Morfologiškai daugiareikšmių formų atsiradimo priežastys

MIGLĖ ŽEMRIETĖ Institute of the Lithuanian Language ORCID ID: 0009-0000-2174-3311 RESEARCH AREAS: corpus linguistics, computational linguistics, automatic morphological analysis, morphological ambiguity and its disambiguation. DOI: doi.org/10.35321/bkalba.2023.96.06 CAUSES OF MORPHOLOGICAL AMBIGUITY IN FORMS KEYWORDS: morphological ambiguity, automatic morphological analysis, corpus, homoforms, dictionary. ABSTRACT Drawing on data from the Dictionary of Lithuanian Homoforms, this article discusses the causes of Lithuanian morphologically ambiguous forms, referred to in the article as homoforms. Based on their causes, homoforms can be classified into the following groups: more theoretical homoforms arising from the specific compilation practices of certain dictionaries; homoforms arising from differences in prosodic features; homoforms arising from the homonymy of different words; homoforms determined by the Lithuanian morphological system; and homoforms with the same lexical meaning arising from assignment to different paradigms. These causes of homoform formation help explain why particular cases of morphological ambiguity arise during annotation by morphological annotators, and why fully accurate automatic annotation results cannot be obtained. ABSTRACT In the article, the reasons for the homoforms of the Lithuanian language are described on the basis of data obtained from the Dictionary of Homoforms of the Lithuanian Language. The author offers a classification of homoforms according to the reasons of their appearance, which is as follows: homoforms that are more theoretical in nature and appeared due to the specifics of creating dictionaries; homoforms resulting from differences in prosodic elements; homoforms resulting from different overlapping words; homoforms that arose due to the specifics of the Lithuanian morphology system; homoforms of the same meaning that appeared under different paradigms. These reasons help us to understand why annotation by morphological annotators often result in different cases of morphological ambiguity. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 107 INTRODUCTION Nowadays, various computer programs are most commonly used in language research. They make it possible to process language, solve linguistic problems, segment text, lemmatize, classify, cluster, or correct textual errors, and translate automatically. Computer programs can also be used to perform semantic, morphological, and syntactic analysis, and to create annotated or unannotated corpora, speech corpora, and so on. The Lithuanian language is also being computerized. One stage in the computerization of Lithuanian is automatic morphological analysis. Morphological ambiguity becomes apparent during this process. Morphological ambiguity is just one small part of ambiguity, which includes: 1) inflected and uninflected word forms or words; 2) word forms or words belonging to different or the same parts of speech; 3) words or their forms that differ in certain forms or prosodic elements, as well as those that are completely identical (Rimkutė 2003a: 63). Words or their forms are considered morphologically ambiguous when two or more lemmas are assigned to the same word form (e.g., the form kovų can be assigned the lemmas kovas and kova) or two or more grammatical features (e.g., naktis — singular nominative or plural accusative). According to Erika Rimkutė (2002: 86), morphological ambiguity revealed through automatic morphological analysis is a very specific case; however, people encounter ambiguous words every day. This usually does not cause communication problems because, by understanding the context, people can readily infer which word or form has been used, while the other options do not even come to mind. In addition, in spoken language, differences in intonation or stress placement often help identify morphologically ambiguous words. Although morphological ambiguity is essentially a solved problem (see Rimkutė, Daudaravičius 2007), even after improvements to Lithuanian morphological taggers, it is not always possible to choose the most likely one from among several morphologically ambiguous forms; moreover, cases are not uncommon in which a morphological tagger incorrectly selects one form from several possible ones. Lithuanian can be analyzed using three fairly comprehensive and publicly available morphological taggers: the tagger available in the System for Syntactic and Semantic Analysis, Semantika.lt1, and the taggers Lemuoklis 1 See https://semantika.lt/Analysis/TextAnalysis. 108 BENDRINĖ KALBA | 96 and Morfuoklis2. The data presented in this article are based on the analysis performed using Lemuoklis, so this annotator will be presented in somewhat greater detail. The aim of this article is to describe the causes of the emergence of Lithuanian homophones (this and other terms will be explained in Section 2). The objectives are to review terms related to morphological ambiguity and, based on the cases of morphological ambiguity analyzed, describe and classify them, and discuss the causes of the emergence of homophones. The words in the Dictionary of Lithuanian Homophones (LKHŽ)3 were analyzed. The study employed qualitative analysis, corpus linguistics, and descriptive methods. The LKHŽ presents one million non-repeating examples of homophones found in a morphologically annotated corpus. The dictionary shows which Lithuanian word forms are morphologically ambiguous. Readers can draw their own conclusions about which forms are more likely and what determines this. The dictionary contains about 36,000 homophones, classified into three categories: 1) homophones of inflecting parts of speech; 2) homophones of invariable parts of speech; 3) homophones of inflecting and invariable parts of speech. Homophones of inflecting parts of speech, as well as homophones of inflecting and invariable parts of speech, are further divided into 53 types (which will be mentioned below). E. Rimkutė, who prepared a morphologically annotated Lithuanian corpus, found that morphological ambiguity encompasses both inflecting and non-inflecting parts of speech (Rimkutė 2006a: 136). According to her, inflecting parts of speech most often coincide, and inflecting and non-inflecting parts coincide least often. Although the dictionary does not provide statistical data4, judging by the types of homophones to which the largest number of word forms are assigned, the following types of homophones can be mentioned as common: nouns and adjectives of the (i)a declension in the singular nominative coinciding with nouns and adjectives of the (i)o declension in the plural accusative (e.g., nemažas, pripažintas, trečias); nouns and adjectives of the (i)o and (i)a declensions coinciding in the singular accusative (e.g., abejingą, aštuntą, parduotą); nouns and adjectives of the (i)o and e declensions coinciding in the singular genitive and plural nominative and vocative (e.g., abatijos, abejingos, abstrakčiausios) (for a more detailed account of the dictionary’s structure and compilation, see the article’s author’s work: Piečytė 2023). 2 See https://sitti.vdu.lt/morfuoklis/. 3 This dictionary is being prepared for publication. It has not yet been made publicly available. The authors are M. Žemrietė and E. Rimkutė. 4 Calculating frequency would require a separate study. However, examination of some homophones in corpora shows that some may occur thousands of times, while others occur only a few times. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 109 The source of the LKHŽ and, consequently, of this article is a one-million-word Lithuanian corpus, morphologically annotated in 2000–2005 using Lemuoklis, a program developed by Vytautas Zinkevičius. Almost half of the words in the morphologically annotated corpus are morphologically ambiguous. This phenomenon came to light only when researchers began to process files annotated by an automatic morphological analysis program and saw how language is treated without contextual knowledge, that is, when each word form is analyzed as entirely separate (Rimkutė 2006a: 136). Thus, this article and the explanations it contains should be useful to those who use Lithuanian morphological annotators. It should help them understand the causes of possible inaccuracies. The LKHŽ is the first dictionary of its kind for Lithuanian. It presents an unconventional view of language through the lens of computer programs for Lithuanian language analysis. The dictionary shows what kind of language computer programs “understand.” Foreigners learning Lithuanian may also form a similar impression of the Lithuanian language. Since the dictionary contains a truly large number of homophones, it opens up further research opportunities for scholars. Although, as mentioned, this is the first such dictionary in Lithuania, similar dictionaries have already been published for other languages. For example, a Russian homophone dictionary was created in 2001 by Zhanna Grigoryevna Anoshkina (Словарь омонимичных словоформ5). This dictionary includes both homonyms and homographs (words that are spelled alike but pronounced differently). Stress is not taken into account, and the letter ė is not used. The dictionary is arranged so that word forms appear on the left and lexemes on the right. Parts of speech are given in parentheses, cf.6: ехидна ехидной ехидна (noun), ехидный (adjective) ехидною ехидны ехидно ехидно (adverb), ехидный (adjective) The Russian homoform dictionary is available online: 5 http://cfrl.ruslang.ru/homoforms/index.htm. 6 This example illustrates the declension and parts of speech of the word ехидна (their abbreviations are given in parentheses). In Russian, this word has more than one meaning, depending on the context. The word ехидна can mean a hedgehog, an Australian snake, or be used to describe an angry, cruel, deceitful person in Lithuanian. 