Browsing by master's degree program "Master's Programme Linguistic Diversity in the Digital Age"

Hypernymy is a relationship between two words, where the hyponym carries a more specific meaning, and entails a hypernym that carries a more general meaning. A particular kind of verbal hypernymy is troponymy, where troponyms are verbs that encode a particular manner or way of doing something, such as “whisper” meaning “to speak in a quiet manner”. Recently, contextualized word vectors have emerged as a powerful tool for representing the semantics of words in a given context, in contrast to earlier static embeddings where every word is represented by a single vector regardless of sense. BERT, a pre-trained language model that uses contextualized word representations, achieved state of the art performance on various downstream NLP tasks such as question answering. Previous research identified knowledge of scalar adjective intensity in BERT, but not systematic knowledge of nominal hypernymy. In this thesis, we investigate systematic knowledge of troponymy and verbal hypernymy in the base English version of BERT. We compare the similarity of vector representations for manner verbs and adverbs of interest, to see if troponymy is represented in the vector space. Then, we evaluate BERT’s predictions for cloze tasks involving troponymy and verbal hypernymy. We also attempt to train supervised models to probe vector representations for this knowledge. Lastly, we perform clustering analyses on vector representations of words in hypernymy pairs. Data on troponymy and hypernymy relationships is extracted from WordNet and HyperLex, and sentences containing instances of the relevant words are obtained from the ukWaC corpus. We were unable to identify any systematic knowledge about troponymy and verb hypernymy in BERT. It was reasonably successful at predicting hypernyms in the masking experiments, but a general inability to go in the other direction suggests that this knowledge is not systematic. Our probing models were unsuccessful at recovering information related to hypernymy and troponymy from the representations. In contrast with previous work that finds type-level semantic information to be located in the lower layers of BERT, our cluster-based analyses suggest that the upper layers contain stronger or more accessible representations of hypernymy.

Computer-Assisted Language Learning (CALL) is one of the sub-disciplines within the area of Second Language Acquisition. Clozes, also called fill-in-the-blank, are largely used exercises in language learning applications. A cloze is an exercise where the learner is asked to provide a fragment that has been removed from the text. For language learning purposes, in addition to open-end clozes where one or more words are removed and the student must fill the gap, another type of cloze is commonly used, namely multiple-choice cloze. In a multiple-choice cloze, a fragment is removed from the text and the student must choose the correct answer from multiple options. Multiple-choice exercises are a common way of practicing and testing grammatical knowledge. The aim of this work is to identify relevant learning constructs for Italian to be applied to automatic exercises creation based on authentic texts in the Revita Framework. Learning constructs are units that represent language knowledge. Revita is a free to use online platform that was designed to provide language learning tools with the aim of revitalizing endangered languages including several Finno-Ugric languages such as North Saami. Later non-endangered languages were added. Italian is the first majority language to be added in a principled way. This work paves the way towards adding new languages in the future. Its purpose is threefold: it contributes to the raising of Italian from its beta status towards a full development stage; it formulates best practices for defining support for a new language and it serves as a documentation of what has been done, how and what remains to be done. Grammars and linguistic resources were consulted to compile an inventory of learning constructs for Italian. Analytic and pronominal verbs, verb government with prepositions, and noun phrase agreement were implemented by designing pattern rules that match sequences of tokens with specific parts-of-speech, surfaces and morphological tags. The rules were tested with test sentences that allowed further refining and correction of the rules. Current precision of the 47 rules for analytic and pronominal verbs on 177 test sentences results in 100%. Recall is 96.4%. Both precision and recall for the 5 noun phrase