Neural network models such as Transformer-based BERT, mBERT and RoBERTa are achieving impressive performance (Devlin et al., 2019; Lewis et al., 2020; Liu et al., 2019; Raffel et al., 2020; Y. Sun et al., 2019), but we still know little about their inner working due to the complex technique like multi-head self-attention they implement. Attention is commonly taken as a crucial way to explain the model outputs, but there are studies argue that attention may not provide faithful and reliable explanations in recent years (Jain and Wallace, 2019; Pruthi et al., 2020; Serrano and Smith, 2019; Wiegreffe and Pinter, 2019). Bastings and Filippova (2020) then propose that saliency may give better model interpretations since it is designed to find which token contributes to the prediction, i.e. the exact goal of explanation. 

In this thesis, we investigate the extent to which syntactic structure is reflected in BERT, mBERT and RoBERTa trained on English and Chinese by using a gradient-based saliency method introduced by Simonyan et al. (2014). We examine the dependencies that our models and baselines predict. 

We find that our models can predict some dependencies, especially those that have shorter mean distance and more fixed position of heads and dependents, even though all our models can handle global dependencies in theory. Besides, BERT usually has higher overall accuracy on connecting dependents to their corresponding heads, followed by mBERT and RoBERTa. Yet all the three model in fact have similar results on individual relations. Moreover, models trained on English have better performances than models trained on Chinese, possibly because of the flexibility of Chinese language. 

Sentiment analysis aims at figuring out the opinions of the users towards a certain service or product. In this research, the aim is at classifying the sentiments of users based on the comments they have posed on Douban movie website. In this thesis, I try two different ways to classify the sentiments: with the first one classifying comments into five classes of ratings from 1 to 5, and with the second one classifying comments into three classes of ratings: negative, neutral and positive. For the latter, the ratings of 1 and 2 are grouped as negative, the ratings of 3 neutral and the ratings of 4 and 5 positive.

First, Term Frequency Inverse Document Frequency (TF-IDF) is used as the feature extraction technique for machine learning algorithms. Chi Square and Mutual Information are used for feature selection. The selected features are fed into different machine learning methods: Logistic Regression, Linear SVC, SGD classifier and Multinomial Naive Bayes. The performance of models with feature selection will be compared with the performance of models without feature selection for 5-class classification as well as 3-class classification.

Also, fastText and Skip-Gram are used as embedding methods for deep learning algorithms LSTM and BILSTM. FastText will also be used for both embedding as well as being a classifier. The aim is to compare different machine learning and deep learning algorithms using different vectorization methods to see which model performs the best regarding both 5-class and 3-class classification.

The two classification strategies will be compared with each other in terms of error analysis. The aim is to figure out the similarities and differences of misclassifications made by two different classification strategies.

Cross-lingual transfer has been shown effective for dependency parsing of some low-resource languages. It typically requires closely related high-resource languages. Pre-trained deep language models significantly improve model performance in cross-lingual tasks. We evaluate cross-lingual model transfer on parsing Marathi, a low-resource language that does not have a closely related highresource language. In addition, we investigate monolingual modeling for comparison. We experiment with two state-of-the-art language models: mBERT and XLM-R. Our experimental results illustrate that the cross-lingual model transfer approach still holds with distantly related source languages, and models benefit most from XLM-R. We also evaluate the impact of multi-task learning by training all UD tasks simultaneously and find that it yields mixed results for dependency parsing and degrades the transfer performance of the best performing source language Ancient Greek.

This thesis aims to perform named entity recognition for English social media texts. Named Entity Recognition (NER) is applied in many NLP tasks as an important preprocessing procedure. Social media texts contain lots of real-time data and therefore serve as a valuable source for information extraction. Nevertheless, NER for social media texts is a rather challenging task due to the noisy context. Traditional approaches to deal with this task use hand-crafted features but prove to be both time-consuming and very task-specific. As a result, they fail to deliver satisfactory performance. The goal of this thesis is to tackle this task by automatically identifying and annotating the named entities with multiple types with the help of neural network methods. In this thesis, we experiment with three different word embeddings and character embedding neural network architectures that combine long short- term memory (LSTM), bidirectional LSTM (BI-LSTM) and conditional random field (CRF) to get the best result. The data and evaluation tool comes from the previous shared tasks on Noisy User-generated Text (W- NUT) in 2017. We achieve the best F1 score 42.44 using BI-LSTM-CRF with character-level representation extracted by a BI-LSTM, and pre-trained word embeddings trained by GloVe. We also find out that the results could be improved with larger training data sets.

