Zoek in de site...

PhD theses

Below you can find an overview of PhD theses written by former PhD candidates of our research group.

Properties, Structures and Operations: Studies on language processing in the brain using computational linguistics and naturalistic stimuli.

Alessandro Lopopolo, 2021: Properties, Structures and Operations: Studies on language processing in the brain using computational linguistics and naturalistic stimuli

Context as Linguistic Bridges. SIKS Dissertation Series No. 2020-04. 198 pp.

Maarten van Gompel, 2020

Context as Linguistic Bridges’ is a studimage001y that focusses on the role of context information in machine translation, i.e. automated translation by computers. The underlying intuition is that the context in which a word or phrase appears is an important cue for the translation of that word or phrase. Consider, for example, the two different meanings of the word “bank” in the sentences “I put my money on the bank” and “The ship got stuck on the bank”. We developed classifier-based solutions that work well in Word Sense Disambiguation tasks like the above example, and integrate these in a Statistical Machine Translation system. Our main question is to find to what extent can we improve automated translation by explicitly modelling such context information.

Extracting Actionable Iformation from Microtexts. SIKS Dissertation Series No. 2019-17. ISBN 978-94-028-1540-5. 154pp.

Ali Hürriyetoğlu, 2019: Extracting actionable information from microtexts

Microblogs such as Twitterimage003 represent a powerful source of information. Part of this information can be aggregated beyond the level of individual posts. Some of this aggregated information is referring to events that could or should be acted upon in the interest of e-governance, public safety, or other levels of public interest. Moreover, a significant amount of this information, if aggregated, could complement existing information networks in a non-trivial way. This dissertation proposes a semi-automatic method for extracting actionable information that serves this purpose.

Speaker Recognition Systems in Forensic Conditions. The Calibration and Evaluation of the likelihood Ratio. 164 pp.

Miranti Indar Mandasaari, 2018: Speaker recognition systems in forensic conditions

This thesis contributes on evaluation image004and enhancement of a modern speaker recognition system such that it can be more usable in forensics. The speaker recognition system is a system developed at Radboud University Nijmegen (RUN) and based on the-state-of-the-art i-vector framework. Evaluation of the speaker recognition system was carried out thoroughly by taking into account aspects in both recognition and calibration performances. Duration and noise conditions are two forensic-motivated conditions investigated within the evaluation. A new approach in calibration is proposed in order to tackle the two aforementioned variabilities. The approach is called quality measure function (QMF) calibration. Here, quality measurements from speech, i.e., signal to noise ratio (SNR) and duration of active speech, are incorporated to the traditional linear calibration. The QMF calibration was then comprehensively evaluated. Results show that this proposed method results in improvement in recognition and calibration aspects. In addition, this thesis includes transfer techniques and approaches from speaker recognition field, where the concept of calibration is introduced to the face recognition community.

Modelling Patterns of Time and Emotion in Twitter. SIKS Dissertation Series No. 2017-11. 217 pp.

Florian Kunneman,2017: Modelling patterns of time and emotion in Twitter

As people come together to celebratimage005e, demonstrate or be entertained, they often leave traces about these events through communicating via online social media. The studies reported in thesis are aimed at the automatic identification and interpretation of such traces in the Dutch Twitter sphere. The outcomes are integrated into an information system that provides insight into real world events and into the contemplations of their crowds. The studies connect to existing work on detecting events, figurative speech or emotion in Twitter messages. The thesis covers new ground in analysing detected events for patterns of periodicity and for patterns of prior and subsequent emotion. Two prominent units of information in tweets are at the basis of all studies: time and hashtags.

Exploiting properties of the human auditory system and compressive sensing methods to increase noise robustness in ASR. SIKS Dissertation Series No. 2017-39. 208pp.

Sarah Amadi, 2017: Exploiting properties of the human auditory system and compressive sensing methods to increase noise robustness in ASR


Automatic speech recognition (ASR) image006refers to the process of converting acoustic features representing speech to a sequence of words. While state-of-the-art ASR systems have achieved high accuracies in quiet environments, their performance drops rapidly when the speech is corrupted by environmental noise, background music or competing speakers. Meanwhile, listening test experiments with noisy speech have shown that, even in the absence of context predictability, human listeners perform much more accurately than ASR systems. This indicates that the human auditory system is able to exploit more cues from the acoustic signals for accurately recognizing noisy speech. In this thesis, we explored different ways to incorporate knowledge about the human auditory system into the design of an ASR system, aiming at improved noise robustness.