110 BENDRINĖ KALBA | 96 An English homophone dictionary was created in 20087; its author is Alfred Aloisi (for more on the relationship between homophones and homoforms, see Section 2 of this article). The dictionary description states that it is continuously expanded, but provides no information about where the words are collected from, how many there are, and so on. Contact with the dictionary’s author revealed that its first version was printed, but it was soon realized that creating an online version would be more practical, since a printed one could quickly become outdated and more than one edition would be needed. The dictionary contains more than 10,000 homophones, and users can also suggest entries. In addition, it provides not only homophone lemmas and their parts of speech, but also definitions—something that is not very common in dictionaries of this type. It includes words that are spelled differently but pronounced alike, for example, moo (Lith. mū) and moue (Lith. grimasa), cf.: moo 1.:: verb-intransitive To emit the deep, bellowing sound made by a cow; low. 2.:: noun The lowing of a cow or a similar sound. moue 1.:: noun A small grimace; a pout. English homophones are also included in printed dictionaries; for example, Leslie Presson’s Dictionary of Homophones was published in 1997. It is intended for learners of English and contains more than 600 pairs of homophones. In 2012, A. D. Reed compiled the dictionary Reed’s Homophones: A Comprehensive Book of Sound-alike Words (the dictionary lists American English homophones, with more than 1,000 pairs. Each word is accompanied by a definition, part of speech, and usage example). Books, textbooks, and scholarly articles on similar topics have also been produced in Lithuania. In 2004, Jonas Šukys published the book Panašiai skambantys žodžiai: normos ir klaidos, which provides an in-depth analysis of paronyms (the term will be explained in Section 2). In the book, the scholar explains the standard meanings of the compared words, identifies and corrects erroneous usage, and also discusses aspects of the use of paronyms and homographs. Some homoforms arise from different words with different meanings that happen to coincide. Rūta Marcinkevičienė’s (2011) textbook Žodžio reikšmė. Žodynai ir tekstynai is devoted to the analysis of such words; one of its chapters discusses polysemy, describes homonyms in detail, and presents their classification. 7 An English homophone dictionary is available online: https://www.homophone.com. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 111 E. Rimkutė has published particularly many works on morphological ambiguity. Her dissertation (Rimkutė 2006a) and several scholarly articles (Rimkutė 2002, 2003a, b) may be mentioned; they describe various types of ambiguity in detail, devoting particular attention to morphological ambiguity and homophones. This article consists of an introduction (Part One), two theoretical sections, two research sections, and a summary. A list of all cited sources is provided at the end of the article. 1. MORPHOLOGICALLY ANNOTATED CORPUS AND THE MORPHOLOGICAL AMBIGUITY REVEALED IN IT The first morphologically annotated corpus of Lithuanian is an electronic database created by Vida Žilinskienė and Laima Grumadienė. It was used specifically to compile frequency dictionaries (Rimkutė 2006b: 34). V. Žilinskienė and L. Grumadienė’s Frequency Dictionary of Contemporary Written Lithuanian: Electronic Corpus (2002) consists of 1.2 million words; the texts it contains are original (not translated) and belong to journalistic, scholarly, literary, or administrative styles. These texts were annotated, and dictionaries were subsequently created on their basis: Frequency Dictionary of Contemporary Written Lithuanian (1997) and Frequency Dictionary of Contemporary Written Lithuanian (1998). The morphologically annotated corpus mentioned in the introduction began to be developed in 2000 at the Centre of Computational Linguistics of Vytautas Magnus University8. It was compiled semi-automatically using the automatic morphological analysis program Lemuoklis. Its first version comprised about 1 million words, but the corpus was later expanded and inaccuracies were corrected9. In the corpus, journalistic texts account for 36 percent, scholarly literature for 24 percent, fiction for 19 percent, administrative-style texts for 2.8 percent, and transcripts of the Seimas of the Republic of Lithuania for 6.8 percent. The corpus was compiled semi-automatically. At the time this corpus was being compiled, the morphological analysis program developed by V. Zinkevičius 8 This corpus is publicly available online: https://clarin.vdu.lt/xmlui/handle/20.500.11821/33. 9 Data for the Dictionary of Lithuanian Homophones were taken from this first version of the corpus, since the aim was to record as many cases of morphological ambiguity in Lithuanian as possible. Later, statistical disambiguation modules were implemented in Lemuoklis (Rimkutė, Daudaravičius 2007), enabling the program to operate more accurately and to produce far fewer morphologically ambiguous forms. 112 BENDRINĖ KALBA | 96 Lemuoklis, which provides lemmas and morphological tags, did not take context into account; the program contained no semantic information, and word forms were usually identified using a list of Lithuanian word roots rather than lists of word forms with their grammatical tags specified (Zinkevičius 2000: 30). This means that the program provided all possible grammatical tags and all possible lemmas (headwords) for word forms (see the examples given in the introduction). It should also be noted that Lemuoklis analyzed written forms without stress marks. For this very reason, homophones could also differ in prosodic features (Rimkutė, Grybinaitė 2004: 75–76). At present, the Semantika.lt annotator is more often used in language technology research because it is more accurate; it can also be maintained, updated, and expanded (Bielinskienė et al. 2017: 2). However, it too does not always annotate correctly—it often fails to recognize loanwords, slang, and abbreviations. Since it cannot perceive the whole language or context in the way a human does, errors occur in annotating morphologically ambiguous words (if a word has several different grammatical meanings, morphological annotators sometimes select the wrong grammatical tag). Very recently, at the end of 2023, a new word and text analysis program, Morfuoklis, based on the Semantika.lt annotator, also became operational. This program can also perform synthesis, i.e., generate a desired word or several words from specified grammatical tags. Most annotation errors occur because of morphological ambiguity: although programs are being improved, they still very often fail to recognize context (or recognize it incorrectly). Recently, considerable attention has been devoted to addressing this problem in the automatic analysis of Lithuanian: various tools have been developed to constrain semantic, syntactic, and morphological ambiguity; rules have been formulated to help disambiguate texts; and morphological ambiguity in Lithuanian has been analyzed and classified (Rimkutė, Grigonytė 2006: 32). Homophones are common not only in Lithuanian. A similar situation has been observed in corpora of other languages: for example, morphologically ambiguous forms account for 46 percent in Czech, about 8.01 percent in Slovenian, 40 percent in Romanian, 38.65 percent in English, and 21.58 percent in Hungarian (Hajič 2004: 173, cited in Rimkutė 2006a: 47). Based on data provided by Estonian scholars (Puolakainen 2012: 194), it can be said that almost half of Estonian words are morphologically ambiguous. Evalda Jakaitienė (2009: 87) teigė, kad būta svarstymų, ar toks reiškinys kaip žodžių daugiareikšmiškumas apskritai gali egzistuoti – esą jeigu viena MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 113 A sequence of sounds has two different meanings; in that case, it should be considered two separate words rather than one polysemous word. It is important to mention that the scholar is speaking here about lexical, not morphological, polysemy. However, this view is not popular—according to Skaistė Aleksandravičiūtė (2008: 275), polysemy is a systematic linguistic phenomenon, since it is linked to the characteristics of the Lithuanian morphological system and case syncretism, and is also connected with the economy and limited capacity of language. 2. KEY TERMS FOR MORPHOLOGICAL POLYSEMY In linguistics, morphological polysemy is most often understood and referred to using several different terms. The following discussion concerns terminology, that is, what to call morphologically polysemous forms: homoforms, homonyms, homographs, or something else (terms associated with other levels of polysemy are not discussed; the focus below is exclusively on cases of morphological polysemy). The broadest term that could in part be used in this article is homonyms, which are most often defined as “words that have the same form and entirely different meanings” (Jakaitienė 2009: 101), as “a word that sounds the same as another word, belongs to the same part of speech, but has a different meaning” (Gaivenis, Keinys 1990), or as “words pronounced alike (there are also homonymic forms and morphemes)” (Šukys 2004: 9). As mentioned earlier, morphologically polysemous forms include inflected and uninflected forms that are pronounced alike or differently. In the analysis of morphological polysemy, lexical meaning is not relevant (this is most often emphasized in descriptions of homonyms); therefore, it is clear that the term homonyms is too broad when referring to morphologically polysemous forms. The term homoforms raises a number of questions for linguists. Some of them (Barauskaitė 1979: 43–44; Barauskaitė et al. 1995: 23) consider homoforms to be different forms of the same word or similarly sounding forms of different words, e.g., aprašai—both the plural nominative of the noun aprašas and the second-person present-tense form of the verb aprašyti. Nikolajus Pavlovičius Kolesnikov also calls homoforms paradigmatic homonyms (Kolesnikov 1987: 7, cited in Nevzorova et al. 2005: 231), while Francis Katamba calls coinciding forms of the same or different parts of speech syncretism (Katamba 1994: 15). 