In this thesis, we explore the impact of M-BERT and different transfer sizes on the choice of different transfer languages in dependency parsing. In order to investigate our research questions, we conduct a series of experiments on the treebanks in Universal Dependencies with UUParser.    

The main conclusions and contributions of this study are as follows:  

First, we train a variety of languages in several different scripts with M-BERT being added into the parsing framework, which is one of the most state-of-the-art deep learning models based on the Transformer architecture. In general, we get advancing results with M-BERT compared with the randomly initialized embedding in UUParser.   

Second, since it is a common way to choose a source language, which is 'close' to the target language in cross-lingual parsing, we try to explore what 'close' languages actually are, as there is not a definition for 'close'. In our study, we explore how strongly the parsing results are correlated with the different linguistic distances between the source and target languages. The relevant data is queried from URIEL Database. We find that the parsing performance is more dependent on inventory, syntactic and featural distance than on the geographic, genetic and phonological distance in zero-shot experiments. In the few-shot prediction, the parsing accuracy shows stronger correlation with inventory and syntactic distance than with others.    

Third, we vary the training sizes in few-shot experiments with M-BERT being added to see how the parsing results are influenced. We find that it is very obvious that few-shot experiments outperform zero-shot experiments. With the source sizes being cut, all parsing scores decrease. However, we do not see a linear drop of the results.

Text normalization plays a crucial role in text-to-speech systems by ensuring that the input text is in an appropriate format and consists of standardized words prior to grapheme-to-phoneme conversion for text-to-speech. The aim of this study was to assess the performance of five text normalization systems based on different methods. These text normalization systems were evaluated on the English Google text normalization dataset. The evaluation was based on the similarity between the ground truth and normalized outputs from each text normalization system. Since multiple ground truth issues occurred during the evaluation, the original similarity scores needed to be manually re-scored. The re-scoring was employed on a sample data semi-randomly extracted from the evaluation dataset. According to the results, the accuracy of these text normalization systems  can be ranked as follows: the Duplex system, the Hybrid system, the VT system, the RS system, and the WFST system. For the two rule-based systems from ReadSpeaker, the VT system performed slightly better than the RS system, with a slight difference in the original similarity score. By analyzing the error patterns produced during the normalization process, the study provided valuable insights into the strengths and limitations of these systems. The findings of this study contribute to the refinement of internal rules, leading to improved accuracy and effectiveness of text normalization in text-to-speech applications.

Multilingual natural language processing (NLP) is increasingly recognized for its potential in processing diverse text-type data, including those from social media, reviews, and technical reports. Multilingual language models like mBERT and XLM-RoBERTa (XLM-R) play a pivotal role in multilingual NLP. Notwithstanding their capabilities, the performance of these models largely relies on the availability of annotated training data. This thesis employs the multilingual pre-trained model XLM-R to examine its efficacy in sequence labelling to open-ended questions on democracy across multilingual surveys. Traditional annotation practices have been labour-intensive and time-consuming, with limited automation attempts. Previous studies often translated multilingual data into English, bypassing the challenges and nuances of native languages. Our study explores automatic multilingual annotation at the token level for democracy survey responses in five languages: Hungarian, Italian, Polish, Russian, and Spanish. The results reveal promising F1 scores, indicating the feasibility of using multilingual models for such tasks. However, the performance of these models is closely tied to the quality and nature of the training set. This research paves the way for future experiments and model adjustments, underscoring the importance of refining training data and optimizing model techniques for enhanced classification accuracy.

Offensive language detection (OLD) has become a crucial area of research in recent years. Despite growing interest, the majority of current studies have focused on English, leaving many other languages, like Chinese, relatively unexplored. In this thesis, we investigate the effectiveness of various intermediate tasks for enhancing the performance of detecting offensive content in Chinese, including sentiment analysis, continued masked language modeling, and cross-lingual transfer learning. By fine-tuning various pre-trained language models with these tasks, our goal is to capture knowledge from the intermediate tasks and transfer it to the target task. Our findings demonstrate that while intermediate training generally improves model performance on the Chinese OLD task, the extent of improvement varies based on the specific intermediate task, model architecture, and the amount of target data.

