etSKELL – Estonian corpus for SKELL

etSKELL: Estonian corpus for SKELL

The Estonian corpus for Learners 2020 (etSKELL) is an Estonian corpus made up of sentences collected from the Estonian NC 2019 corpus, the Estonian Collocation Dictionary, and the Estonian Trends corpus. It was specially built up in order to provide the best sentence examples using GDEX system (see more below). The corpus is created in cooperation with the Institute of the Estonian Language using a GDEX configuration for Estonian developed by Kristina Koppel.

SKELL

SKELL is an abbreviation of Sketch Engine for Language Learning. It is a freely available web interface suitable for Estonian learning.

Part-of-speech tagset

The etSKELL corpus was tagged by PoS tagger developed at the University of Tartu using the estNLTK tools Filosoft part-of-speech tagset.

Good sentence examples (GDEX)

The Estonian corpus for Learners 2020 (etSKELL) consists of separate sentences which were sorted according to their quality. This quality is computed by GDEX system that adds a score to each sentence. The score is mainly based on the sentence length (minimum and maximum length) and the word frequency of particular words which occur in the sentence. The sentences are sorted in the way that the sentences with the highest score are displayed as the first results of a concordance.

What is specific for Estonian GDEX configuration compared to other languages, is that shorter sentences and sentences with wider optimal intervals are preferred. Instead of penalizing long words, the Estonian configuration does not allow words longer than 20 characters. Estonian configuration also penalizes sentences containing more than one adverb, one pronoun, one proper name, one numeral, one conjunction, one comma, or two verbs. There is a classifier that only selects sentences containing a verb and a classifier that penalizes certain non-finite constructions.