Browsing by master's degree program "Kielellisen diversiteetin ja digitaalisten ihmistieteiden maisteriohjelma"

Depressive disorders and especially Major Depressive Disorder (MDD) have recently raised a lot of concerns. According to Bains & Abdijadid, (2022), in 2030 MDD will be the most common concern in terms of mental well-being and Kessler & Bromet (2013) pointed out that MDD “has one of the highest lifetime prevalence among psychiatric disorders”, and for this reason MDD results in a significant burden from an individual and societal perspective (Lépine & Briley, 2011). Anhedonia which is a core symptom of MDD (Gorwood 2008, Liang et al. 2022) can be defined as the “inability to enjoy experiences or activities that normally would be pleasurable” (APA Dictionary of Psychology, n.d.-c). Its effects on cognitive processes such as reward responsiveness has been the interest of various studies. Pizzagalli et al. (2008) demonstrated in their study that individuals suffering from MDD and experiencing anhedonic symptoms have a blunted reward responsiveness compared to some healthy subjects. This master thesis aimed to replicate Pizzagalli et al. (2008) study to reinvestigate anhedonia´s impact on reward responsiveness as well as anxiety´s role since individuals suffering from MDD also face anxiety or anxiety disorder (Xin et al. 2015).Two groups were formed, a clinical group (N=29) based on the DSM-IV criteria and a control group (N=20), and both groups were given the Face Game or the Probabilistic Reward Task (Pizzagalli et al. 2008) to measure reward responsiveness. Reward responsiveness was computed as “response bias” and the perceived difficulty of the task as “discriminability”. The results showed no significant difference in response bias scores between the depressed and healthy participants, and no significant correlations were found between depression scores, anhedonia scores, anxiety scores, and response bias scores. An effect of the experimental task’s blocks on response bias and discriminability scores was found, but no interaction effect (group * block) was found for the response bias and discriminability scores. Thus, we cannot argue in favor of the results obtained by Pizzagalli et al, (2008) that individuals suffering from MDD and especially experiencing anhedonic symptoms are more likely to have a blunted reward sensitivity compared to some healthy subjects. The statistically non-significant results were attributed to the small sample size, the complexity of the task and its design. Therefore, further investigations should focus on getting larger clinical and control groups and they also should redefine some aspects of the experimental task to make it more sensitive to any changes to reward responsiveness among individuals suffering from MDD and experiencing anhedonic symptoms.

This study gives a glimpse on some elements that were used in Neo-Assyrian names. The data set used in this study is ‘A Social Network of the Prosopography of the Neo-Assyrian Empire’ that has been edited by Dr Heidi Jauhiainen and Dr Tero Alstola. The data set is based on the ‘The Prosopography of Neo-Assyria (PNA)’ project directed by Dr Simo Parpola at the University of Helsinki. The data set originally included names in more than ten different languages, but this study focuses solely on names that are in Akkadian, so names of other origins were left out. Also, only names that were identified as male were studied. As this study is a master’s thesis for digital humanities, digital methods were used. Neo-Assyrian names offer an interesting insight to the society of Neo-Assyria and how the Akkadian language was used. The names were short phrases or sentences, so they can tell a lot of the surrounding society. Especially theophoric elements and how they were connected to other elements is something that this study focuses on. Theophoric elements are usually names of deities which were often used in Neo-Assyrian and Neo-Babylonian names. The most common theophoric elements in this data set were Nabû, Aššur, and Bēl. In addition of theophoric elements, other elements were studied too. The most frequent other element was naṣāru ‘to protect’ which was almost only found together with theophoric elements. Other popular elements were nadānu ‘to give’ and šarru ‘king’. The naming patterns of Neo-Babylonia have been studied more than Neo-Assyrian practices. For example, it was quite common for Neo-Babylonians to give their first-born son a name that incorporated the elements aplu ‘heir’ and the theophoric element Marduk. In the names of second-born sons the theophoric element Nabû was often combined with some form of the verb nadānu ‘to give’. Based on this study it seems that a similar practice of naming a second-born son might have been used in Neo-Assyria as well. However, the element aplu was very infrequent in the studied data set.

