Model-Based Inflection Error Diagnosis in Language Learning Systems, Graph-Based and Distributional Models for Multilingual Meaning Representation, and the Use of Annotation in Social Research and the Humanities for Analyses on Health

The presentation will be tripartite: I will talk about my Bachelor thesis, my Master thesis and the project I have been working in for the last three years.

My Bachelor thesis dealt with modelling inflection in human languages in a way that only represents knowledge about their correctness, but nevertheless allows for the diagnosis (detection, explanation and correction) of inflection errors in the context of language learning systems, where language learners have to generate or analyze word forms. Besides providing an adequate representation of the inflection process, this would allow a system to automatically provide a language learner with much more useful information than just ‘this is wrong’, without the system designer having to anticipate possible errors by the user.

My Master thesis was based on the phenomenon that when using a natural language as a pivot for machine translation (like Google Translator presumably does, using English), polysemous words in the pivot language that are not polysemous in the source language, nor in the target language, can lead to incorrect translations. For example, Google Translator translates the German noun{Graf}’ to the Spanish verb {contar}’ (which would be ‘zählen’ in German), although there is no meaning correspondence between the two words.

In the project I have been working in for the last three years, hermA, scholars from the humanities (literature studies) and social sciences (nursing science, cultural anthropology) explored the potential of automatic text processing for supporting their research processes. I will give an overview of the project setup and introduce one or two exemplary use cases where their research has benefitted from automation.

In der Regel sind die Vorträge Teil von Lehrveranstaltungsreihen der Universität Bremen und nicht frei zugänglich. Bei Interesse wird um Rücksprache mit dem Sekretariat unter sek-ric(at)dfki.de gebeten.

zuletzt geändert am 30.07.2019
nach oben