An Inclusive Notion of Text

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 11311–11327 July 9-14, 2023 ©2023 Association for Computational Linguistics An Inclusive Notion of Text Ilia Kuznetsov and Iryna Gurevych Ubiquitous Knowledge Processing Lab (UKP-TUDA) Department of Computer Science and Hessian Center for AI (hessian.AI) Technical University of Darmstadt ukp.informatik.tu-darmstadt.de Abstract Natural language processing (NLP) researchers develop models of grammar, meaning and com- munication based on written text. Due to task and data differences, what is considered text can vary substantially across studies. A con- ceptual framework for systematically captur- ing these differences is lacking. We argue that clarity on the notion of text is crucial for repro- ducible and generalizable NLP. Towards that goal, we propose common terminology to dis- cuss the production and transformation of tex- tual data, and introduce a two-tier taxonomy of linguistic and non-linguistic elements that are available in textual sources and can be used in NLP modeling. We apply this taxonomy to survey existing work that extends the notion of text beyond the conservative language-centered view. We outline key desiderata and challenges of the emerging inclusive approach to text in NLP, and suggest community-level reporting as a crucial next step to consolidate the discussion. 1 Introduction Text is the core object of analysis in NLP. An- notated textual corpora exemplify NLP tasks and serve for training and evaluation of task-specific models, and massive unlabeled collections of texts enable general language model pre-training. To a large extent, natural language processing today is synonymous to text processing. But what belongs to text? More broadly, what information should be captured in NLP corpora and be available to the models during training and inference? Despite its central role, the notion of text in NLP is vague: while earlier work mostly focused on grammatical phenomena and implicitly limited text to written language, the applied NLP of the past years increasingly takes an inclusive approach to text by introducing non-linguistic ele- ments into the analysis. Extensions vary from incor- porating emojis to exploiting document structure and cross-document relationships, and apply to all Figure 1: The same textual document (a) can be seen in many ways (b-d) depending on the assumed notion of text: while a syntax researcher might focus on written language (b), a summarization system can use docu- ment structure (c), and multimodal applications might use non-linguistic elements like tables (d). Systemat- ically capturing the differences between the assumed notions of text (top) requires a taxonomy of inclusive approaches to text. Such taxonomy is currently lacking. major components of the modern NLP infrastruc- ture, including unlabeled text collections (Lo et al., 2020), language models (Aghajanyan et al., 2021) and annotated corpora (Kuznetsov et al., 2022). The assumption that text in NLP solely refers to written language no longer holds. Yet, as Figure 1 illustrates, a systematic approach to capturing the differences between the assumed notions of text is lacking. This is problematic for several reasons. From the reproducibility perspective, machine learning as- sumes …

Content truncated to 500 words