Ihmiset muuttavat äänentuotantoaan kuuluvammaksi meluisassa ympäristössä refleksinomaisesti. Tätä ilmiötä kutsutaan Lombard-efektiksi. Efekti saa puhujan tuottamaan Lombard-puhetta, jota on tutkittu jo yli vuosisadan ajan eri näkökulmista. Lombard-puheen akustiikalle ominaista ovat korotettu äänenpainetaso, korotettu puheäänen perustaajuus, muutokset äänen osataajuuksissa sekä muissa äänen spektrin rakenteissa. Lisäksi Lombard-puheessa vokaalien pituuksilla on tapana kasvaa, ja äärimmäisissä meluolosuhteissa hyperartikulaatiota voi esiintyä. Puhetilanteeseen sisältyvä kommunikatiivinen aspekti on keskeistä ilmiön synnylle. Tämän tutkielman tavoitteena oli tutkia puheentuottoa keskustelutilanteessa, jossa samanaikaisesti toinen keskustelijoista on altistettuna melulle ja tuottaa täten Lombard-puhetta, ja toinen keskustelija kommunikoi hiljaisuudessa ilman taustamelun suoria vaikutuksia, ja selvittää, onko puheen akustiikassa tai ymmärrettävyydessä eroavaisuuksia tällaisessa epäsymmetrisessä tilanteessa verrattuna symmetriseen puhetilanteeseen, jossa molempien puhujien ääniympäristö on sama. Tutkimusta varten kaksi paria suomenkielisiä keskustelijoita (yhteensä neljä osallistujaa, kaikki naisia) ratkoivat pareittain sudokupohjaisia tehtäviä kolmessa eri taustamelutilanteessa: (1) hiljaisuudessa, (2) molempien ollessa taustamelussa (symmetrinen), ja (3) vain toisen keskustelijan ollessa taustamelussa (asymmetrinen). Taustamelu, jota soitettiin koehenkilöille 75 dB äänenpainetasolla, oli laadultaan cocktail-melua, joka sisältää niin kutsuttua puheensorinaa jossa useampi puhuja puhuu päällekkäin. Keskustelut äänitettiin ja niistä kerättiin yhteensä 453 maalitavua, joista kaikista analysoitiin keskimääräinen äänenpainetaso, ja 417 maalitavusta analysoitiin keskimääräinen perustaajuus. Äänenpainetason ja perustaajuuden arvot normalisoitiin ja arvoille suoritettiin keskiarvoja ja variansseja vertailevat tilastolliset testit. Odotetusti kaikki puhujat korottivat äänenpainetasoaan ja perustaajuuttaan siirryttäessä hiljaisesta keskustelutilanteesta symmetriseen taustamelutilanteeseen, jossa molemmat keskustelukumppanit tuottivat Lombard-puhetta. Henkilöt, jotka asymmetrisessä keskustelutilanteessa olivat itse hiljaisuudessa ja kommunikoivat keskustelukumppanille, joka oli melussa, korottivat sekä äänenpainetasoaan että perustaajuuttaan asymmetrisessä keskustelutilanteessa verrattuna hiljaiseen keskustelutilanteeseen. Lisäksi toinen näistä puhujista korotti sekä äänenpainetasoaan että perustaajuuttaan lähes oman Lombard-puheensa tasolle, jota mitattiin symmetrisessä tilanteessa. Puhujat, jotka olivat altistettuna melulle asymmetrisessä tilanteessa, käyttivät keskimäärin matalampaa äänenpainetasoa asymmetrisessä kuin symmetrisessä tilanteessa, vaikka tuottivatkin Lombard-puhetta molemmissa tilanteissa. Väärin kuultuja maalitavuja ei havaittu asymmetrisessä tilanteessa, vaan henkilöt, jotka olivat kyseisessä tilanteessa hiljaisuudessa, onnistuivat korottamaan ääntään tarvittavalle tasolle, jotta ratkaiseva tieto saatiin kommunikoitua melussa olevalle henkilölle. Tämä tutkimus osoitti, että kahden keskustelukumppanin ääniympäristöjen ollessa eriävät, kumpikaan keskustelijoista ei tuota täysin sentyyppistä puhetta, joka olisi sopivaa heidän senhetkiseen ääniympäristöönsä, vaan puheentuottoon vaikuttaa myös välillisesti keskustelukumppanin ääniympäristö. Lisäksi tutkimus osoitti, että siinä missä puhetilanteen kommunikatiivisuus voi lisätä Lombard-efektin vaikutuksia, se voi myös häivyttää niitä. Jatkotutkimuksissa tulisi kerätä enemmän dataa ja suorittaa datalle laajempaa analyysiä.

