• Open problems in computational diversity linguistics

    Author(s):
    Johann-Mattis List (see profile)
    Date:
    2019
    Group(s):
    Classical Philology and Linguistics, Digital Humanists, Digital Humanities East Asia, Linguistics
    Subject(s):
    Historical linguistics, Computational linguistics
    Item Type:
    Presentation
    Meeting Title:
    International Conference of Historical Linguisitcs
    Meeting Org.:
    Australian National University
    Meeting Loc.:
    Canberra
    Meeting Date:
    2019-07-01/2019-07-05
    Tag(s):
    computer-assisted language comparison, open problems
    Permanent URL:
    http://dx.doi.org/10.17613/93t2-mt53
    Abstract:
    Despite a period of almost two decades in which quantitative approaches in historical linguistics have been increasingly used, gaining constantly more popularity even among predominantly qualitatively oriented linguists, we find many problems in the field of computational historical linguistics, which have only sporadically been addressed. In the talk, I will present a previously published list of 10 problems I personally deem important for historical linguistics, quickly explaining why I think that these problems are not solved yet, why they are hard to solve, but why I have confidence that they might be solved in the nearer or farer future. I will then present a computer-assisted framework for the development of problem-solving strategies in computational historical linguistics. The core of the framework is its interdisciplinary perspective, which helps to adapt existing solutions to similar problems in other disciplines to the problems in historical linguistics, while at the same time relying on a rigorous formalization and inspection of existing strategies in the "classical", qualitative approaches to historical language comparison. In contrast to the now very popular machine-learning approaches, which are widely applied to many problems in general and comparative linguistics, the framework is furthermore open with respect to the strategies which are used to solve a given problem, and it generally prefers explicit solutions (if they are available) over black-box strategies, as they are preferred in standard machine-learning frameworks. I will then pick four specific problems (automated morpheme detection, automated borrowing detection, automated induction of sound laws, and automated phonological reconstruction), and present initial strategies for their computer-assisted solution.
    Metadata:
    xml
    Status:
    Published
    Last Updated:
    4 years ago
    License:
    Attribution-NonCommercial
    Share this:

    Downloads

    Item Name: pdf list-2019-ichl-keynote-handout.pdf
      Download View in browser
    Activity: Downloads: 220

    Back to Deposits