Pareto Probing: Trading-Off Accuracy and Complexity

By Tiago Pimentel, Naomi Saphra, Adina Williams, Ryan Cotterell. 2020

71 citations recorded by the source corpus.

The question of how to probe contextual word representations for linguistic structure in a way that is both principled and useful has seen significant attention recently in the NLP literature. In our contribution to this discussion, we argue for a probe metric that reflects the fundamental trade-off between probe complexity and performance: the Pareto hypervolume. To measure complexity, we present a number of parametric and non-parametric metrics. Our experiments using Pareto hypervolume as an evaluation metric show that probes often do not conform to our expectations---e.g., why should the non-contextual fastText representations encode more morpho-syntactic information than the contextual BERT representations? These results suggest that common, simplistic probing tasks, such as part-of-speech labeling and dependency arc labeling, are inadequate to evaluate the linguistic structure encoded in contextual word representations. This leads us to propose full dependency parsing as a probing task. In support of our suggestion that harder probing tasks are necessary, our experiments with dependency parsing reveal a wide gap in syntactic knowledge between contextual and non-contextual representations.

Subfields: Artificial Intelligence, Information Systems, Efficiency & Compression, Evaluation & Benchmarks

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

In Information Systems

In Efficiency & Compression

In Evaluation & Benchmarks

Citations in the corpus

Probing Classifiers: Promises, Shortcomings, and Advances (2021, background): “Taking an information-theoretic perspective on probing, Pimentel et al. (2020b) proposed to use control functions instead of control tasks in order to compare probes.”
Implicit Representations of Meaning in Neural Language Models (2021, neutral): “A phenomenon is judged to be encoded by a model if the probe’s accuracy cannot be explained by its accuracy when trained on control tasks (Hewitt and Liang, 2019) or baseline models (Pimentel et al., 2020).”
When Do You Need Billions of Words of Pretraining Data? (2020, background): “New probing methods related to MDL address different aspects of these problems (Whit-ney et al., 2020; Pimentel et al., 2020a) and may yield different conclusions.”
Emergent Structures and Training Dynamics in Large Language Models (2022, methodology): “Many of these approaches highlight the need to for-208 malize the "effort" required in encoding a linguistic property, often via some form of a control function (Pimentel et al., 2020b).”
Probing Classifiers: Promises, Shortcomings, and Alternatives (2021, background): “Taking an information-theoretic perspective on probing, Pimentel et al. (2020b) proposed to use control functions instead of control tasks in order to compare probes.”
Low-Complexity Probing via Finding Subnetworks (2021, methodology): “This question has been the subject of much recent debate (Hewitt and Liang, 2019; Voita and Titov, 2020; Pimentel et al., 2020b, inter alia ).”
Emergent Linear Representations in World Models of Self-Supervised Sequence Models (2023, background): “Figure 12: Cosine similarity scores between linear probes across layers. as inadequate (Pimentel et al., 2020b; Saphra and Lopez, 2019).”
Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color (2021, methodology): “We vary α across a wide range of settings to examine the effect of probe complexity, which we measure using the nuclear norm of the linear projection matrix W ∈ Rφ×ι; ||W ||∗ = ∑min(φ,ι) i=1 σi(W ), where σi(W ) is the ith singular value of W (Pimentel et al., 2020).”
A Bayesian Framework for Information-Theoretic Probing (2021, methodology): “Finally, the last result goes against one of the claims of Pimentel et al. (2020a), who in light of their ﬂat Pareto curves for POS tagging claimed that we needed harder tasks for probing.”
DirectProbe: Studying Representations without Classifiers (2021, methodology): “Recent work (Pimentel et al., 2020b; Voita and Titov, 2020; Zhu and Rudzicz, 2020) probe representations from an information theoretic perspective.”
Masked Language Modeling and the Distributional Hypothesis: Order Word Matters Pre-training for Little (2021, methodology): “To probe the array of pre-trained models for syntactic, semantic and other linguistic properties, we investigate dependency parsing using Pareto probing (Pimentel et al., 2020a) as well as the probing tasks from Conneau et al.”
Conditional probing: measuring usable information beyond a baseline (2021, methodology): “While Pimentel et al. (2020b) conclude from the data processing inequality that probing is not meaningful, we conclude that estimating mutual information is not the goal of probing.”