Search

Compound adverbs represent an interesting issue in terms of Automatic Morphological Analysis (AMA). The reason is that compound adverbs in Czech are expressions formed by compounding existing words that are different parts of speech without any change in their form. An indicative sign of compound adverbs is that they can always be decomposed again. Compound adverbs may be written as one word but sometimes a multiword form coexists. A word that is originally a different part of speech gains an adverbial meaning and becomes an adverb. This article presents the results of a corpus probe aimed at mapping expressions that are demonstrably compound adverbs and were not recognized by AMA or were incorrectly tagged by AMA as another part of speech. Analysis of data obtained from the Czech National Corpus (ČNK) SYN v3 show that the unrecognized and incorrectly tagged units can be divided into several groups. Based on knowledge of these groups it is possible to refine part of speech tagging by AMA. The corpus probe examined units written in accordance with the current codification as well as substandard units.

The paper describes the construction and testing of an electronic application for automatic morphological analysis of Old English. It introduces resources and methodologies at our disposal based on the state of the art in the field of electronic analysis of Old English and on an overview of Old English morphology. A detailed account of the chosen methodology is offered and a specific description of the implementation is provided: from the acquisition and preparation of the input data and choice of technology to the programming and testing of the results. The resulting recall of 95% can be seen as a success of the project, however, the paper also shows how the recall may be improved. It also discusses further use of the analyser, especially the disambiguation of its results. The paper makes a future semi-automatic morphological tagging of Old English texts a real possibility. Powered by TCPDF (www.tcpdf.org)

Straipsnyje rašoma apie lietuvių kalbos morfologinio anotatoriaus veikimo principus, automatinės morfologinės analizės specifiką. Didžiausias dėmesys skiriamas vienam iš 2007-2008 m. Valstybinio mokslo ir studijų fondo remto projekto "Internetiniai ištekliai: anotuotas lietuvių kalbos tekstynas ir anotavimo priemonės (ALKA2)" įgyvendintų darbų - lietuvių kalbos morfologinio anotatoriaus leksinės duomenų bazės pildymui. Išsamiai aprašoma į morfologinio anotatoriaus leksikos duomenų bazę įtrauktinų žodžių atranka, morfologinio anotavimo etapai, sunkumai, su kuriais susidurta atliekant šį darbą. Morfologinio anotatoriaus leksikos bazė padidinta 24 000 žodžių (daugiausia tikrinių ir bendrinių daiktavardžių), todėl tikimasi, kad gana žymiai pagerės morfologinio anotatoriaus kokybė ir bus išvengta daugybės neatpažintų žodžių. Šiame straipsnyje norėta parodyti anotavimo procesą; atskleisti, kad kyla sunkumų ne tik vertinant, ar nauji žodžiai teiktini, ar neteiktini, reikalingi lietuvių kalbai ar galima apsieiti be jų; sudėtinga ne tik nustatyti naujų žodžių reikšmes, bet taip pat ir analizuoti morfologiškai: nustatyti linksniavimo paradigmą, giminę, kaitymą skaičiais, darybinius vedinius ir pan. The article presents the principles of the morphological annotator and the peculiarities of automatic morphological analysis. The paper focuses on building the lexical database of the Lithuanian morphological annotator, which is one of the completed tasks of the project Internet Resources: Annotated Corpus of the Lithuanian Language and Tools of Annotation (ALKA 2), implemented in 2007-2008 and sponsored by the Lithuanian State Science and Studies Foundation. The selection of words to be included into the lexical database of the morphological annotator is described in detail. The stages of morphological annotation and difficulties in this paper are also discussed. The lexical database of the morphological annotator has increased by 24 000 words (mostly proper and common nouns). Therefore it is expected that the quality of the morphological annotator will improve considerably and many unrecognized words will be avoided. The goal of the article is to show the process of annotation. It reveals that problems arise not only during the evaluation of acceptability of new words for the Lithuanian language and the identification of their meanings, but also during their morphological analysis. It is difficult to determine their declension paradigms, gender, number inflection, derivatives, etc.