agreement rules result in 96.0% in respect to the 34 test sentences. Analytic and pronominal verb, as well as noun phrase agreement patterns, were used to generate open-end clozes. Verb government pattern rules were implemented into multiple-choice exercises where one of the four presented options is the correct preposition and the other three are prepositions that do not fit in context. The patterns were designed based on colligations, combinations of tokens (collocations) that are also explained by grammatical constraints. Verb government exercises were generated on a specifically collected corpus of 29074 words. The corpus included three types of text: biography sections from Wikipedia, Italian news articles and Italian language matriculation exams. The last text type generated the most exercises with a rate of 19 exercises every 10000 words, suggesting that the semi-authentic text met best the level of verb government exercises because of appropriate vocabulary frequency and sentence structure complexity. Four native language experts, either teachers of Italian as L2 or linguists, evaluated usability of the generated multiple-choice clozes, which resulted in 93.55%. This result suggests that minor adjustments i.e., the exclusion of target verbs that cause multiple-admissibility, are sufficient to consider verb government patterns usable until the possibility of dealing with multiple-admissible answers is addressed. The implementation of some of the most important learning constructs for Italian resulted feasible with current NLP tools, although quantitative evaluation of precision and recall of the designed rules is needed to evaluate the generation of exercises on authentic text. This work paves the way towards a full development stage of Italian in Revita and enables further pilot studies with actual learners, which will allow to measure learning outcomes in quantitative terms

It is common for speech to occur in closed spaces. Hence, room acoustics have a significant role in speech communication. In previous studies, effects of reverberation on speech production have been found. However, research on the concerned field is yet scarce. Adverse room acoustics have been observed to expose occupational speakers, such as teachers, to voice disorders. Thus, it is crucial to study what are the room acoustic requirements for economic speaking. The purpose of this study is to examine which speech-acoustic traits change when the speaker is exposed to reverberation, and how. In the present study, two different approaches are taken: variation of reverberation time and removal of the reverberation. The changes in speech are reflected to the Lombard sign (the raise of speech level in a noisy environment). Additionally, differences related to gender and prosody are examined concerning the present topic. In this study, a speech production experiment was conducted with acoustic and statistical analyses. 11 Finnish-speaking volunteers (six females and five males) participated the experiment, where 150 short sentences were recorded from each participant. The sentences were produced in five different room-acoustic conditions. In four out of five, digitally simulated reverberation was played back on headphones worn by the participant with varying reverberation times. The fifth condition was (nearly) anechoic. Out of the recorded sentences, speech rate, creak ratio and harmonics-to-noise ratio were measured along with mean, maximum and movement of intensity and pitch. The measurements were then assessed with various statistical methods. The results of the study show a significant decrease in speech rate caused by an increasing reverberation time. Additionally, speech rate was the highest in the anechoic condition. Moreover, creak ratio decreased greatly when reverberation time increased to more than one second especially on male speakers and end-weighted sentences. Additionally, monotonousness was higher in reverberated conditions than the anechoic condition. However, substantial speaker-dependent differences in the effects of reverberation on speech were found. Moreover, sentence weight was found to influence speech more fundamentally than reverberation. The results suggest that rooms with average reverberation times, rather than particularly long or short, seem the most beneficial for speaking. This observation corresponds to previous studies. Further research on the field is required to extract valuable knowledge needed in acoustical design of spaces, including classrooms. Designing speaker-friendly spaces helps to preserve occupational speakers’ voices throughout their careers.

