Marie Kopřivová

New Spoken Corpora of Czech: ORTOFON and DIALEKT

Abstract The paper introduces the ORTOFON corpus of spontaneous spoken Czech and the DIALEKT corpus of Czech dialects, their design principles and practical solutions adopted during data collection.

Eye of a Needle in a Haystack

We propose a multidimensional taxonomy of multiword expressions (MWEs) as a pattern applicable to entries in a representative lexicon of Czech MWEs. The taxonomy and the lexicon are useful for many reasons concerning lexicography, teaching Czech as a foreign language, and theoretical issues of MWEs as entities standing between lexicon and grammar, as well as for NLP tasks such as tagging and parsing, identification and search of MWEs, or word sense and semantic disambiguation. In addition to the description of various types of idiomaticity, the taxonomy and the lexicon are designed to account for flexibility in morphology and word order, syntactic and lexical variants and even creatively used fragments.

Contribution Towards a Corpus-Based Phraseology Minimum

This paper represents an attempt to put together a list of the most commonly used (most typical) Czech idioms using corpus data with annotated collocations. Collocations are annotated in corpora of contemporary written Czech as well as in a corpus of spoken Czech containing transcripts of intimate conversations. Idioms are selected based on their frequency in different text types (newspapers and magazines, non-fiction, fiction, spoken language) and the resulting list is compiled based on a criterion of occurrence of the given idiom in at least two different text types. A short characteristic of the individual text types is given in terms of which types of idioms are typical for them (according to formal criteria). This study confirms a substantial divide between idiom use in written and spoken language. A smaller difference can be observed between fiction on the one hand and non-fiction and newspapers on the other. The main reason for this is the interactive nature of fiction texts, which leads to them containing idioms with verbal components. These are employed in a fashion similar to spoken languages, in interactions among the individual characters. By contrast, non-fiction and journalistic language tends to be more descriptive, with more nominal idioms.

Experimental Tagging of the ORAL Series Corpora: Insights on Using a Stochastic Tagger

The ORAL series corpora of spontaneous spoken Czech currently contain neither lemmatization nor part of speech tagging. The main reason for this is that readily available NLP tools, designed primarily with written texts in mind, underperform when applied directly to speech transcripts, due to various morpohological and syntactic specificities of informal spoken language and the ways these are captured in transcription. Recently, the highly optimized open-source MorphoDiTa toolchain for training and applying stochastic tagging models was released; MorphoDiTa makes it easy and fast to experiment with incremental changes in the training procedure. The article discusses modifications to the morphological dictionary and training data used by the models which are necessary in order to improve their performance on the ORAL series corpora, as well as challenges which remain to be solved.