This paper presents a novel pipeline for constructing multimodal and multilingual parallel corpora, with a focus on evaluating state-of-the-art ASR tools for verbatim transcription. Our findings indicate that current technologies can streamline corpus construction, with fine-tuning showing promising results in terms of transcription quality compared to out-of-the-box Whisper models. The lowest overall WER achieved for English was 0.180, using a fine-tuned Whisper-small model. As for Italian, the fine-tuned Whisper-small model obtained a lower WER of 0.201 compared to the baseline Whisper-small’s WER of 0.219. While limitations remain, the updated pipeline is expected to drastically reduce the human efforts involved.

In the domain of cuisine, both dishes and ingredients tend to be heavily rooted in the local context they belong to. As a result, the associated terms are often realia tied to specific cultures and languages. This causes difficulties for non-speakers of the local language and ma- chine translation (MT) systems alike, as it implies a lack of the concept and/or of a plausible translation. MT typically opts for one of two alternatives: keeping the source language terms untranslated or relying on a hyperonym/near-synonym in the target language, provided one exists. !Translate proposes a better alternative: explaining. Given a cuisine entry such as a restaurant menu item, we identify culture-specific terms and enrich the output of the MT system with automatically retrieved definitions of the non-translatable terms in the target language, making the translation more actionable for the final user.

Cross-Level Semantic Similarity (CLSS) is a measure of the level of semantic overlap between texts of different lengths. Although this problem was formulated almost a decade ago, research on it has been sparse, and limited exclusively to the English language. In this paper, we present the first CLSS dataset in another language, in the form of CLSS.news.sr – a corpus of 1000 phrase-sentence and 1000 sentence-paragraph newswire text pairs in Serbian, manually annotated with fine-grained semantic similarity scores using a 0–4 similarity scale. We describe the methodology of data collection and annotation, and compare the resulting corpus to its preexisting counterpart in English, SemEval CLSS, following up with a preliminary linguistic analysis of the newly created dataset. State-of-the-art pre-trained language models are then fine-tuned and evaluated on the CLSS task in Serbian using the produced data, and their settings and results are discussed. The CLSS.news.sr corpus and the guidelines used in its creation are made publicly available.