n academic year 2023/24, our university collected over 200,000 student feedback responses evaluating teaching staff and course experiences. The survey included demographic data, 10 Likert scale questions on teaching quality, a question on student attendance, and three open-ended questions about student experiences. This paper explores the integration of Large Language Models (LLM) Gemini for sentiment analysis to evaluate students’ feedback quantitatively and qualitatively. We statistically analyze the Likert scale responses. To address the linguistic diversity of open-ended responses, written in both Cyrillic and Latin scripts with standard and slang expressions in several languages, we employed a preprocessing step using Gemini to standardize the input for further analyses. Sentiment analysis aims to identify various sentiment nuances, including direct answers, contradiction, multipolarity, mixed sentiment, sarcasm, irony, negation, ambiguity, understatement, and over-exaggeration. By comparing these insights with quantitative feedback, we aim to uncover deeper patterns between student perceptions and teaching performance. While the focus is on sentiment analysis, we also discuss the evaluation of the results provided by LLM. For the sentiments with less answers, the evaluation of GenAI was done manually. For the sentiments with more than 1000 entries, we suggest a semi-automated approach for sentiment categorization, to be explored in future work. This study enhances our understanding of student feedback through advanced computational methods, providing a more nuanced perspective on teaching quality and student satisfaction.

Understanding the relation between the meanings of words is an important part of comprehending natural language. Prior work has either focused on analysing lexical semantic relations in word embeddings or probing pretrained language models (PLMs), with some exceptions. Given the rarity of highly multilingual benchmarks, it is unclear to what extent PLMs capture relational knowledge and are able to transfer it across languages. To start addressing this question, we propose MultiLexBATS, a multilingual parallel dataset of lexical semantic relations adapted from BATS in 15 languages including low-resource languages, such as Bambara, Lithuanian, and Albanian. As experiment on cross-lingual transfer of relational knowledge, we test the PLMs’ ability to (1) capture analogies across languages, and (2) predict translation targets. We find considerable differences across relation types and languages with a clear preference for hypernymy and antonymy as well as romance languages.

Macedonian adjectives are inflected for gender, number, definiteness and degree, with in average 47.98 inflections per headword. The inflection paradigm of qualificative adjectives is even richer, embracing 56.27 morphophonemic alterations. Depending on the word they were derived from, more than 600 Macedonian adjectives have an identical headword and two different word forms for each grammatical category. While non-verbal adjectives alter the root before adding the inflectional suffixes, suffixes of verbal adjectives are added directly to the root. In parallel with the morphological differences, both types of adjectives have a different translation, depending on the category of the words they have been derived from. Nouns that collocate with these adjectives are mutually disjunctive, enabling the resolution of inflectional ambiguity. They are organised as a lexical taxonomy, created using hierarchical divisive clustering. If embedded in the future spell-checking applications, this taxonomy will significantly reduce the risk of forming incorrect inflections, which frequently occur in the daily news and more often in the advertisements and social media.

We introduce in this paper a generic approach to combine implicit crowdsourcing and language learning in order to mass-produce language resources (LRs) for any language for which a crowd of language learners can be involved. We present the approach by explaining its core paradigm that consists in pairing specific types of LRs with specific exercises, by detailing both its strengths and challenges, and by discussing how much these challenges have been addressed at present. Accordingly, we also report on on-going proof-of-concept efforts aiming at developing the first prototypical implementation of the approach in order to correct and extend an LR called ConceptNet based on the input crowdsourced from language learners. We then present an international network called the European Network for Combining Language Learning with Crowdsourcing Techniques (enetCollect) that provides the context to accelerate the implementation of this generic approach. Finally, we exemplify how it can be used in several language learning scenarios to produce a multitude of NLP resources and how it can therefore alleviate the long-standing NLP issue of the lack of LRs.