[PLing] Roth, "Automating linguistic tests using pre-trained language models" (19. 1.)

Laura Grestenberger laura.grestenberger at univie.ac.at
Mon Jan 18 11:55:28 CET 2021

Liebe Kolleg*innen,

Ich möchte Euch/Sie herzlich zu einem weiteren Online-Vortrag der Wiener 
Sprachgesellschaft einladen: Benjamin Roth (Universität Wien, Institut 
für Europäische und Vergleichende Sprach- und 
Literaturwissenschaft/Forschungsgruppe Data Mining and Machine Learning) 
spricht morgen (19. 1.) über "Automating linguistic tests using 
pre-trained language models".


This work in progress is based on the premise that a main method for 
linguistic analysis is that of transforming and reformulating utterances 
in conjunction with judging the results of those (hypothetical, 
counterfactual) transformations. Such a process can be called a 
linguistic test.

The concrete form of a linguistic test arises either directly from the 
task at hand (e.g., if an ambiguous word is replaced by synonyms of two 
different readings, which one is less marked) or it follows from a 
specific theory (e.g., language is organized hierarchically based on 
categories, and replacements by elements of the same category should not 
make an utterance ungrammatical).

We argue that the analysis based on linguistic tests is fundamentally 
different from the methods that are typically employed in the automatic 
analysis of language by computers, namely rule-based analysis (using 
grammar formalisms) and machine-learning-based analysis (using annotated 
training examples). We will show initial experiments how linguistic 
tests can be automated using pre-trained neural language models.

Link zum BBB-Raum: 

Zugangscode: 284338

Beginn: 18.30h.

Interessierte sind herzlich willkommen!


Laura Grestenberger

Mag. Laura Grestenberger, PhD
Institut für Sprachwissenschaft
Universität Wien
Sensengasse 3a, A-1090 Wien
T +43 1 4277 41720
Laura.Grestenberger at univie.ac.at

More information about the PLing mailing list