Universella meningsrepresentationer och flerspråkig språkmodellering är heta ämnen inom språkteknologi, specifikt området som berör förståelse för naturligt språk (natural language understanding). En meningsinbäddning (sentence embedding) är en numerisk skildring av en följd ord som motsvaras av en hel fras eller mening, speficikt som ett resultat av en omkodare (encoder) inom maskininlärning. Dessa representationer behövs för automatiska uppgifter inom språkteknologi som kräver förståelse för betydelsen av en hel mening, till skillnad från kombinationer av enskilda ords betydelser. Till sådana uppgifter kan räknas till exempel inferens (huruvida ett par satser är logiskt anknutna, natural language inference) samt åsiktsanalys (sentiment analysis). Med universalitet avses kodad betydelse som är tillräckligt allmän för att gynna andra relaterade uppgifter, som till exempel klassificering. Det efterfrågas tydligare samförstånd kring strategier som används för att bedöma kvaliteten på dessa inbäddningar, antingen genom att direkt undersöka deras lingvistiska egenskaper eller genom att använda dem som oberoende variabler (features) i relaterade modeller. På grund av att det är kostsamt att skapa resurser av hög kvalitet och upprätthålla sofistikerade system på alla språk som används i världen finns det även ett stort intresse för uppskalering av moderna system till språk med knappa resurser. Tanken med detta är så kallad överföring (transfer) av kunskap inte bara mellan olika uppgifter, utan även mellan olika språk. Trots att behovet av tvärspråkiga överföringsmetoder erkänns i forskningssamhället är utvärderingsverktyg och riktmärken fortfarande i ett tidigt skede. SentEval är ett existerande verktyg för utvärdering av meningsinbäddningar med speciell betoning på deras universalitet. Syftet med detta avhandlingsprojekt är ett försök att utvidga detta verktyg att stödja samtidig bedömning på nya uppgifter som omfattar flera olika språk. Bedömningssättet bygger på strategin att låta kodade meningar fungera som variabler i så kallade downstream-uppgifter och observera huruvida resultaten förbättras. En modern mångspråkig modell baserad på så kallad transformers-arkitektur utvärderas på en etablerad inferensuppgift såväl som en ny känsloanalyssuppgift (emotion detection), av vilka båda omfattar data på en mängd olika språk. Även om det praktiska genomförandet i stor utsträckning förblev experimentellt rapporteras vissa tentativa resultat i denna avhandling.

Onomatopoeettiset interjektiot ovat huudahduksenomaisia, ääntä jäljitteleviä sanoja, jotka ovat monessa kielessä suhteellisen marginaaliseksi koettu ilmiö eikä sitä ole siten kovin laajalti tutkittu. Tutkimuksen vähyyttä korostaa terminologian epäyhtenäisyys, sillä tutkimuksesta riippuen sanoja saatetaan kutsua ideofoneiksi, imitatiiveiksi tai ääniefekteiksi. Onomatopoeettiset interjektiot ovat erityisesti sarjakuvista tuttu sanaluokka. Sarjakuvissa esiintyviä onomatopoeettisia interjektioita on kuitenkin aiemmin tutkittu lähinnä käännösratkaisullisista näkökulmista yhden tai kahden kielen osalta laajemman äännesymbolisen vertailun sijaan. Tässä tutkimuksessa vertaillaan onomatopoeettisten interjektioiden käyttöä neljässä Fenno-Baltian alueen kielessä: suomessa, virossa, latviassa ja liettuassa. Aineistona käytetään paralleelikorpusta, joka koostuu suomen-, viron-, latvian- ja liettuankielisistä Aku Ankka -sarjakuvista. Tutkimuksen tavoitteena on selvittää interjektioiden ominaispiirteitä ja äännesymboliikkaa niin yleisesti kuin kielikohtaisella tasolla ja identifioida niistä semanttisen luokittelun avulla äännesymbolisia yksiköitä eli fonesteemejä. Tutkimuksen analyysiosuus koostuu fonologisesta ja semanttisesta osasta. Fonologisessa analyysissä verrataan interjektiosanaston grafeemifrekvenssiä neutraalin asiatekstin grafeemifrekvenssiin sekä tutkitaan kielten keskinäistä korrelaatiota sekä kieltenvälisiä eroja interjektioiden fonologian ja tavurakenteen osalta kielipareittain. Semanttisessa analyysissä interjektiot luokitellaan 15 eri luokkaan semanttisin perustein. Tämän avulla interjektioista identifioidaan tiettyyn semanttiseen käyttöalueeseen rajautuvia fonologisia piirteitä tai fonesteemejä niin kielikohtaisella tasolla kuin yleisesti. Tuloksista selviää, että erityisesti tremulantit ja klusiilit ovat yleisempiä onomatopoeettisissa interjektioissa kuin neutraalissa asiatekstissä. Sen sijaan grafeemi ⟨e⟩ on interjektioissa huomattavasti harvinaisempi, sillä suomen- ja vironkielisessä aineistossa sitä ei esiinny lainkaan. Kieliparien fonologinen korrelaatio interjektioaineistossa on odotettua sattumanvaraisempaa, mikä toisaalta tukee käsitystä äännesymboliikan universaaliudesta. Samaa kielii myös se, että interjektioista identifioidut fonesteemit ovat kielistä riippumatta hyvin samankaltaisia. Toisaalta jokaisesta kielestä löytyi myös yksittäisiä äännesymbolisia piirteitä, joita muissa kielissä ei esiintynyt.