114 BENDRINĖ KALBA | 96 In this article, morphologically polysemous forms are called homoforms. As will become clear from the review of the use of this term below, homoforms is used in this article in a broader sense than is usual in linguistic scholarship. Therefore, without examining all types of homonymy in detail, the following section describes the concept of homoforms in linguistic scholarship. E. Jakaitienė (2009: 103) homoformas apibrėžia kaip atskiras, savo garsine sandara atsitiktinai sutampančias visai nehomonimiškų žodžių poras. Kaip pavyzdį mokslininkė pateikia tokį sakinį: Mama nuo vaiko musės vai-ko, kuriame sutampa daiktavardžio vaikas vnsk. kilmininkas ir veiksmažodžio vaikyti esamojo laiko trečiojo asmens forma. Remiantis mokslininke, gali sutapti ir kai kurių žodžių kelios gramatinės formos, pvz.: lipti, lipa, lipo („judėti aukštyn ar žemyn“) ir lipti, limpa, lipo („kibti, klijuotis“). Čia sutampa veiksmažodžių bendratis ir iš jos padarytos formos. Problemiška tai, kad tokius atvejus kiti priskiria ne homoformoms, o daliniams homonimams (Barauskaitė ir kt. 1995: 24). In Kazimieras Gaivenis and Stasys Keinys’s Dictionary of Linguistic Terms (KTŽ), homoforms are defined as forms of different words that are pronounced and written identically (Gaivenis, Keinys 1990). Thus, cases of case syncretism, in which several cases of the same word coincide, are not classified as homoforms. However, this definition may make it difficult to distinguish homoforms from partial homonyms. The dictionary defines the latter as “homonyms whose morphemes do not all coincide completely, or which belong to different parts of speech” (Gaivenis, Keinys 1990: 80). E. Jakaitienė defines partial homonyms as words “whose grammatical forms either all coincide with some of the forms of another word, or of which at least several forms coincide” (Jakaitienė 2009: 101–102). According to her, in most such cases one word is inflected and the other uninflected; for example, gaila is both a noun meaning pity or regret and an adverb (examples from a morphologically annotated corpus). According to E. Jakaitienė, in analytic languages, partial homonyms are words that may belong to different parts of speech. For example, the English word tender can be a noun meaning an offer, an adjective meaning gentle or sensitive, and a verb meaning “to offer, to submit” (Jakaitienė 2009: 102). Other scholars (Kaljanov 2023: 27) call such cases something slightly different—grammatical homonyms. These are phonetically simi- MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 115 forms of the same word that coincide but have different grammatical meanings, for example, ranka (singular nominative) and ranka (singular instrumental). A definition of homoforms can also be found in the Encyclopedia of the Lithuanian Language (LKE). It states that homoforms are certain coinciding forms of words belonging to different parts of speech, for example, grūdo—the singular genitive of the noun grūdas and the third-person form of the simple past tense of the verb grūsti (LKE 1999: 247–248). This definition does not make clear whether coinciding forms of words belonging to the same part of speech can also be considered homoforms, for example, tapo (in this case, the other principal forms differ: tapyti, tapė and tapti, tampa). Only written, unaccented forms are presented in the morphologically annotated corpus. The accent location or pitch accent of such forms is often relevant—morphologically ambiguous forms frequently differ in prosody and are pronounced differently. Linguists often call such cases homographs. KTŽ (1999: 23–24) states that homographs are “words that are spelled alike but pronounced differently: lìngė (a car part) and liñgė (a bird)”; E. Jakaitienė defines the concept as follows: “words (or their forms) that are spelled alike but pronounced differently (differing in a) accent location or b) pitch accent)” (Jakaitienė 2009: 103). It may also be difficult to understand how homographs differ from homonyms and homoforms. On the basis of Vincas Urbutis’s view cited in LKE (1999: 248) and the definition of homographs given here, homographs are words or their forms that are spelled alike. It is stated that in Lithuanian homographs are usually also homonyms (e.g., the form kartą can be the accusative of both the word karta (i.e., generation) and the word kartas (i.e., time/occasion). The encyclopedia also treats as homographs cases in which word forms belonging to different parts of speech coincide (LKE 1999: 248), for example, apra šai (this form can be either the plural genitive of the noun aprašas or the third-person present-tense form of the verb aprašyti—examples taken from the morphologically annotated Lithuanian corpus). However, there is no general agreement. J. Šukys writes that “words that are otherwise identical and that the DLKŽ numbers as homonyms, but that differ in accent location, accent paradigm, or the pitch accent of the stressed syllable, need not be considered true homonyms, but may be classified as homoforms. Homoforms (also called homonymous forms) occur far more frequently in the language than true homonyms, and often involve entirely different parts of speech: maišo (the genitive of the noun maišas)—maišo (the form of the verb maišyti in the third-person simple past); jos (the pronoun’s 116 BENDRINĖ KALBA | 96 ji plural nominative and singular genitive form)—jos (the third-person future form of the verb joti)” (Šukys 2004: 9). Other scholars share this view, arguing that “there are also homophones that are spelled alike but differ in accent,” e.g., báltų and baltų (báltai and báltas) (Barauskaitė et al. 1995: 248). Klausimų kelia ir N. P. Kolesnikovo pateiktas homografų apibrėžimas (Kolesnikov 1987: 6, cituota iš Nevzorova ir kt. 2005: 231). Pasak jo, homografams gali būti priskirtos: 1) skirtingą gramatinę ir leksinę reikšmę turinčios, bet vienodai rašomos žodžių formos (kaip pavyzdį galima pateikti lietuvišką žodį baltì, t. y. sutampa veiksmažodžio bendratis ir vyr. g. būdvardžio baltas dgsk. vardininko forma); 2) skirtingą leksinę, bet vienodą gramatinę reikšmę turinčios taip pat rašomos žodžių formos (pvz., forma žiedas gali būti trys skirtingi žodžiai, turintys skirtingas leksines reikšmes, t. y. žodis gali reikšti ir augalo dalį, ir papuošalą, ir kelio rūšį); 3) skirtingą gramatinę, bet vienodą leksinę reikšmę turinčios žodžių formos (plg. lietuviško veiksmažodžio turėti esamojo laiko antrojo asmens vnsk. formą turì ir esamojo laiko trečiojo asmens formą tùri). Galima suprasti, kad mokslininkas supina homografų ir homofomų terminus, kadangi šioje klasifikacijoje nagrinėjami ir skirtingų, ir tų pačių reikšmių žodžiai bei jų formos. Tačiau taip suprasti homoformas atrodytų gana logiška, nes morfologiškai anotuotuose tekstuose apie 20 proc. daugiareikšmių žodžių skiriasi būtent prozodiniais elementais, t. y. tarimu (Rimkutė 2002: 7). Jurgis Pakerys uses another term for cases in which differently pronounced forms of the same word coincide: homographic word forms. He emphasizes that these differ from homographs (Pakerys 2009: 130). Vincentas Drotvinas, however, calls homoforms, homographs, and homophones by another term—homonymous word forms (Drotvinas 1986: 91, cited in Rimkutė 2006a: 43). This designation seemed acceptable to E. Rimkutė, the first researcher of morphological ambiguity, but to avoid using such a long term, she decided simply to adopt the term homoforms (Rimkutė 2003b: 36). The author of this article follows the same conception. Another term used for morphologically ambiguous forms is paronyms. These are defined as “words that are pronounced or written similarly but usually have different meanings; sometimes their meanings may be somewhat related, though they are generally not synonymous (in such cases, it is often said that the words share a semantic connection, or belong to the same semantic field)” (Šukys 2004: MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 117 9–10). J. Šukys explains that paronyms in written language may also include homographs, that is, words spelled alike but pronounced differently, e.g., draugė – draugė (“a company, a group”), while in spoken language they may include homophones, that is, words pronounced alike but spelled differently, e.g., rūgti – rūkti. Nevertheless, the term paronym has a fairly clearly defined meaning in linguistics, so it was decided to broaden the meaning of homoform and continue using this particular term. In addition, the author of this article maintains that inflected forms assigned to the same or to different parts of speech, which differ in accent location, pitch accent, or vowel quantity, may be both homoforms and homographs. In other words, some homographs may also be homoforms, and some homoforms may also be homographs (when pronunciation is taken into account and the forms are analyzed morphologically). Moreover, some specific cases remain undefined in linguistic studies; for example, it is unclear what to call instances where two or more invariable parts of speech coincide. Therefore, following E. Rimkutė (2006a: 44), this study decided to classify such instances as homoforms of invariable parts of speech. In the following sections, the previously outlined conception of homoforms is used when discussing them. Chapter Four presents a classification of homoforms—that is, units of morphological ambiguity—based on data from the morphologically annotated corpus. The article uses primarily the terms morphologically ambiguous forms and homoforms. 