An Information Theoretic View on Selecting Linguistic Probes
By Zining Zhu, Frank Rudzicz. 2020
22 citations recorded by the source corpus.
There is increasing interest in assessing the linguistic knowledge encoded in neural representations. A popular approach is to attach a diagnostic classifier -- or "probe" -- to perform supervised classification from internal representations. However, how to select a good probe is in debate. Hewitt and Liang (2019) showed that a high performance on diagnostic classification itself is insufficient, because it can be attributed to either "the representation being rich in knowledge", or "the probe learning the task", which Pimentel et al. (2020) challenged. We show this dichotomy is valid information-theoretically. In addition, we find that the methods to construct and select good probes proposed by the two papers, *control task* (Hewitt and Liang, 2019) and *control function* (Pimentel et al., 2020), are equivalent -- the errors of their approaches are identical (modulo irrelevant terms). Empirically, these two selection criteria lead to results that highly agree with each other.
Subfields: Artificial Intelligence, Natural Language Processing, Information Theory
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Oleg Serikov
- Vivek Srikumar
- Yichu Zhou
- Miruna Clinciu
- Ryan Teehan
- Shachar Mirkin
- Aaron Gokaslan
- Natasha Seelam
In Natural Language Processing
- Oleg Serikov
- Vivek Srikumar
- Yichu Zhou
- Natasha Seelam
- Miruna Clinciu
- Yonatan Belinkov
- Ryan Teehan
- Aaron Gokaslan
In Information Theory
Citations in the corpus
- Factual Probing Is [MASK]: Learning vs. Learning to Recall (2021, neutral): “by the probe (Hewitt and Liang, 2019; Pimentel et al., 2020; Voita and Titov, 2020; Zhu and Rudzicz, 2020).”
- Probing Classifiers: Promises, Shortcomings, and Alternatives (2021, background): “While Pimentel et al. (2020b) posit that their control function are a better criterion than the control tasks of Hewitt and Liang (2019), subsequent work showed that the two criteria are almost equivalent, both theoretically and empirically (Zhu and Rudzicz 2020).”
- DirectProbe: Studying Representations without Classifiers (2021, background): “Recent work (Pimentel et al., 2020b; Voita and Titov, 2020; Zhu and Rudzicz, 2020) probe representations from an information theoretic perspective.”
- Emergent Structures and Training Dynamics in Large Language Models (2022, result): “Recent work suggests that both approaches yield comparable results empirically with similar error terms theoretically (Zhu and Rudzicz, 2020).”