Graduate School of Language Technology in Finland

Kieliteknologian valtakunnallinen tutkijakoulu - Språkteknologiska forskarskolan i Finland

Students

At the KIT Graduate School there are currently five graduate students funded by the Ministry of Education, and some students with other funding arrangements.

Lili Aunimo

Organization: University of Helsinki, Department of Computer Science
Supervisor: Helena Ahonen-Myka
Research topic: Computational Methods for Semantic Representation of Text
Description: The goal of my research is to develop computational methods for representing the contents of text. The research has two parts. The first part consists of applying suitable data mining and machine learning techniques to induce a model that represents the contents of a text. The second part consists of applying a suitable knowledge representation framework in representing the semantics of text in a form easily accessed by a natural language processing application.
Financing: Ministry of Education
Links: Homepage, Doremi Research group

Tom Bäckström

D.Sc. (tech), graduated in April 2004.

Organization: Helsinki University of Technology, Laboratory of Acoustics and Audio Signal Processing
Supervisor: Paavo Alku
Research topic: Spectral models of speech and parametrization of the glottal source
Description: The research is concentrated around two topics. Firstly, we investigate constrained linear predictive models for modeling voiced speech sounds in the spectral domain. The goal is to develop models that more effectively allow inclusion of a priori knowledge while maintaining stability of the model. Secondly, we study different parameters of the glottal source that describe effects of vocal loading. The aim is to gain better understanding of the speech production system as well as develop tools for phoniatricians for clinical analysis of the voice.
Financing: Ministry of Education
Links: Homepage

Mathias Creutz

Organization: Helsinki University of Technology, Laboratory of Computer and Information Science, Neural Networks Research Centre
Supervisors: Krista Lagus, Mikko Kurimo
Research topic: Unsupervised learning of language models
Description: The goal is to study how unsupervised learning methods developed at the Neural Networks Research Centre can be applied to language modelling. The focus is on linguistically motivated solutions that are applicable to several languages. The performance of resulting language models shall be evaluated using speech recognition as a testbench.
Financing: Ministry of Education
Links: Homepage

Osmo Eerola

Organization: University of Turku
Supervisor: Olli Aaltonen
Research topic: To be determined
Financing: Independent

Kimmo Kettunen

Organization: University of Tampere
Supervisor: Kalervo Järvelin
Research topic: Suomen kielen vartalo-ohjelmaperhe tiedonhaun apuna
Financing: Independent

Juha-Pertti Laaksonen

Organization: University of Turku, Department of Phonetics
Supervisors: Olli Aaltonen (Dept. of Phonetics, Univ. of Turku); Risto-Pekka Happonen (Dept. of Oral and Maxillofacial Surgery, Univ. of Turku)
Research topic: Effects of Oral and Maxillofacial Surgical Operations on Speech Acoustics
Description: The purpose of the research is to find out the effects of operations used in oral and maxillofacial surgery on acoustic quality of speech by analyzing the main acoustic features of speech. By studying the effects of distorted control mechanisms (e.g., nerve impairments) and altered vocal tract configurations (e.g., modifications of oral structures) on speech acoustics, I try to get information concerning the control mechanisms of speech production, and the relationship between acoustics and articulation. Also the different theories of speech production are discussed. In addition, the results can be applied to development of speech production modeling.
Financing: Independent
Links: Homepage

Mietta Lennes

Organization: University of Helsinki, Department of Phonetics
Supervisors: Antti Iivonen, Stefan Werner, Martti Vainio
Research topic: Phonetic variability of speech sounds in Finnish informal speech
Description: The goal of my research is to model the phonetic variability of speech sounds in Finnish informal speech. A method will be developed for the description of the amount of articulatory effort used by the speaker. Relationships between prosodic properties and the expected acoustic-phonetic quality of the corresponding speech sounds will be studied. Artificial learning methods will be used to test the model.
Financing: Independent
Links: Homepage

Krister Lindén

Organizations: Helsinki University, Department of General Linguistics;
Helsinki University of Technology, Laboratory of Computer and Information Science, Neural Networks Research Centre;
Tampere University, Department of Information Studies
Supervisors: Lauri Carlson (HU), Krista Lagus (HUT), Kalervo Järvelin (TaU)
Research topic: On Word Sense Disambiguation: Using self-organizing document maps as semantic components in language technology applications
Description: Word sense disambiguation is the task of selecting the most appropriate word meanings or senses in a given context. Self-organizing maps are an efficient way to cluster, visualize and abstract data. When a self-organizing document map is provided with information about words in context a clustering based on similar contexts emerges. The goal is to find methods for exploiting this clustering for word sense disambiguation in language technology applications.
Financing: Ministry of Education
Links: Homepage