Straipsnyje pristatoma automatinė morfologinė lietuvių kalbos analizė ir automatiniu būdu sulemuotas bei morfologiškai anotuotas 1 mln. žodžių tekstynėlis. Tam naudojama kompiuterinė programa „Lemuoklis“, automatiškai nustatanti rašytinės žodžio formos antraštinį pavidalą ir galimas tos formos morfologines pažymas. Nagrinėjant Vytauto Didžiojo universiteto Kompiuterinės lingvistikos centre sudarinėjamą automatiškai anotuotą tekstyną, išryškėjo didelis morfologinis daugiareikšmiškumas (apie 40 proc. visų formų yra morfologiškai daugiareikšmės). Lingvistas, norėdamas vienareikšminti tokias formas, dažnai susiduria su problema, kurią formą palikti. Paaiškėjo, kad nėra aiški nekaitomų žodelių klasifikacija, ne visada aiškiai nustatomos tarnybinių kalbos dalių ribos. Iškilo klausimas, kaip skirti visiškai sutampančias kaitomas ir nekaitomas kalbos dalis, kokios morfologinės kategorijos būdingos vienoms ar kitoms kalbos dalims, kaip galima vienareikšminti morfologiškai daugiareikšmes formas bei žodžius. The article deals with the automatically tagged corpus of the Lithuanian language. The corpus with morphological tags has shown a high level degree ambiguity of the language: about 40 percent of word forms are ambiguous. The corpus linguistics has directed attention to such issues, which have not been analysed using methods of traditional linguistics. The morphological ambiguity of language has become obvious only in this automatically tagged corpus. The computational program „Lemuoklis“, created by V. Zinkevičius, can define lemmas and morphological categories of word forms. The morphological ambiguity has appeared only in texts, which were processed by this program. It is very important for automatic morphological analysis to define clearly parts of speech because the accuracy of such analysis can help to avoid morphological ambiguity. But often it is difficult to choose the right form even for the human tagger, as dictionaries and grammars do not agree about how to define parts of speech and some other morphological categories. Some problematic aspects of the Lithuanian morphology are analysed in this article. Very often it is difficult to decide which part of speech a non-inflective word belongs to, what the boundaries of some words are, how one could separate some ambiguous inflective and noninlective words, what morphological categories some parts of speech have.

Un problema crucial en el desarrollo de los sistemas para el análisis morfológico automático de los idiomas flexivos es el tratamiento de las alternaciones de la base. Los modelos existentes requieren el desarrollo de las reglas correspondientes que especifican qué variantes de la base se pueden generar de la variante dada. Un gran número de tales reglas (por ejemplo, para el lenguaje ruso alrededor de un mil) no tiene ninguna interpretación lingüística razonable. Sugerimos un método que evite el uso de tales reglas gracias a la generación y verificación de las hipótesis sobre las formas gramaticales posibles. Los métodos de este tipo –conocidos como análisis a través de generación– hacen el desarrollo de sistemas mucho más simple que el el enfoque directo estándar. Un sistema para el análisis y la generación morfológica para el lenguaje ruso, desarrollado con nuestro método está disponible sin costo para el uso académico; el sistema para el español está bajo desarrollo. A crucial problem in development of systems for automatic morphological analysis for inflective languages is the treatment of stem alternations. The existing models require development of the corresponding rules that specify what stems can be generated from a given one. Many of such rules (e.g., for Russian about a thousand) do not have any reasonable linguistic interpretation. We suggest a method that avoids the use of such rules by generating and verifying the hypotheses about possible grammatical forms. The methods of such type are known as analysis through generation; they make the system development much simpler than the standard direct approach. A morphological analysis and generation system for Russian developed with our method is freely available for academic use; a Spanish system is being implemented. Work done under partial support of Mexican Government (CONACyT and SNI) and CGEPI-IPN, Mexico.