Component analysis of adjectives in Luxembourgish for detecting sentiments.
Joshgun Sirajzade and Daniela Gierschek

Among different linguistic means, parts of speech and their structure play a very important role in expressing sentiments. Especially adjectives and verbs seem to be crucial in transmitting sentiments, whereas other parts of speech such as prepositions do not carry any information regarding sentiments. In this talk, we will mainly focus on adjectives and their role in conveying sentiment in Luxembourgish. It is a report about the analysis of opinion bearing adjectives regarding their morphological structure. We analyze prefixes like on-, mëss-, ur- and suffixes like -lech, -eg, -los, -haft, bar and -esch in adjectives like onangenehm [unpleasant], ural [ancient], wonnerbar [wonderful] or witzeg [funny]. For this purpose, we utilize a corpus of news articles and user comments written in Luxembourgish. While adjectives are crucial for expressing sentiment, many other features also contribute to those utterances. The features explored here are therefore do not articulate sentiment alone. Syntactical negation like net [not] or keen [nobody, no], for instance, can completely reverse the meaning of an adjective. They are not in the scope of this talk, but are still important for sentiment detection.
We argue that a hybrid system based on linguistic-based rules and machine learning techniques can deliver the best performance for sentiment analysis on Luxembourgish texts. For this reason, it is crucial to investigate language-dependent linguistic structures in order to design good rules and learn successful features. The lexical and morphological structure of adjectives can certainly contribute to this task.