Sinikka Loikkanen

Organization: University of Helsinki
Supervisors: Kimmo Koskenniemi, Lauri Carlson
Research topic: Automaattisest ja puoliautomaattiset menetelmät morfologisen kuvauksen päättelemiseksi kieliopillisesti koodatusta aineistosta
Financing: Independent

Stina Ojala

Organization: University of Turku, Department of Phonetics
Supervisors: Olli Aaltonen, Department of Phonetics, University of Turku; Ritva Takkinen, Department of Languages, University of Jyväskylä
Research topic: The handshapes of the Finnish Sign Language
Description: The topic of this study is the handshapes of the Finnish Sign Language. The research is conducted by means of optical phonetics. The purpose of this study is to define and clarify the processes involved in the production and perception of sign language.
The bases of the present research are the handshapes produced by native signers. These handshapes are observed from digital video samples and then are processed further into motion frame -syntheses. The data is then presented as stimuli in behavioral and psychophysiological tests.
The results of this study are used to model both the production and perception models of sign language. In addition, the results can establish a basis for the coding of movement envelopes into a computer database and in the future in the design of a virtual signer.
Financing: The Alfred Kordelin Foundation, The Finnish Konkordia Fund
Links: Homepage

Jussi Piitulainen

Organization: University of Helsinki
Supervisors: Kimmo Koskenniemi, Lauri Carlson
Research topic: Sanojen distributionaalisen samanlaisuuden suhde niiden merkityksen samanlaisuuteen
Financing: Independent

Harri Saarikoski

Organizations: Helsinki University (General Linguistics) and AAC Global Oy
Supervisors: Lauri Carlson (Helsinki University, General Linguistics) and Pasi Tyrväinen (Jyväskylä University, Computer Sciences)
Research topic: Requirement Specification for Hybrid WSD Component in Demanding NLP Applications
Description: I aim to compile a requirement specification for the Word Sense Disambiguation (WSD) component. My key contribution will be in the design and acquisition of a WSD-enabling central knowledge base (RDF family), as well as in defining the optimal balance between the various WSD knowledge sources and disambiguation tools (ie. knowledge-driven vs computational methods). To add to the feasibility of the results, I will simultaneously look for ways to automate the acquisition of this knowledge base and ways to represent the knowledge in a unified and computationally efficient way. The aim is to find real cures to the WSD bottleneck that stubbornly ails the huge promise of the NLP field. WSD makes possible an astounding variety of demanding industrial NLP applications (such as machine translation, ontology autogeneration and automatic annotation).
Financing: Independent
Links: Homepage
ONTO project for Finnish Ontologies in the Semantic Web (launching in Fall 2003)

Janne Savela

Organization: University of Turku, Department of Phonetics
Supervisor: Olli Aaltonen
Research topic: Attentive and pre-attentive processing of vowels
Description: In my research the pre-attentive and attentive processing of vowel phonemes are discussed. Firstly, the vowel identification data will be analysed. The multilingual aspects of vowel labeling will be discussed. Secondly, the pre-attentive processing of vowel phonemes are discussed.For example, he effects of distinctive features on speech perception are studied.
Financing: Ministry of Education
Links: Homepage

Vesa Siivola

Organization: Helsinki University of Technology, Laboratory of Computer and Information Science
Supervisor: Mikko Kurimo
Research topic: Language modeling for speech recognition
Description: Language modeling is an essential component of a modern speech recognizer. Usually the task is simply to estimate the probabilities for the next word given the already recognized speech. The goal of my work is to improve the performance of this part of the recognizer.
Financing: Independent
Links: Homepage

Janne Tynkkynen

Organization: University of Helsinki
Supervisors: Kimmo Koskenniemi, Lauri Carlson
Research topic: Hybrid Tagging in a Multilingual Environment
Financing: Independent

Anssi Yli-Jyrä

Organization: Helsinki University, Department of General Linguistics; Nordic Language Technology Programme, NorFA
Supervisors: Kimmo Koskenniemi, Lauri Carlson
Research topic: Efficient Parsing with Finite-State Constraint Satisfaction
Description: Finite-state automata are efficient, and therefore widely employed in human language technology. Kimmo Koskenniemi (1990) has proposed an appealing parsing approach that uses finite-state automata. The current implementations of the approach are very slow although the system runs in an asymptotic linear time. My goal is to find out how the parser's implementation can be improved on the basis of comparison to other parsing approaches and introducing new finite-state methods that improve the flexibility of the data-structures.
Financing: NorFA - Nordic Academy for Advanced Study
Links: Homepage

back to top top
Last updated: