Скачать или смотреть Bowern Chirila Lab (Yale Linguistics) LSA Presentation: Digital Archiving (Collections Review)

Here, we report on pilot research on the extent to which language collections in digital linguistic archives are discoverable, accessible, and functional. Using a test case of common tasks in phonetic and phonological documentation, we evaluate a random sample of collections and find substantial problems in all domains.
Linguists have been documenting languages with a view to continued language work from archival materials for many years (cf. Dobrin et al. 2007; Hammarström 2018; Broeder et al. 2011; Himmelmann 1998; McDonnell, Berez-Kroeker & Holton 2019; Henke 2016). Digital archives such as ELAR have been very concerned with models of data accessibility for academics and communities (cf. Nathan 2010). But archives distribute the data that linguists contribute, and if linguists see archiving as the endpoint of a detailed and complex process of data gathering (cf. Bowern 2008:47), not the starting point, issues relevant for future use could be overlooked. At the same time, digital documentation is now pervasive and has allowed the collection of larger and more varied corpora than could have been previously anticipated. However, no one, to our knowledge, has systematically investigated how depositors’ choices in structuring their collections might affect the usability of language resources (though cf. Dipper et al. nd; Innes 2010; Burke et al. 2021; Dobrin et al. 2007).
In this talk, we report on the results of a preliminary review of a random sample of 20 publicly accessible collections from prominent digital archives: ELAR, CLA, PARADISEC, AILLA, and PanGloss. We evaluate the collection’s utility and suitability for different types of tasks, especially phonetic/phonological documentation. Our test was simple: download the files, convert transcripts to Praat TextGrids, and run the Montreal Forced Aligner (McAuliffe et al. 2017) to produce phoneme-level segmentation. We also evaluated the collections for ease of use (e.g. for how discoverable information is, how collections’ structure facilitated or impeded work). We performed these tests and noted any issues encountered in the workflow.