Social media provides huge amounts of potential data for natural language processing but using this data may be challenging. Finnish social media text differs greatly from standard Finnish and models trained on standard data may not be able to adequately handle the differences. Text normalization is the process of processing non-standard language into its standardized form. It provides a way to both process non-standard data with standard natural language processing tools and to get more data for training new tools for different tasks. In this thesis I experiment with bidirectional recurrent neural network models and models based on the ByT5 foundation model, as well as the Murre normalizer to see if existing tools are suitable for normalizing Finnish social media text. I manually normalize a small set of data from the Ylilauta and Suomi24 corpora to use as a test set. For training the models I use the Samples of Spoken Finnish corpus and Wikipedia data with added synthetic noise. The results of this thesis show that there are no existing tools suitable for normalizing Finnish written on social media. There is a lack of suitable data for training models for this task. The ByT5-based models perform better than the BRNN models.

This master's thesis seeks to conceptually replicate psychologist Michael Kosinski's study, published in 2021 in Nature Scientific Reports, in which he trained a cross-validated logistic regression model to predict political orientations from facial images. Kosinski reported that his model achieved an accuracy of 72\%, which is significantly higher than the 55\% accuracy measured in humans for the same task. Kosinski's research attracted a huge amount of attention and also accusations of pseudoscience. Where Kosinski trained his model with facial features containing information for example about head position and emotions, in this thesis I use a deep learning convolutional neural network for the same task. Also, I train my model with Finnish data, consisting of photos of the faces of Finnish left- and right-wing candidates gathered from the 2021 municipal elections. I research whether a convolutional neural network can learn to predict from candidates' faces whether a member of a Finnish party belongs to either the right-wing Coalition Party (Coalition) or the left-wing Left Alliance (Left Alliance) with better than 55\% accuracy, and what is the possible role of color information on the classification accuracy of the model. On this basis, I also consider the wider ethical issues surrounding these types of models and the technological advances they bring. There has been a recent ethical debate on the widespread use of facial recognition technology in relation to issues such as human autonomy, privacy, and civil liberties. In the context of previous scientific findings, there has also been debate about the potential ability of facial recognition technologies to reveal information about our most personal traits, such as sexual orientation, personality, and emotional states. Thus, facial recognition technologies are also closely related to privacy issues. In his original article, Michael Kosinski did not underestimate the many problematic ethical issues that the use of facial recognition technology can raise. He did, however, underline the role of science in trying to determine the function, capability, and accuracy of these technologies. Only through research can we gain insights into these technologies, which can then potentially be used to inform societal decision-making. This research approach is also the aim of this Master's thesis.

There are comprehensive requirements in Finland for procurement by any government organization to go through a tendering process where information about each tender is made available not only to vendors and service providers, but to everyone else in Finland as well. This is accomplished through the website Hilma and should make tenders easy to find. Moreover, in Finnish, variance in domain terminology is not thought to be the problem that it is in English. For instance, the last four years of tenders on Hilma never refer to jatkuva parantaminen as toiminnallinen erinomaisuus whereas “continuous improvement” and “operational excellence” could be used interchangeably in English. And yet, it is considered very difficult for a vendor or service provider to find applicable tenders on Hilma. Rather than lexical variability being the cause as it might be in English, the differences in concept paradigms between the private and public sectors in Finland pose the challenge. Whereas a taxi company representative would be looking for tenders about transportation services, a public officer could be posting a tender about social equity for the disabled. The second difficulty is that the Hilma search engine is purely Boolean with restrictive string match criteria rather than inviting natural language questions. Finally, the Hilma search engine does not account for Finnish being a highly inflecting and compounding language where single words usually morph instead of taking on adpositions, and where compound words are affixed together without hyphenation. Many information retrieval approaches would look outside the corpus for query expansion terms. Natural language processing might also offer the potential to look for paraphrases in existing parallel corpora on tenders throughout the European Union rather than in Hilma. However, this thesis focuses on clustering the tenders posted in Finnish on Hilma, applying the comprehensive workflow of the very recent BERTopic package for Python. All documents in each cluster are concatenated and the highest TFIDF-scoring words in the concatenated document are slated to be “search extension terms.” If one of the terms were to be entered by a Hilma user, the user would be invited to perform parallel searches with the remaining terms as well. The first main contribution of this thesis is to use state of the art models and algorithms to represent the corpus, reduce dimensionality of the representations and hierarchically cluster the representations. Second, this thesis develops analytical metrics to be used in automatic evaluation of the efficacy of the clusterings and in comparisons among model iterations that programmatically remove more and more distractions to the clustering that are discovered in the corpus. Finally, this thesis performs case studies on Hilma to demonstrate the remarkable efficacy of the search extension terms in generating tremendous numbers of additional useful matches, addressing paradigm-based differences in terminology, morphovariance and affixation.