3. CLASSIFICATION OF LITHUANIAN HOMOFORMS ACCORDING TO COINCIDING PARTS OF SPEECH As mentioned above, coinciding forms belonging to the same or different parts of speech are called homoforms in this article. E. Rimkutė (2006a: 40) distinguished three types: 1) homoforms of inflecting parts of speech; 2) homoforms of inflecting and uninflecting parts of speech; 3) homoforms of uninflecting parts of speech. Homoforms of inflecting parts of speech are morphologically coinciding forms of the same or different parts of speech (Rimkutė 2006a: 52). E. Rimkutė gives the following sentences as examples: lišk į būdą (the singular accusative of the noun būda) and sit vs. he found a new method of treatment (the singular accusative of the noun būdas). As can be seen, the singular accusative forms of the nouns būda and būdas coincide here. However, it is not necessarily the forms of different words belonging to the same part of speech that must coincide 118 BENDRINĖ KALBA | 96 — forms of two different inflecting parts of speech, or two different forms of the same word, may also coincide. E. Rimkutė gives the following example: English kalbà (noun) vs. some Lithuanian politicians constantly kaĩba (verb) absurdus. As can be seen, some forms of the noun kalba and the third-person present-tense form of the verb kalbėti coincide in these sentences. E. Rimkutei (2006a: 48) išanalizavus 10 000 morfologiškai anotuotų žodžių formų, esančių morfologiškai anotuotame tekstyne, paaiškėjo, kad kaitomųjų kalbos dalių homoformos sudaro didžiausią morfologinio daugiareikšmiškumo dalį – net 76,2 proc. Homoforms of inflecting and uninflecting parts of speech are coinciding forms of two or more different parts of speech, one of which is inflecting and the other or others uninflecting (Rimkutė 2006a: 59). This category includes homoforms involving nouns, adverbs, prepositions, conjunctions or particles; adjectives and adverbs; verbs, adverbs, prepositions or particles, and other coinciding parts of speech. They account for 6.3 percent of all cases of morphological ambiguity in the morphologically annotated corpus. As an example, the scholar gives the following word combinations: námo (noun) roof vs. I am going namõ (adverb). It can be seen that a noun and an adverb coincide here. Homoforms of inflecting and uninflecting parts of speech most often consist of three or more coinciding parts of speech, that is, they are multiple-member forms. Homoforms of uninflecting parts of speech are two or more coinciding uninflecting parts of speech. This category includes coincidences involving conjunctions, particles, adverbs, and so on. Homoforms of uninflecting parts of speech are the second most frequent type of homoform, accounting for 17.5 percent of all homoforms in the morphologically annotated corpus (Rimkutė 2006a: 63). In compiling the LKHŽ, these three categories were subdivided into 53 more specific types. The classification proposed by E. Rimkutė (2006a) was essentially followed, but expanded. An excerpt from the LKHŽ is presented in Figure 1. In the dictionary, a coinciding word or word form is printed in bold; the next line indicates which parts of speech the form may be classified as; the headword is given in angle brackets, followed by detailed grammatical labels. Uninflecting parts of speech are presented in alphabetical order; inflecting forms, as well as forms of inflecting and uninflecting parts of speech, are ordered according to the parts of speech that coincide, and within each category alphabetically. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 119 FIG. 1. Excerpt from the Dictionary of Lithuanian Homophones Some homophones may be assigned to several groups; for example, vieši is both the nominative plural form of the adjective viešas and a homophone of the verb viešėti; it is also the syncretic present-tense form of the second and third person singular of the verb viešėti. pirmiausia is both an adjective and a numeral, and also a syncretic form of those parts of speech in the feminine singular nominative and instrumental cases and in the neuter gender. Therefore, the same word may appear several times in the dictionary under different types of homophones. This means that the dictionary does not immediately list all possible grammatical tags for a given form, but only those that apply to a particular type of syncretism, such as syncretism between nouns and adjectives, and so on. Nevertheless, most cases of morphological ambiguity in the dictionary can be assigned to only one type of homophone. The syncretic forms of inflecting parts of speech in LKHŽ are divided into 41 types. Homophones involving inflecting and uninflecting parts of speech are divided into 11 types in the dictionary. Syncretic forms of uninflecting parts of speech are not further subdivided in the dictionary, since its compilation was based specifically on E. Rimkutė’s (2006a) classification. She assigned all syncretic forms of uninflecting parts of speech to a single general type. 120 BENDRINĖ KALBA | 96 4. CLASSIFICATION OF LITHUANIAN HOMOPHONES ACCORDING TO THEIR CAUSES Since LKHŽ contains a truly large number of morphologically ambiguous words (about 36,000), it is possible to attempt to identify the causes of homophones in Lithuanian. Taking into account the data in the dictionary of homophones and the morphological analyses offered by Lemuoklis, Lithuanian homophones can be classified into five groups according to their causes. All these groups are presented in detail below. 4.1. Theoretically possible homophones arising from the specifics of dictionary compilation When the morphological ambiguity of language began to be analyzed by computer, an important distinction became apparent: morphological ambiguity may be expected or unexpected. Expected ambiguity is that which people most commonly perceive. Unexpected ambiguity can be of two types: forms generated by a morphological analyzer and words that are rarely used (or no longer used at all) in contemporary language. Drawing on its morphological resources, Lemuoklis produces words that are not used in actual language, some of whose forms coincide with frequently used parts of speech or their forms in Lithuanian. This happens because dictionaries do not include all derivatives, even those formed regularly. In 2006, while working on a morphologically annotated corpus, E. Rimkutė compiled a list of unreal words generated by the morphological analyzer. This list was incorporated into the morphological analyzer as words to be ignored, so the problem of Lemuoklis generating words on its own has now been essentially solved. The lexical basis of Lemuoklis consists of words found in the Dictionary of Contemporary Lithuanian (as well as in other dictionaries), which includes many words now used especially rarely, e.g.: rata, girdas, metinys, ložėti, baisis, graudis, vajys, etc. If forms of these words coincide with forms of words commonly used in speech, Lemuoklis lists all the variants, which makes corpus processing considerably more difficult for linguists. The second problem, in which archaic words coincide with common ones, remains only partly unresolved. Such homophones are found mainly among syncretic forms of inflecting and uninflecting parts of speech; for example, in Lemuoklis’s analysis, the word forms bravo, ate, arba are treated as homophones, with these words also identified as nouns, since such information can also be found in dictionaries (online- MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 121 In the International Dictionary of Foreign Words (TŽŽ), the word bravas is defined as “in seventeenth- and eighteenth-century Italy, a hired assassin, an adventurer capable of using any kind of violence”; atas as “one hundredth of the Lao monetary unit, the kip”; or as “1. a covered wooden two-wheeled cart used by the peoples of Central Asia; used in Afghanistan to transport heavy loads; 2. a four-wheeled cart for transporting grain, used in the Caucasus and southern Ukraine.” It usually takes time to distinguish such words from unexpected homophones, because each of them has to be checked in various dictionaries and corpora. It is also difficult to decide whether they should be included in the LKHŽ, since it is no longer sufficient to rely solely on data from other dictionaries. This cause of homophone occurrence reveals a dilemma: the larger the lexicon of a language-analysis program, the more words and their forms the program should be able to recognize, but at the same time this means that unforeseen overlaps—cases of morphological ambiguity—may arise. Thus, to obtain the most accurate results possible, the principles for selecting words used in computer programs must be carefully considered. 