With podcasts being a fast growing audio-only form of media, an effective way of promoting different podcast shows becomes more and more vital to all the stakeholders concerned, including the podcast creators, the podcast streaming platforms, and the podcast listeners. This thesis investigates the relatively little studied topic of automatic podcast trailer generation, with the purpose of en- hancing the overall visibility and publicity of different podcast contents and gen- erating more user engagement in podcast listening. This thesis takes a hotspot- based approach, by specifically defining the vague concept of “hotspot” and designing different appropriate methods for hotspot detection. Different meth- ods are analyzed and compared, and the best methods are selected. The selected methods are then used to construct an automatic podcast trailer generation sys- tem, which consists of four major components and one schema to coordinate the components. The system can take a random podcast episode audio as input and generate an around 1 minute long trailer for it. This thesis also proposes two human-based podcast trailer evaluation approaches, and the evaluation results show that the proposed system outperforms the baseline with a large margin and achieves promising results in terms of both aesthetics and functionality.

In this paper, we present our process to establish a PICO and a sentiment annotated corpus of clinical trial publications. PICO stands for Population, Intervention, Comparison and Outcome - these four classes can be used for more advanced and specific search queries. For example, a physician can determine how well a drug works only in the subgroup of children. Additionally to the PICO extraction, we conducted a sentiment annotation, where the sentiment refers to whether the conclusion of a trial was positive, negative or neutral. We created both corpora with the help of medical experts and non-experts as annotators.

Foreign words, such as names, locations or sometimes entire phrases, are a problem for any system that is meant to convert graphemes to phonemes (g2p; i.e.converting written text into phonetic transcription). In this thesis, we investigate both rule-based and neural methods of phonetically transcribing English words found in Icelandic text, taking into account the rules and constraints of how foreign phonemes can be mapped into Icelandic phonology.

We implement a rule-based system by compiling grammars into finite-state transducers. In deciding on which rules to include, and evaluating their coverage, we use a list of the most frequently-found English words in a corpus of Icelandic text. The output of the rule-based system is then manually evaluated and corrected (when needed) and subsequently used as data to train a simple bidirectional LSTM g2p model. We train models both with and without length and stress labels included in the gold annotated data.

Although the scores for neither model are close to the state-of-the-art for either Icelandic or English, both our rule-based system and LSTM model show promising initial results and improve on the baseline of simply using an Icelandic g2p model, rule-based or neural, on English words. We find that the greater flexibility of the LSTM model seems to give it an advantage over our rule-based system when it comes to modeling certain phenomena. Most notable is the LSTM’s ability to more accurately transcribe relations between graphemes and phonemes for English vowel sounds.

Given there does not exist much previous work on g2p transcription specifically handling English words within the Icelandic phonological constraints and it remains an unsolved task, our findings present a foundation for the development of further research, and contribute to improving g2p systems for Icelandic as a whole.

This study of Silius Italicus’ Punica is executed by means of the analytical tools provided in Hegel’s Aesthetics together with the Weberian concept of the Idealtypus. The Punica and in particular a group of its protagonists are examined in relation to the “Hegelian Idealtypus of epic”, in order to identify the driving forces in the narrative and the pivotal conflict of the poem.

I argue that although the Punica displays numerous traditional epic features regarding formal criteria and subject matter, its historical situation must be taken into consideration as well as its text-internal qualities. Accordingly, the Punica must be considered a Kunstepos.

Furthermore, in the frequent occurrences of conflict and disintegration within the epicizing narrative, the poetic principle of drama emerges. This quality is shared by Livy’s reflective and pragmatical history. The dramatic principle, however, is not allowed its full realization. The narrating subject of the Punica appropriates the space of the objective narrator of the ideal-typical epic, but is substantially the counterpart to historiography’s ideological epitomator. The narrating subject’s point of view is closely connected to the notions of Roman fides, ius and mos, which are construed as opposites to Carthaginian unpredictability and primitivism. However, several of the central Roman protagonists also have problematic relationships to these notions and fail to comply with the expectations imposed on them as state officials. Above all, this becomes obvious through their defeat and death.

