades indigenas deben superar primero el complejo de inferioridad y deben dejar de ver al otro –Alijuna- como superior.

Abstract;In this report, we discuss the development of a CALL system for self assessment of Japanese pronunciation by English-speaking learners. Different from the conventional method for CALL system which focuses on average students, the system we developed focuses more on ...

Vous trouverez ci-après la Table des matières et mes remarques sur l’article de Louis Begioni et Alvaro Rocchetti qui soulève un certain nombres de questions en phylogénie du langage considérée seulement pour les langues romanes exclusivement en Europe et donc tardivement dans l’après glaciation, excluant toute connexion avec le Grec (voire l’Arménien), mais faisant deux ou trois fois une comparaison avec l’anglais qui est une langue germanique très créolisée, il est vrai, au contact conquérant du français normand. Le français est fortement signalé comme exceptionnel sans faire référence au contact fort avec le Celte, gaulois, breton et autres, et les langues germaniques (spécificités de contacts importants pour la langue d’Oïl et le Picard).

Dr. Jacques Coulardeau, Université Paris I Panthéon Sorbonne.

Table des matières/ Les petits fours apéritifs

Introduzione

Louis Begioni
Alvaro Rocchetti
L’évolution de l’ordre des mots dans la phrase du latin aux langues romanes ; comparaison avec les langues agglutinantes et isolantes 11

Luciana T. Soliman
Les formes verbales voici/voilà en contexte narratif fictif 27

Sophie Saffi
La phrase en contexte : étude d’une bande dessinée en français et en italien 47

Mohammed Nabih
De l’organisation compositionnelle du discours médiatique électronique 91

Guy Achard-Bayle
Ondrej Pešek
La phrase en cotexte et la hiérarchie des unités mésotextuelles 121

Thomas Franck
Quand citer c’est faire. Analyse de la doxa existentialiste à partir de ses citations décontextualisées 161

Francesco Parisi
Le Schizo et les langues : la frase tra letteratura, linguistica e filosofia 177

Houda Landolsi
Le Rassemblement National et les prénoms. Construction du sens intégral dans le discours sur la francisation des prénoms : une affaire de… contexte ? 201

本研究では,ベトナム人日本語学習者(VS)の音声コミュニケーション上の問題点の分析、発音熟達者のラーニングヒストリーの分析、シャドーイングの効果の検証を行い、以下の成果を得た。1)VSの音声は他の母語話者(に比べ自然さも印象も低い. 2)VSの音声には喉頭の緊張が頻繁に起こり、苦しそうに聞こえる原因となっている。1)2)共に母語の影響が考えられる 3)シャドーイングを用いた発音訓練前後のVSの発音の韻律的特徴を分析したところ、発話速度などにおける問題点が明確になり、これらの多くにはシャドーイングの訓練が有効である。4)発音熟達者のラーニングヒストリー調査を通じて,発音熟達者は共通して「高さ」に対する意識が強い.The difficulties of Japanese pronunciation for Vietnamese learners of Japanese are found not only at the segmental level but also at the suprasegmental/prosodic level. For example, the articulation of each vowel and consonant, interference from tones in Vietnamese, articulatory tension in the larynx, and the realization of short/long mora. In this research project, we carried out a perception experiment on the level or degree of clarity of the learners’ pronunciation by native Japanese speakers. We also conducted a survey on the learning history of Vietnamese “goo d learners” to discuss a more effective learning method of Japanese pronunciation.研究課題/領域番号:23652114, 研究期間(年度):2011-201

This thesis analyses 'incomplete sentences' in languages which utilise distinctively agglutinative components in their morphology. In the grammars of the languages dealt with in this thesis, there are certain types of sentences which are variously referred to as 'elliptical sentences' (Turkish eksiltili cümleler), 'incomplete sentences' (Uzbek to'liqsiz gaplar), 'cut-off sentences' (Turkish kesik cümleler), etc., for which the grammarians provide elaborated semantic and syntactic analyses. The current work attempts to present an alternative approach for the analysis of such sentences. The distribution of morphemes in incomplete sentences is examined closely, based on which a system of analysis that can handle a variety of incomplete sentences in an integrated manner is proposed from a morphological point of view. It aims to aid grammarians as well as researchers in area studies by providing a simple description of incomplete sentences in agglutinative languages. The linguistic data are taken from Turkish, Uzbek, and Japanese, with special reference to (Bukharan) Tajik.