Tutkielma tarkastelee selkokielen nominitaivutuksen periaatteiden toteutumista informoivissa selkoteksteissä. Tavoitteena on selvittää, toteutuvatko selkokielen mittarissa olevat nominitaivutuksen kriteerit teksteissä. Selkokielen mittarin avulla voidaan arvioida tekstin selkokielisyyttä. Mittarin kriteereissä kehotetaan suosimaan nominien perusmuotoja sekä välttämään vajaakäyttöisiä sijoja. Lisäksi selvitetään, mitä keinoja selkokirjoittajat ovat käyttäneet muuttaessaan nominien taivutusmuotoja perusmuodoiksi. Tavoitteena on kytkeä tulokset osaksi selkokielen mittarin kehitystyötä. Tutkielman taustalla on funktionaalinen kielikäsitys. Teoriaosa esittelee selkokielen kehittämistä, selkokielen ohjeistusta sekä luetun ymmärtämistä. Aineisto on kerätty korpus- ja kyselytutkimuksen menetelmin. Tutkielmassa vertaillaan kolmea selkokielistä esitettä tai opasta ja yleiskielisiä lähtötekstejä. Kustakin selkokielisestä tekstistä aineistoon kuuluu 1200–1300 sanaa. Kyselyllä selvitetään selkokielen kirjoittajien (n = 21) näkemyksiä sekä keinoja rakentaa tekstiin perusmuotoja. Kvantitatiivisessa tutkimuksessa lasketaan nominien perusmuotojen ja taivutusmuotojen esiintymien määrät. Perusmuotojen rakentumista selvitetään kvalitatiivisin menetelmin analysoimalla yleis- ja selkokielisten tekstien sanapareja sekä kyselytutkimuksen vastauksia. Tulosten mukaan nominien perusmuotoja esiintyy kaikissa selkokielisissä teksteissä enemmän kuin lähtöteksteissä. Myös erikseen tarkasteltujen substantiivien perusmuotoja esiintyy selkokielessä enemmän kuin yleiskielessä. Vajaakäyttöisiä sijoja ei selkoteksteissä ole käytetty. Nominitaivutuksen kriteereiden periaatteet toteutuvat aineistoon kuuluvissa teksteissä. Perusmuodon rakentamisen keinoina on käytetty esimerkiksi lausekkeiden muuttamista sivulauseiksi, lausetyypin muutoksia, sanavalintoja ja aktiivimuotoisia lauseita. Korpus- ja kyselytutkimuksen tulokset tukevat toisiaan, mutta tuloksia ei voida yleistää koskemaan kaikkia informoivia selkotekstejä. Tulosten perusteella selkokielen mittarin nominitaivutuksen kriteereihin esitetään joitakin tarkennuksia.

This thesis investigates the interaction between lexical tones and pitch reset in Akan, a Kwa language with about 8.1 million native speakers in Ghana (Eberhard et al., 2020). Experimental studies on Akan prosody are limited, although the language has a large first and second language speakers. This study seeks to increase our knowledge of the tone-intonation structure of the Akan language. In an earlier study on Akan complex declarative sentences, pitch reset occurred at the beginning of the content word that followed the clausal marker of an embedded clause (Kügler, 2016). Following a pilot study, a hypothesis was formed for the present study that pitch reset in complex declarative utterances in Akan also occurs within the clausal marker of the dependent clause and not only in the following content word. Focusing on the Asante Twi dialect, a controlled material consisting of 64 complex sentences were created. Five native speakers of Asante Twi were recorded as they produced the 64 sentences and additional 32 complex sentences used as fillers. The Mean f_0 values of the syllables of the subordinate conjunction and the syllables of the word before and after the conjunction were extracted and analysed in R; the statistical analysis was based on a linear mixed model. As expected, a reset in the pitch contour consistently occurred within the subordinate conjunction, contrasting the earlier study. The conjunction was phrased prosodically with the dependent clause to signal the syntactic relationship between the two. The degree of pitch register reset was also dependent on the tonal structure; reset was more significant when the initial tone of the conjunction was High but lesser when the conjunction began with a Low tone. Thus, the results show that lexical tones interact to determine the f_0 contour of Akan utterances and that the intonational contour of utterances is complex in the Akan language.