Hence, the pivotal conflict of the Punica is the struggle to yoke together epic heroism and historically oriented rationality. The parties of the conflict can be expressed as “the law”, as a metonym for the ideological demands of the Roman state, and “the hero” embodied by the generals that fight and die, but without recognition.

In four close readings, the characters Gracchus, Flaminius, Paulus and Marcellus are studied. They express in different ways the incompatibility between the heroic inner driving forces and the demands of usefulness and rationality associated with the Roman state.

This paper presents a brief account of major lexicalization patterns of breaking and cutting events (jointly called disconnection) in Swedish, based on a general model (meaning matrix) of the disconnection situation. The place of Swedish in a general lexical typology of disconnection is sketched on the basis of data from a couple of non-European languages and corpus-based contrastive analysis of a number of genetically and/or areally closely related languages.

This thesis investigates the vocabulary skills and the non-word repetition (NWR) performance of 99 typically developing (TD) 4­­–7-year-old Arabic-Swedish-speaking children and 11 Arabic-Swedish-speaking children with a diagnosis of Developmental Language Disorder (DLD). The children’s early language development, family backgrounds and language exposure patterns are explored through parental questionnaires, and for the DLD children also via interviews with parents, teachers and speech-language pathologists regarding their developmental history, language skills and communicative behaviour. Vocabulary comprehension and production is assessed with Cross-linguistic Lexical Tasks (CLT; Haman et al. 2015) in Arabic and Swedish. Phonological working memory is assessed with four different NWR tasks with varying item length, phonological complexity and language-likeness (Radeborg et al. 2006, Chiat 2015, Abou Melhem et al. 2011). For vocabulary, differences between the two languages (Arabic and Swedish) and differences between comprehension and production are explored, as well as effects of age, exposure and socioeconomic status (SES). For NWR, effects of age, task, item length and phonological complexity are investigated, as well as effects of vocabulary and exposure.

Results: Vocabulary comprehension and production scores were found to increase with age in both Arabic and Swedish. Daily language exposure predicted comprehension and production scores in Arabic, but only production scores in Swedish. Length of exposure to Swedish was the most important predictor of Swedish vocabulary scores. SES (parental education) did not predict vocabulary scores in either language. For NWR, scores increased with age on all tasks. There were also task and item effects. Factors related to NWR performance were type of task, item length, phonological complexity and vocabulary skills.

At group level, the DLD children scored below their TD peers on both vocabulary and NWR tasks. Many DLD children had particularly low vocabulary scores in their first language (Arabic), despite extensive and continuous exposure from birth. There was substantial overlap between the TD and the DLD groups on NWR performance, and not all DLD children scored low on NWR. Having a history of language delay or language difficulties in the family was more common among the DLD children than the TD children. The study underscores the importance of considering patterns of language exposure and developmental history when assessing the language skills of bilingual children with potential DLD.

As in many other countries, baseline data concerning the linguistic development of bilingual children in Sweden are lacking, and suitable methods for identifying developmental language disorder (DLD) in bilinguals are lacking as well. This study presents reference data from 108 typically developing (TD) Turkish/Swedish-speaking children aged 4;0–8;1, for a range of language tasks developed specifically for the assessment of bilinguals (LITMUS test battery, COST Action IS0804). We report on different types of nonword repetition (NWR) tasks (language-specific and language-independent), receptive and expressive vocabulary (Cross-Linguistic Lexical Tasks, CLTs), and narrative macrostructure comprehension and production (Multilingual Assessment Instrument for Narratives, MAIN) in Turkish, the children’s home language, and in Swedish, the language of schooling and society. Performance was investigated in relation to age, language exposure, type of task, and (for NWR and narratives) vocabulary size. There was a positive development with age for all tasks, but effects of language exposure and vocabulary size differed between tasks. Six bilingual Turkish/Swedish children with DLD were individually compared to the TD children. TD/DLD performance overlapped substantially, particularly for NWR, and more so for the production than the comprehension tasks. Surprisingly, the discriminatory potential was poor for both language-specific and language-independent NWR. DLD case studies underscored the importance of interpreting language scores in relation to exposure history, and the need for an increased emphasis on functional language skills as reported by parents and teachers when assessing and diagnosing DLD in bilinguals.

