Nothing will tear us apart? Uncovering grammatical features of Polish interpretese with supervised learning models

Abstract

In this paper, we applied a data-driven approach to investigate unique grammatical characteristics of Polish interpreter discourse (interpretese) in the comparable Polish Interpreting Corpus (PINC). Following, in part, the previously adopted methodology, we first applied the Boruta feature selection method (Kursa & Rudnicki, Citation2010) to retrieve a set of dependency POS bigrams that allow differentiating between interpreted and non-interpreted language best. Then, we trained a random forest model to establish the accuracy with which the selected features allow the model to tell interpreted and non-interpreted Polish discourse apart. Finally, we applied Multidimensional analysis to group the key features into meaningful dimensions. We found that Polish interpretese exhibits a distinct grammatical profile compared to non-interpreted Polish. Polish interpretese is very verbal and relies heavily on overt linking devices, particularly subordinating conjunctions and prepositional phrases introducing pronouns. This tendency is more pronounced in interpretations of impromptu speeches, which also exhibit greater use of pronouns and personal verb forms, suggesting a more involved and context-dependent style. The study suggests that the mode of delivery of the source speech impacts the syntactic features of both the original speeches and interpretations.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 20 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.