4.2. Homophones arising from differences in prosodic features Homophones arising from differences in prosodic features (e.g., stress placement, pitch accent, vowel length) occur only in written language. Since Lithuanian morphological taggers analyze written word forms without stress marks, homophones often differ prosodically. Examples of such homophones are most numerous among overlaps between inflected parts of speech, and they occur in most of the smaller categories. Often, two or more forms of the same word overlap. There are 11 types of such examples in the dictionary: overlaps between nominative singular forms of nouns of the (i)o declension and their instrumental and vocative singular forms (e.g.: agentūra vs. agentūra, mama vs. mama); overlaps between genitive singular and nominative plural forms of nouns of the (i)o, ė declensions (e.g.: absoliučios vs. absoliučios; adekvačios vs. adekvačios; aikštės vs. aikštės); overlaps between instrumental and vocative singular forms of nouns of the ė declension10 (e.g.: katė vs. kâte, dumblė vs. dùmblė); overlaps between infinitives and nominative plural masculine forms of past passive participles (e.g.: 10 It should be noted that the vocative is possible only theoretically in many cases. When compiling the homophone dictionary, noun semantics were not taken into account; it was assumed that vocative forms are possible for all nouns. 122 BENDRINĖ KALBA | 96 apdìrbti vs. apdirbtì; apèiti vs. apeitì); coincidences between the genitive plural forms of the conditional mood and passive past tense participles (e.g.: aplẏtų vs. aplytų; dėngtų vs. dengtų); i)a declension nouns’ singular locative and vocative coincidences (e.g.: aidė vs. áidė; garažė vs. garaže); i)u and i declension nouns’ singular nominative and plural accusative coincidences (e.g.: apimtìs vs. ãpimtis; ãpskritis vs. apskritis); coincidences between the third- and second-person singular present-tense forms (e.g.: neprigirdì vs. neprigirdi; negãli vs. negali); coincidences between the feminine singular dative and neuter forms of pronouns (e.g., tái vs. taĩ, šìtai vs. šitaĩ; of course, šitai can also be a particle, but such coincidences in the dictionary fall into another category—that of coincidences between inflected and uninflected parts of speech—because, as mentioned, the dictionary provides only grammatical labels appropriate to a specific type); i)u declension nominals’ singular accusative and plural genitive11 coincidences (e.g.: ãlų vs. alų; lietų vs. lietų); coincidences between verb infinitives and second-person singular present-tense forms (e.g.: ìnkšti vs. inkšti; klýsti vs. klysti). Homophones caused by differences in prosodic features also arise when two or more different inflected parts of speech coincide. There are 7 types of such homophones in the dictionary: overlaps between nouns and verbs (e.g.: Adomo—the genitive singular form of the proper noun Adõmas12 coincides with the genitive singular form of the passive participle ãdomas; apsaugos—the genitive singular and accusative plural forms of the noun apsauga coincide with the future-tense form of the verb apsaugoti, cf. apsaugõs vs. apsáugos); overlaps between nouns and adjectives (e.g., atlaidus—the accusative plural form of the noun atlaidai coincides with the nominative singular form of the adjective atlaidus, cf. ãtlaidus vs. atlaidùs); overlaps between verbs and adjectives (e.g., ãntrinę, asmeninę vs. antrinę, asmeninę—the feminine singular accusative forms of the past active frequentative participles of the verbs antrinti, asmeninti coincide with the feminine singular accusative forms of the adjectives antrinis, -ė, asmeninis, -ė); overlaps between verbs and pronouns (e.g.: kitai vs. kitai; kuria vs. kurià—forms of the verbs kisti, kurti coincide with forms of the pronouns kitas, kuri); overlaps between nouns, adjectives, and verbs (e.g.: balti—the vocative singular form báltį of the noun baltis; the nominative plural form balti of the adjective baltas; the infinitive bálti); overlaps between nouns and numerals or pronouns (e.g.: Anos—the proper noun 11 Sometimes (as in the cases given below), the genitive plural is possible only theoretically. 12 It is true that the tagger can distinguish uppercase from lowercase letters. Nevertheless, morphological ambiguity could arise in this case if the participle were also written with an uppercase initial, for example, at the beginning of a sentence. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 123 Ana and the pronoun ana forms, cf. Ānos vs. anõs; kelias – the noun kẽlias and pronoun keliàs coincide); coincidences between nouns, verbs and pronouns, adjectives or numerals (e.g., dvejos – the numeral dvẽjos coincides with the singular genitive form dvejõs and plural nominative form dvẽjos of the noun dveja, as well as the future-tense verb dvejõs). Such coincidences have also been observed among homoforms of inflecting and uninflecting parts of speech; the dictionary contains 8 types: coincidences between verbs, adjectives and adverbs (e.g., blogai – the singular second-person past simple form blógai of the verb blogti coincides with the feminine singular dative form blõgai of the adjective blogas, -a and the adverb blogai); coincidences between adjectives and adverbs (e.g., aklai, efektyviai – the feminine dative forms ãklai, efektỹviai of the adjectives coincide with the adverbs aklai, efektyviai); coincidences between nouns and adverbs (e.g., seniai – the plural nominative form sẽniai of the noun senis coincides with the adverb seniai); coincidences between nouns, adjectives or pronouns and adverbs (e.g., aiškiai, aukštai – the plural nominative forms aiškiai, aukštai of the nouns aiškis, aukštas coincide with the feminine dative forms áiškiai, áukštai of the adjectives aiškus, -i, aukštas, -a and the adverbs aiškiai, aukštai); coincidences between nouns, verbs, adjectives and adverbs (e.g., bukãi – the plural nominative and singular vocative forms bùkai of the noun bukאס coincide with the singular second-person past simple form bukãi of the verb bukti, the feminine singular dative form bùkai of the adjective bukאס, and the adverb bukãi); coincidences between numerals, prepositions and adverbs (e.g., pirma – the feminine singular nominative and instrumental forms and the neuter forms pirmà, pìrma of the numeral pirmas, -a coincide with the preposition and adverb pìrma); coincidences between verbs and adverbs (e.g., apmaudžiau – the singular first-person past simple form apmáudžiau of the verb apmaudyti coincides with the comparative adverb apmaudžiau), coincidences between nouns, pronouns and adverbs or particles (e.g., Ana – the singular nominative, instrumental and vocative forms of the proper noun Ana coincide with the forms ana of the pronoun and particle, cf. Anà, Āna and anà). If Lemuoklis and other morphological taggers were able to analyze texts with marked stress13, morphological ambiguity would likely be considerably reduced. 13 This can be done automatically without difficulty, for example, using the Kirčiuoklis program, which is available at https://kalbu.vdu.lt/mokymosi-priemones/kirciuoklis. On the other hand, this program relies on the results of an automatic morphological analysis program, creating a vicious circle: the quality of the stress-marking program depends on how well the morphological analyzer works. It should be noted that Kirčiuoklis provides several stress variants for morphologically ambiguous forms, so it would not be particularly helpful in solving the problem of morphological ambiguity. 124 BENDRINĖ KALBA | 96 There have also previously been attempts to classify homoforms according to prosodic features. E. Rimkutė (2002: 91–98) analyzed 10,000 grammatically annotated words used in articles in Lietuvos ryto. It was found that 83 percent of homophones are phonetically identical, while 17 percent differ in stress placement. The researcher further divided homoforms that do not differ in stress placement into two types: those differing in vowel quantity (e.g., the verb mãno and the pronoun màno) and those differing in intonation (e.g., the verbs kláusė (lemma – klausyti) and kláusė (lemma – klausti)). The examples given earlier also discussed homoforms that differ in stress placement and intonation. The latter category also included cases differing in vowel quantity, although they were not distinguished separately. 4.3. Homoforms arising from the coincidence of different words This category includes some homoforms of inflecting parts of speech, as well as homoforms of inflecting and uninflecting parts of speech. In theory, all cases in the category of homoforms of inflecting and uninflecting parts of speech are also words with different meanings, since they have different lemmas. However, this is not always true, because, for example, there is considerable semantic overlap among coincidences between adverbs and adjectives, and one form is often derived from the other. Therefore, homoforms arising from polysemous words could be further divided into two smaller categories: homoforms with different lexical meanings and homoforms with partially shared lexical meaning. 