The Arabic-speaking community in Sweden is large and diverse, yet linguistic reference data are lacking for Arabic-Swedish-speaking children. This study presents reference data from 99 TD children aged 4;0-7;11 on receptive and expressive vocabulary in the minority and the majority language, as well as for three types of non-word repetition (NWR) tasks. Vocabulary scores were investigated in relation to age, language exposure, and socio-economic status (SES). NWR performance was explored in relation to age, type of task, item properties, language exposure, and vocabulary. Eleven children with DLD were compared to the TD group. Age and language exposure were important predictors of vocabulary scores in both languages, but SES did not affect vocabulary scores in any language. Age and vocabulary size had a positive effect on NWR accuracy, whilst increasing item length and presence of clusters had an adverse effect. There was substantial overlap between the TD and DLD children for both vocabulary and NWR performance. Diagnostic accuracy was at best suggestive for NWR; no task or type of item was better at separating the two groups. Reports from parents and teachers on devleopmental history, language expoosure, and functional language skills emerged as important facotrs for correctly identifying DLD in bilinguals.

The present study reports reference data for two nonword-repetition tasks for Swedish-speaking adults, and investigates effects of test-related factors (nonword type and length, consonant clusters). Twenty-three native speakers of Swedish (8 men, 15 women) aged 19 to 41 (M = 27.8 years) participated in two nonword repetition tasks, one language-specific with high syllable complexity (consonant clusters, both open and closed syllables) and one quasi-universal with low syllable complexity (no consonant clusters, only open syllables).The participants scored close to maximum on both tasks but significantly higher on the languages-specific test (M language-specific = 92%, M quasi-universal = 91%), despite the fact that the language-specific items had a higher degree of syllable complexity. Accuracy was significantly lower on longer items (items with 2–4 syllables ≈ 100%, items with five syllables ≈ 70%), but whether language-specific items contained consonant clusters had no effect. The results indicate that adults without a diagnosis of developmental language disorder or dyslexia can be expected to perform close to maximum on both tests. Future studies should investigate clinical groups and investigate effects of vocabulary on nonword repetition.

The thesis presents findings from five readability studies performed on mobile devices. The dynamic Rapid Serial Visual Presentation (RSVP) format has been enhanced with regard to linguistic adaptation and segmentation as well as eye movement modeling. The novel formats have been evaluated against other common presentation formats including Paging, Scrolling, and Leading in latin-square balanced repeated-measurement studies with 12-16 subjects. Apart from monitoring Reading speed, Comprehension, and Task load (NASA-TLX), Eye movement tracking has been used to learn more about how the text presentation affects reading.

The Page format generally offered best readability. Reading on a mobile phone decreased reading speed by 10% compared to reading on a Personal Digital Assistant (PDA), an interesting finding given that the display area of the mobile phone was 50% smaller. Scrolling, the most commonly used presentation format on mobile devices today, proved inferior to both Paging and RSVP. Leading, the most widely known dynamic format, caused very unnatural eye movements for reading. This seems to have increased task load, but not affected reading speed to a similar extent. The RSVP format displaying one word at time was found to reduce eye movements significantly, but contrary to common claims, this resulted in decreased reading speed and increased task load. In the last study, Predictive Text Presentation (PTP) was introduced. The format is based on RSVP and combines linguistic chunking and adaptation with eye movement modeling to achieve a reading experience that can rival traditional text presentation.

It is explained why readability on mobile devices is important, how it may be evaluated in an efficient and yet reliable manner, and PTP is pinpointed as the format with greatest potential for improvement. The methodology used in the evaluations and the shortcomings of the studies are discussed. Finally, a hyper-graeco-latin-square experimental design is proposed for future evaluations.

This paper explores how interfaces that fully uses our ability to communicate through the visual, auditory, and tactile senses, may enhance mobile interaction. The first step is to look beyond the desktop. We do not need to reinvent computing, but we need to see that mobile interaction does not benefit from desktop metaphors alone. The next step is to look at what we have at hand, and as we will see, mobile devices are already quite apt for multimodal interaction. The question is how we can coordinate information communicated through several senses in a way that enhances interaction. By mapping information over communication circuit, semiotic representation, and sense applied for interaction; a framework for multimodal interaction is outlined that can offer some guidance to integration. By exemplifying how a wide range of research prototypes fit into the framework today, it is shown how interfaces communicating through several modalities may enhance mobile interaction tomorrow.

