Creating bilingual dictionaries is very demanding in terms of human effort. From the compilation of the headword list to finding correct and relevant translations and providing information on how to use them in idiomatic target language sentences, lexicographic knowledge and experience are required. Although mono-and bilingual corpora are increasingly present in the process, human effort is still in the center and makes the creation of dictionaries a labor-intensive endeavor. This is particularly true when it comes to lesser-used language pairs, where finding competent bilingual speakers is already a challenge.The scope of the present thesis is to investigate to what extent state of the art language technology methods and resources can be exploited to help lexicographers construct bilingual dictionaries. As the main result of our research, we propose and evaluate a complete methodology to automatically produce "proto-dictionaries" for lesser-used language pairs, and show that not only our method presents economical advantages by reducing the amount of human effort needed, but it also addresses several methodological difficulties or inconsistencies with success. Moreover, a customizable dictionary query interface is presented whose features make the proto-dictionaries useful not only for lexicographers, but also for end users: Language learners as well as advanced users, such as professional translators.The methodology we propose relies on parallel corpora as a resource and exploits word alignment to produce translation candidates. We suggest that conditional probabilities, as estimated via the word alignment process, can be conceived as modelling translation relation between SL and TL units. We argue that, from a theoretical viewpoint, conditional probability is very well fit to quantify translation relation because of its ability to capture the gradual, asymmetrical nature of it.Moreover, the method implicitly produces a partition over the senses of the SL lemma by assigning different translation candidates to different contexts. This implicit feature addresses an omnipresent problem in lexicographic work: The subjective nature of word sense definitions and distinctions. We show via different word sense disambiguation experiments that the more we rely on contextual information, the less sense characterizations are prone to subjectivity.The second part of the thesis deals with the methodology of creating a proto-dictionary from a list of translation candidates obtained from the corpus. Proto-dictionaries result from a filtering of candidates based on three parameters: SL and TL lemma frequencies and the conditional probabilities. These parameters were set via a series of experiments and evaluations, at the end of which a Hungarian-Slovenian and a Hungarian-Lithuanian proto-dictionary were available. On top of that, we confirmed that our method allows to easily produce the reversed dictionaries, which gave us the Slovenian-Hungarian and the Lithuanian-Hungarian resources as well. Contextual information is also pro...