4.3.1. Homoforms with different lexical meanings This category includes coincident forms that are not derivationally related. These include 8 types of homoforms between inflecting and uninflecting parts of speech: verb, adjective and adverb homoforms, for example, griežtai – in this case, the feminine singular dative form of the participle griežtas, -a (lemma – griežti) coincides with the adverb griežtai, which has a completely different meaning, and the feminine singular dative form of the adjective griežtas, -a. Thus, in this case the participle is linked to the adverb and adjective by different lexical meanings, whereas the adjective and adverb are linked by partially overlapping meaning, since they are derived from one another (see section 4.3.2 for details); noun homoforms with adverbs or with prepositions, conjunctions, particles, interjections and ideophones (e.g., dėlei – the singular dative form of the noun dėlė coincides with the form of the preposition dėlei; opa – the singular nominative, vocative and instrumental forms of the noun opa coincide with an ideophone); homoforms of pronouns with adverbs or particles and conjunctions MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 125 (e.g., kam—the homophony of an adverb and a pronoun); homophones of nouns, adverbs, and adjectives or pronouns (e.g., the feminine noun kandis, meaning “a small moth whose caterpillars cut clothes and eat grain, flour, and plants” (LKŽ 2005), whose singular dative form kandžiai coincides with the plural nominative form kandžiai of the masculine noun kandus, -i, meaning “biting, a bite, a bitten spot” (LKŽ 2005); this form also coincides with the feminine singular dative of the adjective kandžiai and the adverb kandis (the adjective and adverb are partly coincident in lexical meaning, whereas nouns arti are related to them by a different lexical meaning); homophones of nouns, verbs, adjectives, adverbs, and prepositions (e.g., artis coincides in form with the singular vocative of the noun árti, the infinitive of the verb artus, -i, the feminine singular nominative of the adjective, and the adverb and preposition arti); homophones of verbs, nouns, and adverbs (e.g., kartais—the plural instrumental forms of the participle kártas and the noun kārtas coincide with the adverbial form kartais); homophones of verbs and prepositions or particles, conjunctions, interjections, onomatopoeic words, and abbreviations (e.g., ak—an interjection coincides with the third-person singular imperative form of the noun akti; užvis—the third-person future form of the verb užvisti coincides with the adverb užvis; vysk—the graphic abbreviation of the word vyskupas coincides with the third-person singular imperative form of the verb vysti); homophones of nouns, pronouns, and adverbs or particles (e.g., ana—the forms of the proper noun Ana, the pronoun, and the particle ana coincide). Homophones with different lexical meanings may also occur among homophones belonging to inflected parts of speech. In LKHŽ they comprise 8 types: homophones of nouns, verbs, and adjectives with different lexical meanings (e.g., anglių—the plural genitive forms of the nouns anglys and anglė coincide); homophones of nouns and verbs (e.g., adomą—the singular accusative form of the participle adomas coincides with the singular accusative form of the proper noun Adomas); homophones of verbs and pronouns or numerals (e.g., jos—the third-person future form of the verb joti coincides with the plural nominative and singular genitive forms of the pronoun ji; kitai—the second-person past simple form of the verb kisti coincides with the singular dative form of the pronoun kita; trejos—the third-person future form of the verb trejoti coincides with the nominative form of the numeral treji, -os); homophones of nouns, adjectives, and verbs (e.g., kaltų—the third-person subjunctive form of the verb kalti, the plural genitive form of the feminine and masculine participle káltas, -a, the plural genitive form of the feminine and masculine adjective káltas, -a, and 126 BENDRINĖ KALBA | 96 the plural genitive form of the noun káltas; kartus—the masculine plural accusative form of the participle kártas, the singular nominative form of the adjective kartus, and the plural accusative form of the noun kartas coincide); homophones of adjectives, pronouns, numerals, and nouns (e.g., viena—the proper noun Viena coincides with the singular nominative, instrumental, and vocative forms of the numeral, pronoun, and adjective vienas, -a); homophones of nouns, verbs, and pronouns or numerals (e.g., dvejos—the singular genitive and plural nominative forms of the noun dveja, the third-person future form of the verb dvejoti, and the forms of the numeral dvejos coincide; maną—the plural nominative form of the active present participle derived from the verb manyti coincides with the feminine and masculine singular accusative form of the pronoun manà and the singular accusative form of the noun mãna); homophones of adjectives and pronouns (e.g., juodu—the dual nominative and accusative form of the pronoun jis coincides with the singular genitive form of the adjective juodas). 4.3.2. Homoforms with partially overlapping lexical meaning Such homophones have different lemmas, but are semantically and derivationally related, e.g.: absoliutiškai—the forms of the adverb absoliutiškai and the adjective absoliutiškas coincide; ilgai—the forms of the adverb ilgai and the adjective ilgas coincide. These are most often adverbs derived from adjectives, i.e., words with the same root, and can be regarded as words with different meanings only in part (when viewed semantically rather than in formal morphological terms). Partial lexical homophones in the dictionary comprise 8 types of homophones belonging to inflected parts of speech, specifically: substantiva mobilia nouns, for example, administratorių (the plural genitive forms of the words administratorius and administratorė coincide), bėglių (the plural genitive forms of the words bėglys and bėglė coincide); homophones of nouns and verbs (e.g., apkaustai—the second-person present form of the verb apkaustyti coincides with the nominative form of the noun apkaustai; atsakai—the plural nominative form of the noun atsakas coincides with the second-person present form of the verb atsakyti); homophones of nouns and adjectives (e.g., absoliutus—the plural nominative form of the noun absoliutas coincides with the singular nominative form of the adjective absoliutus; abstraktu—the singular instrumental form of the noun abstraktas coincides with the neuter form of the adjective abstraktus, -i); homophones of verbs and adjectives (e.g., magnetinę—the past simple plural nominative form of the participle magnetinės coincides with the feminine singular accusative form of the adjective magnetinis, -ė; masinę—the participle masinės MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 127 plural accusative and adjective masinis, -i feminine singular accusative forms); noun, adjective and verb homophones (e.g., kaltą – the masculine and feminine singular accusative forms of the adjective káltas and participle káltas coincide with the singular accusative form of the noun káltas; in the latter case, however, two distinct causes are intertwined. There is indeed partial lexical overlap between the participle and the noun, since their meanings are related, whereas the participle and noun káltas coincide with the adjective káltas for prosodic reasons); adjective and pronoun homophones (e.g., pirmiausia – the feminine singular nominative, instrumental and vocative forms of the neuter-gender adjective pirmiausias, -ia coincide with the form of the numeral pirmas); noun and numeral or pronoun homophones (e.g., aštuonetą (noun and numeral coincide), maniškė (noun and pronoun coincide)); noun, verb and pronoun or adjective, numeral homophones (e.g., niūri – the singular vocative form of the noun niūris, the feminine singular nominative form of the adjective niūrus, -i, and the second- and third-person present-tense forms of the verbs niūrėti and niurti coincide). Examples of partial lexical polysemy can also be found among overlaps between inflected and uninflected parts of speech—in homophones of verbs, adjectives and adverbs (e.g., aklinai – the second-person singular past simple form of the verb aklin- ti, the adverb aklinai, and the feminine singular dative form of the adjective aklinas, -a coincide), adjective and adverb homophones (abejingai, aistringai, cikliškai); noun and adverb homophones (e.g., dovanai, laukan, pusbalsiu); noun, adverb and adjective or pronoun homophones (e.g., bailiai – the plural nominative form of the noun bailys, the feminine singular dative form of the adjective bailus, -i, and the adverb bailiai coincide, although the stress differs in the noun bailiai and the adjective and adverb bailiai. If stress were disregarded, all three forms would be linked by partially overlapping lexical meanings. Also worth mentioning are noun, verb and adverb or preposition homophones (e.g., uždarai – the forms of the participle uždaras, -a derived from the verb uždaryti (singular dative), the plural nominative of the noun už-daras, the feminine singular dative of the adjective uždaras, -a, and the adverb uždarai coincide; again, some words differ in stress, cf. uždarai ir uždarai), verb, noun and adverb homophones (e.g., apsiaustai – the plural nominative form of the noun apsiaustas, the feminine singular dative form of the participle apsiaustas, -a, and the adverb apsiaustai coincide); numeral and adverb homophones (e.g., keturiomis, pirmiau), 128 BENDRINĖ KALBA | 96 verb and adverb homophones (e.g., apibrėžtai, apibendrintai – the feminine singular dative forms of the participles apibrėžtas, –a, apibendrintas, –a coincide with the adverbs apibrėžtai, apibendrintai), noun, pronoun and adverb homophones (e.g., nieko – the forms of the noun and pronoun niekas coincide with the adverb nieko), adjective, numeral and adverb homophones (e.g., pirmiausiai – depending on context, this word may be an adjective, an adverb or a numeral). 