Can readability on small screens be improved by using adaptive Rapid Serial Visual Presentation (RSVP) that adapts the presentation speed to the characteristics of the text instead of keeping it fixed? In this paper we introduce Adaptive RSVP, describe the design of a prototype on a mobile device, and report findings from a usability evaluation where the ability to read long and short texts was assessed. In a latin-square balanced repeated-measurement experiment, employing 16 subjects, two variants of Adaptive RSVP were benchmarked against Fixed RSVP and traditional text presentation. For short texts, all RSVP formats increased reading speed by 33% with no significant differences in comprehension or task load. For long texts, no differences were found in reading speed or comprehension, but all RSVP formats increased task load significantly. Nevertheless, Adaptive RSVP decreased task load ratings for most factors compared to Fixed RSVP. Causes, implications, and effects of these findings are discussed.

This thesis investigates the vocabulary and narrative macrostructure skills of 102 typically-developing (TD) 4- to 7-year-old Turkish-Swedish bilingual children (cross-sectional), the development of these skills over time from age 4 to 6 in a subgroup of 10 children (longitudinal), and six Turkish-Swedish children with a language impairment (LI) diagnosis (clinical). The children’s health, family and language backgrounds, their language use and input patterns are explored through parental questionnaires, family interviews, and interviews with teachers and speech-language pathologists. In both Turkish and Swedish, comprehension and production are assessed with comparable materials: Cross-Linguistic Lexical Tasks (CLT; Haman et al., 2015), and Multilingual Assessment Instrument for Narratives (MAIN; Gagarina et al., 2012). For vocabulary (CLT), age development, language differences (Turkish vs. Swedish), differences between comprehension and production, and effects of language use and input are explored. For narrative macrostructure (MAIN), age development, language differences and task differences (Cat/Dog vs. Baby Birds/Baby Goats) are analyzed. LI children’s scores are compared with TD children, with additional focus on the LI children’s communicative, linguistic and social behavior. 

In both vocabulary comprehension and production, the youngest TD groups performed better in Turkish than in Swedish, but by age 6, Turkish and Swedish vocabulary scores matched due to rapid improvement in Swedish. Factors related to vocabulary scores were: daily language input, parents’ language use with each other and with the child, and child’s language with the sibling(s). For narratives, comprehension was ahead of production. There was no difference between Turkish and Swedish MAIN comprehension, but for both languages a task effect was found (higher scores on Cat/Dog than Baby Birds/Baby Goats). Narrative production scores were generally low for both languages, but increased more with age in Swedish than in Turkish. The longitudinal study largely confirmed the patterns found in the cross-sectional data.

The majority of the LI children performed far below their TD peers in both their languages. Some LI children performed very low in only one language, despite extensive and long-term exposure to that language. In contrast, TD children with very low scores in one language usually had very limited exposure to that language. LI children were also reported to have difficulties with word learning, pragmatics, and attention, and a family history with language problems. It is suggested that bilingual children with potential language impairment should be assessed holistically in both their languages and extensive information about their family backgrounds and language input characteristics be collected.

Research into Automatic Speech Recognition (ASR), the task of transforming speech into text, remains highly relevant due to its countless applications in industry and academia. State-of-the-art ASR models are able to produce nearly perfect, sometimes referred to as human-like transcriptions; however, accurate ASR models are most often available only in high-resource languages. Furthermore, the vast majority of ASR models are monolingual, that is, only able to handle one language at a time. In this thesis, we extensively evaluate the quality of existing monolingual ASR models for Swedish, Danish, and Norwegian. In addition, we search for parallels between monolingual ASR models and the cognition of foreign languages in native speakers of these languages. Lastly, we extend the Swedish monolingual model to handle all three languages.