Modernit sanaupotusmenetelmät, esimerkiksi Word2vec, eivät mallinna leksikaalista moniselitteisyyttä luottaessaan kunkin sanan mallinnuksen yhden vektorirepresentaation varaan. Näin ollen leksikaalinen moniselitteisyys aiheuttaa ongelmia konekääntimille ja voi johtaa moniselitteisten sanojen käännökset usein harhaan. Työssä tarkastellaan mahdollisuutta mallintaa moniselitteisiä sanoja merkitysupotusmenetelmän (sense embeddings) avulla ja hyödynnetään merkitysupotuksia valvomattoman konekäännösohjelman (unsupervised machine translation) opetuksessa kieliparilla Englanti-Saksa. Siinä missä sanaupotusmenetelmät oppivat yhden vektorirepresentaation kullekin sanalle, merkitysupotusmenetelmän avulla voidaan oppia useita representaatioita riippuen aineistosta tunnistettujen merkitysten määrästä. Näin ollen yksi valvomattoman konekääntämisen perusmenetelmistä, sanaupotusten kuvaus joukosta lähde- ja kohdekielten yksikielisiä vektorirepresentaatioita jaettuun kaksikieliseen vektoriavaruuteen, voi tuottaa paremman kuvauksen, jossa moniselitteiset sanat mallintuvat paremmin jaetussa vektoriavaruudessa. Tämä mallinnustapa voi vaikuttaa positiivisesti konekäännösohjelman kykyyn kääntää moniselitteisiä sanoja. Työssä merkitysupotusmalleja käytetään saneiden alamerkitysten yksiselitteistämiseen, ja tämän myötä jokainen konekäännösmallin opetusaineistossa esiintyvä sane annotoidaan merkitystunnisteella. Näin ollen konekäännösmalli hyödyntää sanaupotusten sijaan merkitysupotuksia oppiessaan kääntämään lähde- ja kohdekielten välillä. Työssä opetetaan tilastollinen konekäännösmalli käyttäen tavanomaista sanaupotusmenetelmää. Tämän lisäksi opetetaan sekä tilastollinen että neuroverkkokonekäännösmalli käyttäen merkitysupotusmenetelmää. Aineistona työssä käytetään WMT-14 News Crawl -aineistoa. Opetettujen mallien tuloksia verrataan aiempaan konekäännöstutkimuksen automaattisessa arvioinnissa hyvin menestyneeseen tilastolliseen konekäännösmalliin. Lisäksi työssä suoritetaan tulosten laadullinen arviointi, jossa keskitytään yksittäisten moniselitteisten sanojen kääntämiseen. Tulokset osoittavat, että käännösmallit voivat hyötyä merkitysupotusmenetelmästä. Tarkasteltujen esimerkkien perusteella merkitysupotusmenetelmää hyödyntävät konekäännösmallit onnistuvat kääntämään moniselitteisiä sanoja sanaupotusmenetelmää hyödyntävää mallia tarkemmin vastaamaan referenssikäännöksissä valittuja käännöksiä. Näin ollen laadullisen arvioinnin kohdistuessa yksittäisten moniselitteisten sanojen kääntämiseen, merkitysupotusmenetelmästä näyttää olevan hyötyä konekäännösmallien opetuksessa.