Natural Language Processing (NLP) research is still in its infancy in Africa. Most of languages in Africa have few or zero NLP resources available, of which Igbo is among those at zero state. In this study, we develop NLP resources to support NLP-based research in the Igbo language. The springboard is the development of a new part-of-speech (POS) tagset for Igbo (IgbTS) based on a slight adaptation of the EAGLES guideline as a result of language internal features not recognized in EAGLES. The tagset consists of three granularities: fine-grain (85 tags), medium-grain (70 tags) and coarse-grain (15 tags). The medium-grained tagset is to strike a balance between the other two grains for practical purpose. Following this is the preprocessing of Igbo electronic texts through normalization and tokenization processes. The tokenizer is developed in this study using the tagset definition of a word token and the outcome is an Igbo corpus (IgbC) of about one million tokens. This IgbTS was appl...

国立国語研究所 研究系 音声言語研究領域 非常勤研究員国立国語研究所 研究系 音声言語研究領域国立国語研究所 研究系 音声言語研究領域 技術補佐員千葉大学国立国語研究所 研究系 音声言語研究領域Adjunct Researcher, Spoken Language Division, Research Department, NINJALSpoken Language Division, Research Department, NINJALTechnical Staff, Spoken Language Division, Research Department, NINJALChiba UniversitySpoken Language Division, Research Department, NINJAL国立国語研究所共同研究プロジェクト「大規模日常会話コーパスに基づく話し言葉の多角的研究」では，2016年度から大規模な日常会話コーパスの構築に取り組んでいる。日常生活において自然に生じる活動に埋め込まれた多様な会話を収録するために，本コーパスでは挑戦的な収録法を採用している。この収録法では，調査協力者にビデオカメラやICレコーダーなどの収録機材を2～3ヶ月間ほど貸し出し，日常生活における多様な場面での会話を自ら収録してもらう。プロジェクトメンバーは収録場面に立ち会わないため，調査協力者は，調査の主旨や公開方法などを他の会話参加者に説明し，データの収録及び公開に関する同意書を取得する必要がある。本稿では，テスト収録を踏まえて確立した収録法について報告する。また，これまでの収録状況についても合わせて報告する。In 2016, we launched a new corpus project, “A Multifaceted Study of Spoken Language Using a Large-scale Corpus of Everyday Japanese Conversation,” in which we built a large-scale corpus of everyday Japanese conversation collected in a balanced manner. In order to record various...

The paper describes automatic speech recognition experiments and results on the spontaneous Hungarian MALACH speech corpus. A novel morph-based lexical modeling approach is compared to the traditional wordbased one and to another, previously best performing morph-based one in terms of word and letter error rates. The applied language and acoustic modeling techniques are also detailed. Using unsupervised speaker adaptations along with morph based lexical models 14.4%-8.1% absolute word error rate reductions have been achieved on a 2 speakers, 2 hours test set as compared to the speaker independent baseline results.

In this paper, we re-evaluate morph (data-driven subword) and word lexical models used for large vocabulary continuous speech recognition of agglutinative languages. Since such speech recognition systems are applied mostly for information retrieval purposes we ...

異なる速さで読み上げられた音声がどのような時間的特性を持つのかについて明らかにするために、6 種類のテキストを3 種類の速さで読み上げた音声を分析した。その結果、発話速度は意図した速さにしたがって変化するが、調音速度はあまり変化しない場合があった。また、句点位置にあるポーズは速さにしたがって時間長が変化するが速さに関わらずほぼ全ての位置にポーズが挿入された。それに対し、読点位置にあるポーズは速くなってもその時間長はあまり変化しない代わりに、速さが速くなるにしたがってポーズが挿入されにくくなる傾向が見られた

Korean predicative verb forms obligatorily denote the three categories speech level, mood and sentence type which are not handled by most of the automatic word form recognition systems for this language. These categories are marked by special endings. This paper examines predicative verb forms concentrating on the lexical description of these endings' in the framework of Left-Associative Grammar (LAG). Additionally this paper suggests a system to analyse verb forms in these aspects. The results of this study have been implemented using Malaga2 and integrated into an automatic word form recognition system for Korean called KMM (Korean Malaga Morphology).