The research conducted in this thesis project is divided into two main sections, namely monolingual and multilingual models. In the former, we analyse and compare the performance of monolingual ASR models for Scandinavian languages in monolingual and cross-lingual settings. We compare these results against the levels of mutual intelligibility of Scandinavian languages in native speakers of Swedish, Danish, and Norwegian to see whether the monolingual models favour the same languages as native speakers. We also examine the performance of the monolingual models on the regional dialects of all three languages and perform qualitative analysis of the most common errors. As for multilingual models, we expand the most accurate monolingual ASR model to handle all three languages. To do so, we explore the most suitable settings via trial models. In addition, we propose an extension to the well-established Wav2Vec 2.0-CTC architecture by incorporating a language classification component. The extension enables the usage of language models, thus boosting the overall performance of the multilingual models.

The results reported in this thesis suggest that in a cross-lingual setting, monolingual ASR models for Scandinavian languages perform better on the languages that are easier to comprehend for native speakers. Furthermore, the addition of a statistical language model boosts the performance of ASR models in monolingual,  cross-lingual, and multilingual settings. ASR models appear to favour certain regional dialects, though the gap narrows in a multilingual setting. Contrary to our expectations, our multilingual model performs comparably with the monolingual Swedish ASR models and outperforms the Danish and Norwegian models.

The multilingual architecture proposed in this thesis project is fairly simple yet effective. With greater computational resources at hand, further extensions offered in the conclusions might improve the models further. 

We present the Uppsala University system for SemEval-2022 Task 1: Comparing Dictionaries and Word Embeddings (CODWOE). We explore the performance of multilingual reverse dictionaries as well as the possibility of utilizing annotated data in other languages to improve the quality of a reverse dictionary in the target language. We mainly focus on characterbased embeddings. In our main experiment, we train multilingual models by combining the training data from multiple languages. In an additional experiment, using resources beyond the shared task, we use the training data in Russian and French to improve the English reverse dictionary using unsupervised embeddings alignment and machine translation. The results show that multilingual models occasionally but not consistently can outperform the monolingual baselines. In addition, we demonstrate an improvement of an English reverse dictionary using translated entries from the Russian training data set.

We present an analysis of a number of coreference phenomena in English-Croatian human and machine translations. The aim is to shed light on the differences in the way these structurally different languages make use of discourse information and provide insights for discourse-aware machine translation system development. The phenomena are automatically identified in parallel data using annotation produced by parsers and word alignment tools, enabling us to pinpoint patterns of interest in both languages. We make the analysis more fine-grained by including three corpora pertaining to three different registers. In a second step, we create a test set with the challenging linguistic constructions and use it to evaluate the performance of three MT systems. We show that both SMT and NMT systems struggle with handling these discourse phenomena, even though NMT tends to perform somewhat better than SMT. By providing an overview of patterns frequently occurring in actual language use, as well as by pointing out the weaknesses of current MT systems that commonly mistranslate them, we hope to contribute to the effort of resolving the issue of discourse phenomena in MT applications.

Das Weltall, die Gesamtheit des in der Welt Vorhandenen, wird in den Brāhmaṇas gewöhnlich mit dem Ausdruck idaṃ sarvam ›das alles hier‹ bezeichnet…«, heißt es in der Doktorarbeit von Konrad Klaus Die altindische Kosmologie (1986). Die Vollendung seines 65. Lebensjahres – zugleich die Vollendung von zwei Jahrzehnten als Universitätsprofessor in Bonn – ist uns willkommener Anlass, Konrad Klaus mit der vorliegenden Festschrift zu ehren. »Das alles hier« darf gerne auch im Hinblick auf das bisherige Lebenswerk des Geehrten gedeutet werden: Ein reiches akademisches Wirken mit vielfachen Aktivitäten in Lehre, Forschung und Wissenschaftsmanagement mit einer großen Zahl brillanter Publikationen zu philologischen und kulturgeschichtlichen Fragestellungen als sichtbare Zeichen. Konrad Klaus hat einen würdigen Platz in der wissenschaftlichen Traditionslinie der Indologie, die in Deutschland institutionell mit der Einrichtung des ersten der indischen Philologie gewidmeten Lehrstuhls an der neugegründeten Universität Bonn 1818 begann. Es wäre verfehlt, zu glauben, die 200-Jahr-Feier 2018 sei eine Art vorgezogenes Begräbnis gewesen. Der Übergang zur Südasienkunde mit einem erneuerten Profil gehört mit zum Lebenswerk von Konrad Klaus, der, obwohl selbst klassischer Indologe, diese Neuausrichtung voll unterstützte und wohlwollend begleitete.