A technique for translating computational problems from a natural language into the NUT language was worked on. The particular natural language considered was Estonian and the realm of problems targeted was elementary probability theory.
Of various competing semantic formalisms, Fillmore's deep case frames turned out to be the most well-suited one for representation of computational problems.
The idea of Fillmore's formalism is to classify affirmative sentences according to the kind of action, process or state they describe (e.g.\ BREAK). Sentences in any given class can be regarded as made up of phrases with certain conceptual roles (e.g. AGENT, OBJECT, INSTRUMENT, LOCATION). These are called deep cases. Each class is characterized by a deep case frame, i.e. a frame whose attributes are deep cases. The meaning of sentences in a class can be given by instantiating the corresponding deep case frame.
Three classes of problems from probability theory were chosen for consideration. For each of these, a deep case frame was formulated. Also, an algorithm was developed for transforming problem texts into instances of these frames. From frames, one can easily derive classes in the NUT language, and frame filling then becomes object creation. The diploma thesis (ref) is a report of this research.
The challenges of natural language processing from a wider angle of view (general-purpose models and techniques) were started studying and this research is intended to be continued as a master's thesis project under the supervision of Jaan Penjam.