This thesis investigates the roots of Koryo-mar, an endangered language currently spoken by Korean diaspora communities (also called Koryo-saram, lit. ‘Koryo-people’) in post-Soviet countries, such as Kazakhstan and Uzbekistan, to which they had been deported in 1937 from the Russian Far East (RFE). By the end of the 19th century, the language of the first Korean immigrants to the RFE, for the most part peasants from the Northeastern province of Hamgyŏng, had already caught the interest of a few scholars, among which can be distinguished Mikhail P. Pucillo, the author of the first Russian-Korean dictionary. Such volume, the Opyt Russko-Korejskogo Slovarja (1874) was based on the speech of the Koreans he was in contact with, and represents one of the earliest sources for the study of Koryo-mar. On the other hand, the actual research on the Koryo-saram begun in earnest only in the second half of the 20th century, and the work of Songmoo Kho, Koreans in Soviet Central Asia (1987), is often considered one of the most complete surveys on the history and the culture of Central Asian Koreans. It also contains a chapter devoted to Koryo-mar, that includes several wordlists of both native items and loanwords from Russian and other local languages. By comparing the data extracted from the above-mentioned texts, on a lexical and orthographic level, and highlighting their similarities and variations, this research aims at illustrating the genetic ties between the dialects of Northern Hamgyŏng and Koryo-mar, as well as the relationship between the spoken and the written language. Additional information has also been drawn from secondary sources in Early Modern Korean, such as Hŏ Kyung’s Hong Kiltong Chŏn (16th-17th century), Chang Kyehyang’s Ŭmsik Timibang (1670s), and J. S. Gale’s translation of J. Bunyan’s The Pilgrim’s Progress (1895). The research concludes that the forty elements common to Pucillo (1874) and Kho (1987) can be framed into five different relationship scenarios, which describe not only the continuation of lexical and orthographic material between the two texts and their languages, but also assess the latter’s connection to Standard Korean and its historical variants. Moreover, the examination of Pucillo’s usage of Cyrillic in transcribing Korean provides relevant details concerning the articulatory habits of Northeastern Koreans in the late 1800s, which are contextualised within and in contrast to the writing practices of Early Modern Korean, that are known to be extremely inconsistent. This study, based on a philological approach, can aid in outlining a literary tradition that encompasses the historical development of Koryo-mar, which is still an unrecognised minority language, while the analysis of orthographic peculiarities can shed light on the chronology of the spread of certain phonological changes across the whole Korean peninsula.

This study examines the use of the Latin script in languages where the Latin script is not the official writing system. The aim of the research is to get a general perception of where and why the Latin script is used in those languages. The study concentrates on the use of the Latin script of individual writers, especially in informal writing situations online. The study also examines if the individuals have seen the Latin script used by others. In addition, the study finds out about the keyboard functionality when writing in the Latin and non-Latin scripts. Many case studies have covered the phenomenon of digraphia, which is a situation where one language uses two writing systems. However, a broader crosslinguistic study has not been done on this topic. The Latin script is the dominant script on the Internet which might have an impact on the script choice in online writing. Internet language and chatting are broadly studied but because of the continuously changing nature of the Internet, more research is required on these topics. This study aims to fill in some of the gaps that are left open by previous research. I gathered data by an online questionnaire and by four semi-structured interviews. By means of the questionnaire, I got open question answers and quantitative data from 142 respondents. Additionally, I had four interviews with representative users of different writing systems to get more detailed information about the use of the Latin script. The data was analysed and compared with previous research from the point of view of five groups of different scripts. The general finding of this study is that the respondents choose to write in the Latin script mostly when the non-Latin keyboard in not available or when fast and easy informal writing style is required. The Latin script has many different functions. It is mostly used in online chatting, texting with friends or when one’s personal information needs to be written in travelling documents. However, people prefer to write in the official non-Latin script even though it might sometimes feel difficult or slow to use. The study points out that it cannot be predicted whether a respondent uses the Latin script or not, since the results are mostly very variable. The need or even the eagerness to use the Latin script when it is not the official script, reflects the functions and facilities that are available or unavailable for different scripts.