4.4. Homophones determined by the morphological system of Lithuanian Some homophones are determined by the morphological system of Lithuanian. Nominal overlaps result from syncretic case forms in certain declensions and from forms of different genders or numbers coinciding in some cases (e.g., naujų – the plural genitive form for both feminine and masculine); verb person, number, tense, etc., may also coincide. The lexical meanings of such words are identical. This category as a whole comprises only overlaps among inflected parts of speech. Nominal overlaps are the most numerous and can be divided into number, gender and case syncretism. Case-form overlaps are of 7 types: overlaps of the singular nominative, instrumental and vocative forms and the neuter gender in nouns of the (i)o declension (e.g., nouns abatija, data, emulsija; adjectives faktiška, fatališka, galinga); overlaps of the singular instrumental and vocative forms of nouns in the ė declension (e.g., nouns beteise, dėde, bedante); overlaps of the plural nominative and vocative forms of nouns in the (i)a, (i)u declensions (e.g., nouns abituriantai, akiniai, angelai); overlaps of the singular locative and vocative forms of nouns in the (i)a declension (e.g., albume, bate, čempione); overlaps of the nominative and accusative forms of numerals and pronouns (e.g., pronouns judu, mudu, numerals abi, dvi (it is worth noting that in the Bendrinės lietuvių kalbos žodyne (BLKŽ), the word abi is considered a pronoun); overlaps of the plural nominative and vocative forms of nouns or substantivized participles and adjectives in the i declension (e.g., akmenys, anty s, baltieji, kaltinamieji); overlaps of the nominative, accusative and instrumental forms of numerals or pronouns (e.g., numerals aštuoniolika, dvylika, trylika and the pronoun keliolika). Gender overlaps in the dictionary are of 7 types: overlaps of the masculine and feminine plural genitive forms of nominals (e.g., participles abejojančių, akcentuojančių, apibendrinančių, adjectives apynau jų, birių, blogų – masculine and feminine forms coincide); (i)a MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 129 syncretism between masculine singular nominative forms of declension-pattern nominal words and (i)o-declension feminine plural accusative forms (e.g., adjectives abejingas, brokuotas, dieviškas; participles diferencijuotas, forsuotas, gautas); syncretism between singular accusative forms of (i)o and (i)a-declension nominal words (e.g., adjectives abejingą, aklą, darganą; participles dažomą, draudžiamą, gaunamą); syncretism between masculine plural nominative forms of active participles and neuter forms (e.g., abejoja, amžina, apdarą); syncretism between feminine and masculine nominative forms of numerals (e.g., trys); syncretism between feminine and masculine accusative forms of numerals (e.g., tris); syncretism between masculine and feminine dative and/or instrumental forms of numerals or pronouns (e.g., duiem, judviem); and syncretism between feminine and masculine nouns and the common gender (e.g., giminė, mazgotė—the common gender coincides with the feminine; ligonis—the common gender coincides with the masculine); As noted, the number of inflecting parts of speech may also coincide. There are four types of such cases in the dictionary: syncretism between singular genitive and plural nominative and vocative forms of (i)o, ė-declension nominal words (e.g., nouns abejonės, abėcėlės, baterijos; adjectives basos, birios, dabartinės; participles decentralizuotos, degeneruotos, eiliuotos); syncretism between singular nominative and plural accusative forms of (i)u and i-declension nominal words (e.g., administratorius, gentis, kalendorius); syncretism between singular accusative and plural genitive forms of (i)u-declension nouns (e.g., agresorių, derlių, išminčių); syncretism between singular vocative and plural nominative forms of masculine adjectives of (i) declension (e.g., artimesni, garesni, mažesni). Verb forms also often coincide. There are 7 types of such coincidences in the dictionary. These are: homoforms of infinitives and of the nominative plural masculine forms of the past passive voice participles (e.g., to absorb, to dismantle, to varnish); homoforms of third-person conditional mood verbs and genitive plural forms of past passive voice participles (e.g., adapted, manufactured, heard); homoforms of the third and second persons of present-tense verbs (e.g., is reflected, remembers, flinches); coincidences between participles and gerunds or half-participles (e.g., adoring, fearing—in this case, reflexive gerunds coincide with the non-reflexive nominative singular masculine form of the present-tense participle; the half-participle being givencoincides with present-tense participles—both with the accusative plural feminine form and the nominative singular masculine form); coincidences between infinitives and the second-person singular forms of verbs (e.g., not to err, to slip, to gush); coincidences between finite and non-finite verb forms (e.g., artificially, inclusively, imploringly—the dative singular forms of feminine participles coincide with adverbial forms; returns, investigated—verb- 130 BENDRINĖ KALBA | 96 apklau­siant, pratęsiant, trešiant—the future- and present-tense converb forms coincide. 4.5. Same-meaning homoforms arising from assignment to different paradigms Some nouns, adjectives, verbs, pronouns, and adverbs with the same meaning may belong to two different paradigms, e.g., Audrių could be the singular accusative of either Audrius or Audrys; būčiau could belong to the paradigm būti, yra, buvo, to būti, būna, buvo, or to būti, būva, buvo; grįuvus could belong to either the grįuti, grįūna, grįuvo or the grįūti, grįūva, grįuvo paradigm; the lemma of the form stebukladario could be either stebukladaris or stebukladarys; mano could be a form of either the pronoun with lemma mano or the pronoun with lemma manas. The LKHŽ assigns one separate type to such cases. 4.6. Relationships between causes The five causes of occurrence mentioned above may be interrelated; moreover, the occurrence of homoforms in some words may be determined by more than one cause. For example, abstrãkčiai vs. abstrakčiai—here, ambiguity is caused both by differences in prosodic elements (the stress position differs) and by partial overlap in meaning (one word is derived from the other). Another example is the coincidence between the adjective káltas and the noun káltas. These words could fall into both the categories of homoforms determined by prosodic elements and those determined by different lexical meanings. Moreover, different homoforms belonging to a single type in the dictionary are not necessarily caused by the same factors. For example, some coincidences between conditional forms and passive participles (e.g., adaptuotų, brandintų) arise from differences in prosodic elements (cf. aplýtų and aplytų), while others arise from the morphological system of Lithuanian, that is, they are simply two coinciding verb forms (e.g., automatizuotų, derėtų). The five causes of homoform formation help us better understand why cases of morphological ambiguity arise when annotating with Lemuokliu (or other morphological annotators). By supplementing M I G L Ė Ž E M R I E T Ė. Morfologiškai daugiareikšmių formų atsiradimo priežastys 131 This information could be supplemented with statistical data to identify the most likely cause. This would also make it possible to improve LKHŽ. Understanding what causes morphological ambiguity would make it possible to add new data to the dictionary; for example, labels could be introduced to indicate the likelihood of ambiguity, among other things. Furthermore, more detailed analysis accompanied by statistical data could reveal additional causes of the emergence of homophones. This is a possible topic for future research. SUMMARY The article describes one stage in the computerization of the Lithuanian language—automatic morphological analysis. The problem of morphological ambiguity becomes apparent in this process. The LKHŽ currently being compiled has made it possible to identify the reasons why homophones occur in Lithuanian. According to their causes, homophones can be classified as follows: 1) theoretical homophones arising from the specific features of the compilation of certain dictionaries; 2) homophones arising from differences in prosodic features; 3) homophones arising from the overlap of different words; 4) homophones determined by the morphological system of Lithuanian; 5) homophones with the same meaning arising from assignment to different paradigms. Theoretical homophones complicate morphological analysis when archaic or (almost) obsolete words in contemporary Lithuanian coincide with common words. This problem shows that morphological taggers require differently compiled word lists. Homophones determined by prosodic features occur only in written language. Most examples are found among overlaps between inflected parts of speech, often involving forms of the same word (e.g., sg. nominative mamà and vocative māmą; sg. nominative apimtis and pl. accusative āpimtis), although two different parts of speech can also coincide (e.g., apsaugos—the sg. genitive and pl. nominative forms of the noun apsauga coincide with the third-person future form of the verb apsaugoti). Such overlaps have also been observed between inflected and uninflected parts of speech, for example, overlaps between verbs, adjectives, and adverbs (e.g., blogai—the second-person singular simple past form of the verb blogti, the feminine singular dative form of the adjective blogas, –a, and the adverb blogai coincide). 