The purpose of this thesis is to examine the prosodic features of English dialects using WaveNet. The exact goal is to investigate whether the differences in prosody between the dialects are present in the data and the results, and whether the geographical distance between the cities included in the data has any influence on this. Another aim is to see how the prosodic features of the sentence types present in the data and their possible differences are manifested in the data and the results. Prosody is concerned with those characteristics of speech which cover more than just individual sounds. Prosodic features can further be divided into paralinguistic features, such as the rate of speech and pausing, and linguistic features, like intonation. Parameters useful for analysing prosody are fundamental frequency (f0), intensity and voice quality – we are interested in the first two. Fundamental frequency is the speed of the vibration of the vocal folds while speaking. Intensity in turn is connected to the changes of air pressure while speaking. The data used for this study is the IViE corpus (Intonational Variation in English), comprising of recordings done in nine British cities – Belfast, Bradford, Cambridge, Cardiff, Dublin, Leeds, Liverpool, London and Newcastle, with approximately 12 speakers per city. In three of the cities, Bradford, Cardiff and London, the dialect is that of a minority. The part of the corpus chosen for this study is a set of 22 sentences consisting of five sentence types. The analysis was performed using WaveNet, a convolutional neural network. It uses causal convolutions to ensure the data is processed correctly. In addition to being conditioned on the output of the network itself, it can also be conditioned using embeddings. The WaveNet implementation used here has two embedding layers – target and normalisation embeddings. Before the analysis the data was pre-processed and the relevant information concerning the fundamental frequency and intensity were extracted from the sound files. A corresponding *.time file was also created for each of the sound files, with the aim of minimising the influence of the possible differences in length between sentences and thus improve the network's ability to recognise the intonation contours correctly. The results are presented in the form of dendrograms, depicting the relationships between the dialects and sentence types – both separately and as a combination of the dialects and sentence types. It was shown, that the differences in prosody were in fact manifested in the data for both dialects and sentence types, although not exactly as expected. The geographical proximity did not seem to influence the dialectal similarities as much as was assumed – in addition to other influences this might also be due to some of the dialects being minority dialects in the cities, and therefore not necessarily so easily comparable to the dialects of the neighbouring area as the majority dialects might have been.

Hostility in the player communication of video games (and by extension, mobile games) is a well-documented phenomenon that can have negative repercussions for the well-being of the individual being subjected to it, and the society in general. Existing research on detecting hostility in games through machine learning methods is scarce due to the unavailability of data, imbalanced existing data (few positive samples in a large data set), and the challenges involved in defining and identifying hostile communication. This thesis utilizes communication data from the Supercell game Brawl Stars to produce two distinct machine learning models: a support vector classifier and a multi-layer perceptron. Their performance is compared to each other as well as to that of an existing sentiment analysis classifier, VADER. Techniques such as oversampling and using additional data are also used in an attempt to reach better results by improving the balance of the data set. The support vector classifier model was found to have the best performance overall, with an F1 score of 64.15% when used on the pure data set and 65.74% when combined with the SMOTE oversampling algorithm. The thesis includes an appendix with a list of the words that were found to have the strongest influence on the hostile/non-hostile classification.

Smiling is fundamentally human but a more complex phenomenon than might appear at first glance. Studies in the field of language sciences have explored smiling in the context of speech and found that speaking while smiling has perceivable effects on the voice, and this phenomenon is commonly known as “smiling voice”. Although this phenomenon is widely recognised, there is no clear consensus on the precise acoustic characteristics that cue listeners to the presence of a smile. This study aims to investigate whether listeners can identify smiling voice based only on audio stimuli, what prosodic cues or characteristics they might be using to do so, and whether those cues can be extracted and used to replicate smiling voice using speech synthesis. Another aim of this study is to determine whether the level of perceived smiliness can be controlled in synthetic speech. These issues are addressed with the objectives of adding to the understanding of smiling voice in the field of phonetics and exploring the potential of speech synthesis technology for producing expressive speech. A corpus of Finnish speech was used to conduct a preliminary listening experiment where participants compared neutral and positive utterances in a questionnaire and indicated whether the speaker was smiling in the latter. Utterances that were identified as smiley were analysed acoustically to detect prosodic differences between neutral and smiley speech. Based on the results, formant frequencies F2 and F3 and centre of gravity were selected as prosodic cues to control smiling in speech synthesis. The speech synthesiser was a Tacotron 2 system, including a reference encoder, which was already trained on the speech corpus used. Synthesis evaluation was conducted with a second questionnaire where participants listened to the synthesised utterances and indicated how strongly the speaker was smiling. The results of the first questionnaire showed that listeners were able to distinguish neutral and smiley speech, and subsequent acoustic analyses indicated significant effects of smiling on fundamental frequency, formant frequencies, and centre of gravity. Speech synthesis evaluation results further indicated that F2, F3, and centre of gravity can be used to control the level of perceived smiling at least to the extent of a binary distinction. However, the evaluation showed that more sophisticated control of the level of smiling voice was not achieved.

