To Boldly Query What No One Has Annotated Before? The Frontiers of Corpus Querying

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6307–6321 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6307 To Boldly Query What No One Has Annotated Before? The Frontiers of Corpus Querying Markus G¨artner, Kerstin Jung Institute for Natural Language Processing University of Stuttgart {markus.gaertner,kerstin.jung}@ims.uni-stuttgart.de Abstract Corpus query systems exist to address the mul- tifarious information needs of any person inter- ested in the content of annotated corpora. In this role they play an important part in making those resources usable for a wider audience. Over the past decades, several such query sys- tems and languages have emerged, varying greatly in their expressiveness and technical details. This paper offers a broad overview of the history of corpora and corpus query tools. It focusses strongly on the query side and hints at exciting directions for future development. 1 Introduction Annotated corpora have always been the backbone for many fields in NLP and other disciplines related to linguistics. Whether serving as an invaluable source of empirical evidence for foundational re- search or doubling as gold-standard training input for fueling the furnaces of our machine learning fac- tories, their importance cannot be overemphasized. But especially for the empirically motivated user base, corpora are only ever as good as the means available to explore them. And the primary means of exploring linguistically annotated corpora have always been (dedicated) corpus query tools and corpus query languages in their manifold shapes. In this paper we intend to give a thorough chronology of the major interplay between corpus progression and query tool evolution, with a strong focus on the latter. We start with an overview on relevant aspects of corpora and how they changed over the past ~30 years in Section 2. Section 3 elaborates on the observable phases in query tool development. In Section 4 we discuss alternative corpus query approaches based on general pur- pose data(base) management solutions and provide pointers to related work in Section 5. Section 6 summarizes some of our observations and with Section 7 we finally hint at our vision for future directions in corpus query system development. 2 Once Upon a Corpus – Trends in Corpus Evolution Though corpus linguistics dates back further, ma- jor online catalogs such as those from LDC1 and ELRA2 list corpora starting from the early 1990s. In the following decades corpus trends have var- ied along several dimensions, both technical and content-related. This section discusses such fea- tures and gives examples for their evolution. Since this overview is an introduction to digital corpus query systems, we mainly focus on written and annotated corpora. With a focus on written corpora, character encoding is a decisive factor when estimating the publication date. Starting from plain ASCII (Everts, 20003, Graff and Cieri, 2003) and lan- guage/script specific encodings, such as ISO/IEC 8859 (Armstrong-Warwick et al., 1994; Federico et al., 2000), nowadays many corpora come with a (mostly) language independent UTF-8 encoding (Ion et al. (2012); Prasad et al. (2019) and compare …

Content truncated to 500 words