|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ПсковГУ |
||
Parametrizing number variation in Russian noun phrases with experimental studies and language modelingстатья
Статья опубликована в журнале из списка Web of Science и/или Scopus
- Автор: Студеникина К.А.
- Журнал: Научный результат. Вопросы теоретической и прикладной лингвистики
- Том: 9
- Номер: 1
- Год издания: 2023
- Издательство: Федеральное государственное автономное образовательное учреждение высшего образования "Белгородский государственный национальный исследовательский университет"
- Местоположение издательства: Белгород
- Первая страница: 192
- Последняя страница: 205
- DOI: 10.18413/2313-8912-2023-9-1-1-2
- Аннотация: The study deals with the problem of agreement with coordinated constructions and number form alternation. The target, agreeing with two conjoined singular nouns, copies either plural or singular number feature. This paper focuses on the syntax of Russian noun phrases with coordinated modifiers which demonstrate the number variation of the agreement controller. If two conjoined singular adjectives have split interpretations, both singular and plural nouns are acceptable. In contrast to previous studies relying on introspection and corpus, we parametrize the number variation based on the results of the self-paced acceptability experiments (Likert scale 1-7). We compare the data about human perception with the language probabilities predicted by a neural model for text generation ruGPT-3. Two case studies were conducted to analyze morphological and syntactic factors parametrizing variation. The first study considers the impact of noun morphology on number alternation. The second study examines the effect of the premodifier attributive agreement. Both offline acceptability scores and online reading time demonstrate that the observed morphological and syntactic factors should be considered while parametrizing number variation in Russian noun phrases with coordinated modifiers. The ruGPT-3 language model, trained on a vast collection of Russian texts, manages to predict the correct probability for highly acceptable and highly unacceptable sentences, but it fails to assign accurate probability values to the cases of variation.
- Добавил в систему: Студеникина Ксения Андреевна