Issues in Data Attribution

Transcription

1 Issues in Data Attribution Gary Holton Alaska Native Language Archive University of Hawai i at Mānoa

2 Attribution to refer to the practice of valuing data sets created by other researchers. For the most part discussions of data citation focus on the practice of referencing data collected by the researcher or by a member of her research team. To inspire reproducible linguistic research we must also facilitate citation of data collected by other researchers. Attribution of others research is key to the process, but in general data sets are not valued as highly as other forms of linguistic publications for the purposes of peer review. Thus, attribution is crucially about giving credit for the work of preparing, archiving, and making available linguistic data sets.

3 attribution citation

4 Reproducible research Attribution is key to inspiring reproducible linguistic research preparing, archiving, and making available linguistic data sets involves a significant amount of work if we don t give people credit for this work then they are less likely to make that data available in our current model data sets are not valued as highly as other forms of linguistic publications for the purposes of peer review.

5 Data has scholarly merit shift in practice has broadened the range of scholarly work to include... archives of primary data, electronic databases, corpora... (LSA Resolution Recognizing the Scholarly Merit of Language Documentation, May 2011) the Australian Research Council (ARC) made it clear that the ARC accepts that curated corpora can legitimately be seen as research output (Margetts et al., to appear)

6 Existing solutions Proxy publications create citable publication which describes the collection Co-authorship include data providers as co-authors

7 Proxy publication Common solution for software tools MrBayes (Ronquist and Huelsenbeck 2003) Ronquist, F. & J.P. Huelsenbeck MRBAYES 3: Bayesian phylogenetic inference under mixed models. 19:. Bioinformatics 19: Collection guides published article providing overview of archival data, becomes citable object (Woodbury 2014) May not sufficiently reflect the work put into creation of the archival data set

8 Co-authorship A blunt instrument doesn't distinguish between contributing data and actually collaborating on the analysis not very fine-grained. someone who collects a little data on lots of languages may be overly cited doesn't scale well collaborative projects will require numerous coauthors

9 Additional solutions Review article peer review of archival data as if it were a publication forthcoming in Language Documentation & Conservation Data paper create venues for archival publication (Haspelmath & Michaelis 2014) Peer review of archival data establish bodies to review deposits (Margetts et al., to appear)

10 Archiving as publication Problem: it is not only well-annotated data which needs to be recognized but also raw and less annotated primary data (cf. Margetts et al., to appear) Publication model might encourage cherry-picking of datasets, as linguists choose only the most richlyannotated or prettiest data for inclusion in the archival record Haspelmath & Michaelis 2014 explicitly emphasize selectivity

11 Athabascan (Dene) astronomy For more than 150 years the Gwich in [gwi] term yahdii was reported in published literature as denoting the Big Dipper Yet the archival record reveals that this constellation is actually much larger

12 Athabascan (Dene) astronomy Yahdii cover all the sky Interview with Frank John Ginnis, Fort Yukon, May 29, Recorded by Craig Mishler. Tape 6, side B. Alaska Native Language Archive (KU972Mi1998). Fairbanks. Jules Jetté (1900) also hints at a similar whole-sky constellation in Koyukon, though this reference is not found in the later edition of the dictionary (Jones & Jetté 2000) [Ahtna field notes.] De Laguna and McClellan (1960) report similar constellation based on body-part terms in Ahtna (AT954DM1960, Fairbanks: ANLA)

13 Athabascan (Dene) astronomy Mega-constellation Yahdii, with body-part asterisms (Cannon 2014) Could not have been recovered without archival data right foot tail left foot knife body right hand left ear right ear left hand eyes snout

14 Peer review of archival data Requires development of new infrastructure Allows the establishment of metrics specifically relevant to archival data quality of the repository accessibility of the data (cf. Callaghan 2015) format of the data data richness / amount of annotation diversity of content / representativeness uniqueness of contribution

15 References Callaghan, S. (2015). Data without Peer: Examples of Data Peer Review in the Earth Sciences. D-Lib Magazine, 21(1/2). Cannon, Chris Alaska Athabascan Stellar Astronomy. MA Thesis, U of Alaska Fairbanks. Haspelmath, Martin and Suzanne Maria Michaelis Annotated corpora of small languages as refereed publications: a vision. Diversity Linguistics Comment [Internet], posted March 11, Online: Margetts, Anna, Nick Thieberger, Stephen Morey, and Simon Musgrave. to appear. Assessing annotated corpora as research output. Australian Journal of Linguistics. Woodbury, Anthony