Metadata
- Author: ruder.io
- Title: ACL 2022 Highlights
- Reference: https://ruder.io/acl2022/
- Category: #article
Page Notes
Highlights
-
They also mentioned a positive development, that researchers are becoming more aware of the value of high-quality datasets. Overall, the panelists emphasised that working with such languages requires respect—towards the speakers, the culture, and the languages themselves. — Updated on 2022-06-26 14:36:59 — Group: #Public
-
In the past, there was pressure not to work on tasks that did not achieve high inter-annotator agreement; similarly, in traditional sentiment analysis, the neutral class is often discarded. Understanding cannot just be crammed into simple categories. Annotator opinions bias language models (Sap et al., 2021) and ambiguous examples improve generalization (Swayamdipta et al., 2020). — Updated on 2022-06-26 14:50:59 — Group: #Public
-
Yejin argued that language, knowledge, and reasoning are also not separate areas but exist on a continuum. — Updated on 2022-06-26 14:51:17 — Group: #Public