Summary of The Role Of Syntactic Span Preferences in Post-hoc Explanation Disagreement, by Jonathan Kamp et al.
The Role of Syntactic Span Preferences in Post-Hoc Explanation Disagreement
by Jonathan Kamp, Lisa Beinborn, Antske Fokkens
First submitted to arxiv on: 28 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper investigates post-hoc explanation methods for increasing model transparency, which are essential for users. Currently used methods for attributing token importance often yield diverging patterns. From a linguistic perspective, this study finds that different methods systematically select different word classes and that methods agreeing with other methods and humans display similar preferences. Token-level differences between methods are smoothed out when comparing them at the syntactic span level. The paper also explores the interaction between k (a fixed subset size) and spans, proposing an improved configuration for selecting important tokens. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The study looks into why different explanation methods give different answers about what’s important in a text model. It seems that these methods pick out different types of words and are more likely to agree with each other and humans if they share similar preferences. When we look at the same information at a higher level, like sentences or phrases, instead of individual words, the differences between methods become smaller. The paper also suggests ways to improve how we choose important tokens by looking at the relationships between these tokens. |
Keywords
» Artificial intelligence » Token