132 BENDRINĖ KALBA | 96 Homophones arising from the overlap of different words may have different lexical meanings (overlapping words that are not derivationally related, e.g., arti—the adverb, preposition, infinitive, and participle coincide; cf. artì and ártì) or partially overlapping meanings (they have different lemmas but are semantically and derivationally related, e.g., absoliutiškai—the adverb absoliutiškai and the feminine singular dative form of the adjective absoliutiškas, -a coincide). Homophones arising from the morphological system of Lithuanian are determined by Lithuanian grammar, for example, syncretic cases and forms shared by different genders or numbers (e.g., naujų—a form shared by both feminine and masculine); the person, number, tense, etc. of verbs may also coincide. Such words have the same lexical meaning; in other words, different inflected forms of the same word coincide. This category consists entirely of overlaps among inflected parts of speech. Homophones may also arise because the same word is assigned to different paradigms, for example, būčiau—this form could belong to the paradigms būti, yra, buvo, būti, būna, buvo, or būti, būva, buvo. These causes help explain why various cases of morphological ambiguity arise during automatic annotation. In addition, this information could be used to improve LKHŽ. Understanding what causes morphological ambiguity would make it possible to add further information to the dictionary; for example, labels could be introduced to indicate the likelihood of ambiguity or other data. LITERATŪRA Aleksandravičiūtė Skaistė 2008: Paieškos sistemos „Google“ naudojimo galimybės tiriant kalbos reiškinius. – Bendrinė kalba 81, 266–282. Barauskaitė Janina 1979: Leksikologijos pradmenys. Leksinė semantika, Vilnius: Lietuvos TSR aukštojo ir specialiojo vidurinio mokslo ministerija. Barauskaitė Jūratė, Čepaitienė Giedrė, Mikulėnienė Danguolė 1995: Lietuvių kalba. Leksikologija, fonetika, akcentologija, dialektologija, rašyba 1, Vilnius, Mokslo ir enciklopedijų leidykla. Bielinskienė Agnė, Boizou Loïc, Rimkutė Erika 2017: Lietuvių kalbos morfologiškai ir sintaksiškai anotuoti tekstynai. – Bendrinė kalba 90, 1–30. MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 133 BLKŽ 2022: Bendrinės lietuvių kalbos žodynas (rengiamas), vyr. red. D. Liutkevičienė, Vilnius: Lietuvių kalbos institutas. Prieiga internete: https://ekalba.lki.lt/bendrines-lietuviu-kalbos-zodynas. DLKŽ 2021: Dabartinės lietuvių kalbos žodynas, vyriausiasis redaktorius S. Keinys, 8-as patais. ir papild. leidimas (elektroninis variantas), Vilnius: Lietuvių kalbos institutas. Prieiga internete: https://ekalba.lt. Gaivenis Kazimieras, Keinys Stasys 1990: Kalbotyros terminų žodynas, Kaunas: Šviesa. Grumadienė Laima 2002: Dabartinės rašomosios lietuvių kalbos dažninis žodynas ir jo bazė. – Lituanistika 46, 19–37. Jakaitienė Evalda 2009: Leksikologija, Vilnius: Vilniaus universiteto leidykla. Kaljanov Ali 2023: Grammatical Homonyms in Modern Karakalpak Language. – The International Journal of Language and Cultural 5(1), 26–31. Prieiga internete: https://www.growingscholar.org/journal/index.php/TIJOLAC/article/view/322. Katamba Francis 1994: English Words, London: Routledge. LKE 1999: Lietuvių kalbos enciklopedija, Vilnius: Mokslo ir enciklopedijų leidybos institutas. LKŽ 2005: Lietuvių kalbos žodynas, t. 1–20 (elektroninis variantas), vyr. red. G. Naktinienė, Vilnius: Lietuvių kalbos institutas, 2005. Prieiga internete: https://ekalba.lt/lietuviu-kalbos-zodynas. Marcinkevičienė Rūta 2011: Žodžio reikšmė. Žodynai ir tekstynai, Kaunas: Vytauto Didžiojo universitetas. Nevzorova Olga, Zin’kina Julia, Pjatkin Nicolaj 2005: Applied Problems of Functional Homonymy Resolution for Russian Language. – International Journal ITA 12(3), 231–235. Prieiga internete: http://www.foibg.com/ijita/vol12/ijita12-3-p05.pdf. Pakerys Jurgis 2009: Būtiniųjų predikatyvų žymėjimas lietuvių kalbos veiksmo pavadinimo konstrukcijose. – Gramatinių funkcijų prigimtis ir raiška (Acta Salensia 1), Vilnius: Vilniaus universitetas, 125–145. Piečytė Miglė 2023: Morfologinio daugiareikšmiškumo analizė ir klasifikacija: baigiamasis magistro darbas, Kaunas: Vytauto Didžiojo universitetas. Puolakainen Tiina 2012: How Does the Choise of Morphological Analyser Influence the Quality of Syntactical Analysis? – Human Language Technologies – The Baltic Perspective: Proceedings of the Fifth International Conference Baltic HLT, Amsterdam, Berlin, Tokyo, Washington, DC: IOS Press, 193–200. Rimkutė Erika 2002: Homoformos dabartinės lietuvių kalbos tekstyne. – Lituanistika 2, 86–101. 134 BENDRINĖ KALBA | 96 Rimkutė Erika 2003a: Morfologinio daugiareikšmiškumo tipologija. – Lituanistika 4, 60–78. Rimkutė Erika 2003b: Homonimai, homoformos ir homografai tekstynų lingvistikos požiūriu. – Terminologija 10, 33–41. Rimkutė Erika 2006a: Morfologinio daugiareikšmiškumo ribojimas kompiuteriniame tekstyne: daktaro disertacija, Kaunas: Vytauto Didžiojo universitetas. Rimkutė Erika 2006b: Dabartinės lietuvių kalbos gramatinių formų vartosena morfologiškai anotuotame tekstyne. – Lituanistika 2, 34–55. Rimkutė Erika, Grybinaitė Rasa 2004: Dažniausios lietuvių kalbos morfologinio daugiareikšmiškumo rūšys ir jų automatinis vienareikšminimas. – Kalbų studijos 5, 74–78. Prieiga internete: https://www.vdu.lt/cris/handle/20.500.12259/53446. Rimkutė Erika, Grigonytė Gintarė 2006: Automatizuotas lietuvių kalbos morfologinio daugiareikšmiškumo ribojimas. – Lituanistika 9, 30–37. Rimkutė Erika, Daudaravičius Vidas 2007: Morfologinis „Dabartinės lietuvių kalbos tekstyno“ anotavimas. – Kalbų studijos 11, 30–35. Šukys Jonas 2004: Panašiai skambantys žodžiai: normos ir klaidos, Vilnius: Gimtasis žodis. TŽŽ 2023: Tarptautinių žodžių žodynas (elektroninis variantas). Prieiga internete: https://tzz.lt/. Zinkevičius Vytautas 2000: Lemuoklis – morfologinei analizei. – Darbai ir dienos 24, 246–273. Įteikta 2023 04 05 Priimta 2023 06 29 MIGLĖ ŽEMRIETĖ THE REASONS FOR THE MORPHOLOGICALLY AMBIGUOUS FORMS Summary Although the problem of morphological ambiguity in the Lithuanian language has been solved for all practical purposes morphological annotators, even when are improved, are not always capable of choosing the most probable form from several morphologically polysemous forms. Moreover, such cases are not uncommon when the morphological annotator incorrectly chooses one form from several possible ones. Therefore, a decision was made to conduct a study that would analyze and classify cases of morphological polysemy, hereinafter referred to as homoforms of the Lithuanian language. The aim of the study is to discuss their MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 135 possible causes based on the classified homoforms through the application of quantitative analysis, textual linguistics, and descriptive. When describing the reasons for homoforms, reference is made to the Dictionary of Homoforms of the Lithuanian language, prepared by the author of this article and Erika Rimkutė. Since the Dictionary of Homoforms of the Lithuanian Language consists of a really large number (about 36,000) of morphologically polysemous words, it becomes possible to define the reasons for the appearance of homoforms in the Lithuanian language. Homoforms can be classified according to the reasons for their appearance as follows: (1) homoforms that are rather theoretical in nature and appeared in the process of the compilation of dictionaries; (2) homoforms resulting from differences in prosodic elements; (3) homoforms resulting from different overlapping words; (4) homoforms that came into existence due to the specifics of the Lithuanian morphology system; (5) homoforms of the same meaning that appeared under different paradigms. These five reasons help us to better understand why automatic annotation often produces a kind of morphological ambiguity. Based on this information, it would be possible to improve the Dictionary of Homoforms of the Lithuanian Language on its own right. By understanding what determines morphological polysemy, new data could be added to the dictionary, such as notations that indicate the probability of ambiguity or other data. KEYWORDS: morphological ambiguity, automatic morphological analysis, corpus, homoforms, dictionary. MIGLĖ ŽEMRIETĖ Lietuvių kalbos institutas Petro Vileišio g. 5, LT-10308 Vilnius [email protected] MIGLĖ ŽEMRIETĖ MORFOLOGIŠKAI DAUGIAREIKŠMIŲ FORMŲ ATSIRADIMO PRIEŽASTYS Santrauka Nors morfologinis daugiareikšmiškumas yra iš esmės išspręsta problema, net ir patobulinus lietuvių kalbos morfologinius anotatorius, jiems ne visada iš kelių morfologiškai daugiareikšmių formų pavyksta pasirinkti tikėtiniausią. Be to, nereti ir tokie atvejai, kai morfologinis anotatorius neteisingai parenka vieną formą 136 BENDRINĖ KALBA | 96 iš kelių galimų. Todėl nutarta išanalizuoti ir suklasifikuoti morfologinio daugiareikšmiškumo atvejus, juos vadinant lietuvių kalbos homoformomis. Tyrimo tikslas – remiantis suklasifikuotomis homoformomis, aptarti galimas jų priežastis. Taikyti kiekybinės analizės, tekstynų lingvistikos, aprašomasis metodai. Aprašant homoformų priežastis, remiamasi Lietuvių kalbos homoformų žodynu. Jį sudaro apie 36 tūkst. homoformų, todėl galima apibrėžti homoformų lietuvių kalboje atsiradimo priežastis. Šiame straipsnyje homoformos pagal atsiradimo priežastis suklasifikuotos į penkias grupes: 1) teorinės homoformos, atsiradusios dėl kai kurių žodynų sudarymo specifikos; 2) homoformos, atsiradusios dėl prozodinių elementų skirtumų; 3) homoformos, atsiradusios dėl sutampančių skirtingų žodžių; 4) homoformos, nulemtos lietuvių kalbos morfologijos sistemos; 5) tos pačios reikšmės homoformos, atsiradusios dėl priskyrimo skirtingoms paradigmoms. Ši klasifikacija padeda geriau suvokti, kodėl morfologiniams anotatoriams anotuojant atsiranda vienokių ar kitokių morfologinio daugiareikšmiškumo atvejų. Remiantis šia informacija, būtų galima patobulinti ir patį Lietuvių kalbos homoformų žodyną. Suvokiant, kas lemia morfologinį daugiareikšmiškumą, į žodyną būtų galima įtraukti naujų duomenų, pavyzdžiui, galėtų atsirasti pažymos, kurios rodytų daugiareikšmiškumo tikimybę ar kitus duomenis. ESMINIAI ŽODŽIAI: morfologinis daugiareikšmiškumas, automatinė morfologinė analizė, tekstynas, homoformos, žodynas. MIGLĖ ŽEMRIETĖ Lietuvių kalbos institutas Petro Vileišio g. 5, LT-10308 Vilnius [email protected] MIGLĖ ŽEMRIETĖ. Morfologiškai daugiareikšmių formų atsiradimo priežastys 137