International audienceWe conclude this book by briefly highlighting its most important contributions to the creation of new typological knowledge on the syntax of the indigenous languages of Mesoamerica from the light shed by their RC structure. This exercise leads us to further propose some lines for future research that we believe are essential to cover if we want our knowledge of this area to be more complete

In experimental psycholinguistics, one clue into the architecture of lexical memory comes from the presence of robust frequency effects in lexical decision task (LDT), in which subjects judge whether a written stimulus is a real word or a nonword, and processing complexity is measured by reaction time (RT). For example, in LDT the visual word recognition process is facilitated (or inhibited) by word frequency as measured from the representative corpus. Our study verifies the word frequency effect in standard (“yes/no task”) LDT performed by Khalkha Mongolian subjects. The results showed strong weight of word frequency as RTs predictor (R2 = .631, F (1, 28) = 50.57, p &lt; .000, β = .802, t = 7.111, p &lt; .000). Our experimental results also correspond to experimental findings on word frequency effects for Japanese Katakana (syllabic) and Kanji (logographic) words in standard LDT. Such lexical decision “script moderation” could be the actual clue for further LDT experiments (e. g., ...

Resumen En este artículo presentamos el trabajo realizado en la extracción automática de información sobre la aparición de complementos y adjuntos para un conjunto de 1.400 verbos a partir de un corpus periodístico de un millón y medio de palabras. Los resultados han sido evaluados, obteniéndose una precisión y cobertura satisfactorias. Estos datos se usarán para la adquisición manual y automática de información sobre subcategorización verbal.

In experimental psycholinguistics, one clue into the architecture of lexical memory comes from the presence of robust frequency effects in lexical decision task (LDT), in which subjects judge whether a written stimulus is a real word or a nonword, and processing complexity is measured by reaction time (RT). For example, in LDT the visual word recognition process is facilitated (or inhibited) by word frequency as measured from the representative corpus. Our study verifies the word frequency effect in standard ("yes/no task") LDT performed by Khalkha Mongolian subjects. The results showed strong weight of word frequency as RTs predictor (R2 = .631, F (1, 28) = 50.57, p < .000, β = .802, t = 7.111, p < .000). Our experimental results also correspond to experimental findings on word frequency effects for Japanese Katakana (syllabic) and Kanji (logographic) words in standard LDT. Such lexical decision “script moderation” could be the actual clue for further LDT experiments (e. g., relatively “deep” Mongolian script vs. "shallow” Cyrillic Mongolian).

国立国語研究所 研究系 音声言語研究領域 非常勤研究員国立国語研究所 研究系 音声言語研究領域国立国語研究所 研究系 音声言語研究領域 技術補佐員千葉大学国立国語研究所 研究系 音声言語研究領域Adjunct Researcher, Spoken Language Division, Research Department, NINJALSpoken Language Division, Research Department, NINJALTechnical Staff, Spoken Language Division, Research Department, NINJALChiba UniversitySpoken Language Division, Research Department, NINJAL国立国語研究所共同研究プロジェクト「大規模日常会話コーパスに基づく話し言葉の多角的研究」では，2016年度から大規模な日常会話コーパスの構築に取り組んでいる。日常生活において自然に生じる活動に埋め込まれた多様な会話を収録するために，本コーパスでは挑戦的な収録法を採用している。この収録法では，調査協力者にビデオカメラやICレコーダーなどの収録機材を2～3ヶ月間ほど貸し出し，日常生活における多様な場面での会話を自ら収録してもらう。プロジェクトメンバーは収録場面に立ち会わないため，調査協力者は，調査の主旨や公開方法などを他の会話参加者に説明し，データの収録及び公開に関する同意書を取得する必要がある。本稿では，テスト収録を踏まえて確立した収録法について報告する。また，これまでの収録状況についても合わせて報告する。In 2016, we launched a new corpus project, “A Multifaceted Study of Spoken Language Using a Large-scale Corpus of Everyday Japanese Conversation,” in which we built a large-scale corpus of everyday Japanese conversation collected in a balanced manner. In order to record various...

Extensive development of web 2.0 has led to production of gigantic amount of user generated data. These data consist of many useful information. Manual analyzing these data and classifying sentiment in them, is an exhausting task, thus opinion mining method is needed. Opinion mining approach uses natural language processing where Part-of-Speech (POS) Tagging is a crucial part. The performance of any NLP system depends on the accuracy of a POS tagger. Two main issues that affect the accuracy of POS tagger are unknown words and ambiguity. Although research on POS tagging has been back dated few decades ago, yet they have been mostly focused on English. Research on Malay language is still in the early stage. Also, online Malay Text differs from proper Malay text, in the sense of structure and also grammar. Online users tend use a lot of abbreviations and short forms in their text. Besides this, the "BahasaRojak" phenomena complicate tagging process even further. Thus taking all these into consideration, in this study, we will review stochastic and rule-based POS tagging methodologies to deal with ambiguous and unknown words on online Malay text.