When the FrameNet suit does not fit the Swedish language

A talk by Karin Friberg Heppin, Department of Swedish, University of Gothenburg

Mon, February 3, 2014 | BUR 337

3:30 PM - 4:30 PM

The Swedish FrameNet, SweFN, has over 900 frames, nearing the number of frames in the original Berkeley FrameNet, BFN. The construction of Swedish frames, equivalent to the English ones is sometimes easy, sometimes not. In this talk I will describe some of the difficulties we have encountered during the construction of SweFN.

The conceptual structures or frames, which make up the framenets, are to a degree language independent, making it possible for creators of new framenets to reuse the BFN frames, but populate them with language specific lexical units and annotated sentences. However, problems arise when language structures differ to the degree that equivalents cannot be found, or only near equivalents with perhaps very different distribution.

To give an example: Swedish, like German, is a language with very productive compounding, and compounds written as solid orthographic words. This affects the population of the framenet frames with lexical units. Whereas in English, most compounds are written with separated constituents that may populate frames individually, in Swedish, each compound, as a whole, constitutes one solid lexical unit. This entails that there are lexical units in BFN the Swedish equivalents of which only occur as parts of words: these can be, for example, relational adjectives such as nuclear (scientist) or criminal (lawyer), or adjectives resulting from conversions: working-class (hero). These all have Swedish noun modifiers as equivalents. In many cases when Swedish equivalents are of different grammatical categories they are unsuitable for the BNF frames. For example, the frame SOCIAL_DESIRABILITY is populated by adjectives only. Here we find the LUs upper-class and working-class whose Swedish equivalents are nouns describing categories of people, and do not belong in such a frame.

Other mismatch examples may be when English has a lexical unit for something which is expressed by a construction in Swedish, when there are many-to-many relations, not only between lexical units, but also between concepts. The list of mismatches can be made long and I will serve you some of them.

Bookmark and Share