Constructing EPTIC: A Modular Pipeline and an Evaluation of ASR for Verbatim Transcription
This paper presents a novel pipeline for constructing multimodal and multilingual parallel corpora, with a focus on evaluating state-of-the-art automatic speech recognition 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.
Dec 6, 2024
Expanding the European Parliament Translation and Interpreting Corpus: A Modular Pipeline for the Construction of Complex Corpora
The present paper introduces an expanded version of the European Parliament Translation and Interpreting Corpus (EPTIC), a multimodal parallel corpus comprising speeches delivered at the European Parliament along with their official interpretations and translations (see Bernardini et al., 2016; Bernardini et al., 2018). Constructing multimodal and parallel corpora for translation and interpreting studies (TIS) has been acknowledged as a “formidable task” (Bernardini et al., 2018), which – if automated, as we propose – involves a number of subtasks such as automatic speech recognition (ASR), multilingual sentence alignment, and forced alignment, each of which poses its own challenges. Yet tackling these subtasks also offers a unique way to evaluate state-of-the-art natural language processing (NLP) tools against a unique, multilingual benchmark. In this paper we discuss the development of a modular pipeline adaptable for each of these subtasks and address the broader implications of this work for the field of corpus construction.
Sep 15, 2024
A Corpus for Sentence-Level Subjectivity Detection on English News Articles
We develop novel annotation guidelines for sentence-level subjectivity detection, which are not limited to language-specific cues. We use our guidelines to collect NewsSD-ENG, a corpus of 638 objective and 411 subjective sentences extracted from English news articles on controversial topics. Our corpus paves the way for subjectivity detection in English and across other languages without relying on language-specific tools, such as lexicons or machine translation. We evaluate state-of-the-art multilingual transformer-based models on the task in mono-, multi-, and cross-language settings. For this purpose, we re-annotate an existing Italian corpus. We observe that models trained in the multilingual setting achieve the best performance on the task.
May 25, 2024
Decoding Medical Dramas: Identifying Isotopies through Multimodal Classification
Classifying audiovisual content using unimodal and multimodal transformer-based models. The study compares two classification strategies: a single multiclass classifier and a one-vs-the-rest approach, examining their performance in both unimodal and multimodal settings. Results show the multiclass multimodal approach achieves the best performance, with an F1 score of 0.723, outperforming the unimodal text-based one-vs-the-rest method.
Dec 23, 2023