Remarkable Similarity of Clausal Coordinate Ellipsis in Russian Compared to Dutch, Estonian, German, and Hungarian

Elliptical constructions can help to avoid repetition of identical constituents during natural-language generation. From grammar books, it is not easy to extract executable rules for ellipsis—in our case in Russian. Therefore we follow a different strategy. We test the accuracy of a rule set that has been evaluated for the two Germanic languages, Dutch and German, and the two Finno-Ugric languages, Estonian and Hungarian. For a Russian test corpus of about 100 syntactically annotated coordinated sentences that systematically vary the conditions of rule application, our Java program can automatically produce all elliptical variants. Overand undergeneration in the resulting lists have been tested in two experiments with native speakers. Basically, the rules work very well for Russian. Within the four target languages, Russian works best with the Estonian amendments. Here we report two slight deviations partially known from the linguistic literature.