SEMI-COMPLEX SENTENCES WITH ING-FORMS IN MEDICAL DISCOURSE: CHALLENGES FOR NEURAL MACHINE TEANSLATION

The study is related to semi-complex sentences with ing-forms that are typical for the English language. Being an analytic language with very little inflection and strict word order, English has a wide range of constructions requiring mental operations for understanding their meanings. Nowadays with the increasing amount of texts being processed and translated using artificial intelligence, it is of particular interest to analyse how these sophisticated constructions function in the text and in which ways does the machine “understand” them.

The main objective of this study is to reveal the features of semi-complex sentences in particular (medical) discourse and identify some patterns (if there are any) of neural machine translation of these sentences from English into Russian.

Two types of methods are used in the study in order to achieve the goal: computational (Coh-Metrix tool, Neural Machine Translation) and manual (Analysis of Translation Errors). Coh-Metrix is a tool focused on linguistic features closely associated with deeper level of comprehension. It provides the text analysis on multiple characteristics and language-discourse levels. Neural Machine Translation (NMT) is one of the machine translation approaches which uses the large artificial neural network and deep learning models.

Five medical research articles on topics related to different spheres such as Lifestyle Medicine, Vaccination, Intensive Care, Medical Ethics and Speech Therapy were analysed, processed with Coh-Metrix tool, translated by PROMT neural machine translation system with medical profile and evaluated.

The study found that 1) semi-complex sentences with ing-forms (gerunds, participles and verbal nouns) are frequently used in medical research articles, contributing in diversity (but not always complication) of grammatical structures; 2) gerunds and verbal nouns are more typical for medical discourse due to prevalence of noun phrases over verb phrases; 3) neural machine translation system demonstrated certain patterns in translating ing-forms from English into Russian in some cases potentially causing mistakes; 4) a number of frequent word combinations have their own patterns in NMT translation presumably due to their existence in NMT translation memory 4) there is a necessity to improve the overall quality of the neural machine translation in terms of translating semi-complex sentences with ing-forms.

Keywords: semi-complex sentences, medical discourse, neural machine translation, non-finite verbs