Explaining Contextualized Word Embeddings in Biomedical Research – A Qualitative Investigation

Miletic, Marko; Sariyar, Murat (2022). Explaining Contextualized Word Embeddings in Biomedical Research – A Qualitative Investigation Studies in Health Technology and Informatics, 295, pp. 289-292. IOS Press 10.3233/SHTI220719

[img]
Preview
Text
SHTI-295-SHTI220719.pdf - Published Version
Available under License Creative Commons: Attribution-Noncommercial (CC-BY-NC).

Download (174kB) | Preview

Contextualized word embeddings proved to be highly successful quantitative representations of words that allow to efficiently solve various tasks such as clinical entity normalization in unstructured texts. In this paper, we investigate how the Saussurean sign theory can be used as a qualitative explainable AI method for word embeddings. Our assumption is that the main goal of XAI is to produce confidence and/or trust, which can be gained through quantitative as well as quantitative approaches. One important result is related to the fact that the differential structure of language as explained by Saussure corresponds to the possibility of adding and subtracting word embeddings. On the other hand, these mathematical structures provide insights into the inner workings of natural language.

Item Type:

Journal Article (Original Article)

Division/Institute:

School of Engineering and Computer Science > Institut für Medizininformatik I4MI
School of Engineering and Computer Science

Name:

Miletic, Marko and
Sariyar, Murat

Subjects:

Q Science > QA Mathematics
Q Science > QA Mathematics > QA75 Electronic computers. Computer science

ISSN:

1879-8365

ISBN:

9781643682907

Series:

Studies in Health Technology and Informatics

Publisher:

IOS Press

Language:

English

Submitter:

Murat Sariyar

Date Deposited:

20 Dec 2022 15:33

Last Modified:

15 Jan 2024 15:26

Publisher DOI:

10.3233/SHTI220719

ARBOR DOI:

10.24451/arbor.18476

URI:

https://arbor.bfh.ch/id/eprint/18476

Actions (login required)

View Item View Item
Provide Feedback