Niin kutsuttu helsinkiläinen ässä on 1800-luvulta lähtöisin oleva sosiolingvistinen ilmiö, jonka mukaan helsinkiläisillä suomen kielen puhujilla on muualla Suomessa asuvia terävämpi, sihisevämpi [s]-äänne. Ilmiön pitkästä historiasta huolimatta siitä ei ole aiemmin tehty foneettista tutkimusta joka selvittäisi, onko helsinkiläisten puhujien tuottama [s]-äänne akustisesti tarkasteltuna tavallista suomen kielen [s]-äännettä terävämpi. Tämän maisterintutkielman tavoitteena on tarkastella helsinkiläisten puhujien tuottamia [s]-äänteitä akustisin menetelmin. Tutkielmassa selvitetään, mitkä tekijät vaikuttavat pitkän [s]-äänteen terävyyteen, onko terävyydessä eroa mies- ja naispuhujien välillä ja ovatko helsinkiläisten tuottamat [s]-äänteet tavanomaista terävämpiä akustisesti mitattuna. Aineistona käytettiin Kielipankin tarjoaman Helsingin puhekielen pitkittäiskorpuksen vuonna 2013 kerättyä osakorpusta. Koehenkilöitä oli 13 ja heidän puheestaan poimittuja, analysoitavia pitkiä [s]-äänteitä 622 kpl. Pitkistä [s]-äänteistä mitattiin Centre of Gravity -arvo (COG), joka kuvaa, mille taajuusalueelle [s]-äänteen energia on keskimääräisesti sijoittunut. Kyseisen arvon voidaan ajatella kuvaavan [s]-äänteen terävyyttä, sillä terävässä [s]-äänteessä energia on sijoittunut korkeille taajuuksille ja vähemmän terävässä eli ääntöpaikaltaan takaisemmassa [s]-äänteessä matalammille taajuuksille. Työssä tarkasteltiin puhujan ominaisuuksien (sukupuoli, ikä, koulutustausta) sekä pitkää [s]-äännettä edeltävän vokaalin ominaisuuksien (etisyys/takaisuus, suppeus-/väljyysaste sekä pyöreys/laveus) vaikutusta [s]-äänteen COG-arvoon. Lisäksi tarkasteltiin COG-arvoltaan erityisen korkeiden (COG-arvo yli 6000 Hz) pitkien [s]-äänteiden osajoukkoa ja sitä, mitkä tekijät olivat COG-arvoltaan erityisen korkeiden [s]-äänteiden taustalla. Pitkästä [s]-äänteestä mitatun Centre of Gravity -arvon korkeuteen vaikuttivat vähintään tilastollisesti merkitsevästi niin puhujan sukupuoli, ikä, koulutustausta kuin [s]-äännettä edeltävän vokaalin etisyys/takaisuus, suppeus-/väljyysaste kuin sen pyöreys/laveus. COG-arvoltaan kaikista korkeimmat pitkät [s]-äänteet löydettiin ammattikoulutaustaisilta keski-ikäisiltä naispuhujilta edeltävän vokaalin ollessa etinen, lavea ja suppea tai puolisuppea. Naisten tuottamien [s]-äänteiden COG-arvo oli korkeampi kuin miesten, mikä tukee käsitystä, jonka mukaan helsinkiläinen ässä on erityisesti naisten puheen piirre. Tämän tutkimuksen perusteella ei voida todeta, että helsinkiläisten tuottamat [s]-äänteet olisivat keskimääräistä suomen [s]-äännettä terävämpiä, sillä aineistoon sisältyi niin Centre of Gravity -arvon kuin auditiivisen arvion perusteella monenlaisia pitkiä [s]-äänteitä. Mukana oli kuitenkin myös useita COG-arvoltaan hyvin korkeita ja auditiivisesti arvioituna hyvin teräviä [s]-äänteitä, mikä puolestaan viittaa sekä siihen, että COG-arvo olisi kelvollinen mittari [s]-äänteen terävyyden tarkastelemiseksi että ennen kaikkea siihen, että helsinkiläisellä ässällä voisi olla myös foneettista pohjaa.