Information-Theoretic Probing with Minimum Description Length

By Elena Voita, Ivan Titov. 2020

324 citations recorded by the source corpus.

To measure how well pretrained representations encode some linguistic property, it is common to use accuracy of a probe, i.e. a classifier trained to predict the property from the representations. Despite widespread adoption of probes, differences in their accuracy fail to adequately reflect differences in representations. For example, they do not substantially favour pretrained representations over randomly initialized ones. Analogously, their accuracy can be similar when probing for genuine linguistic labels and probing for random synthetic tasks. To see reasonable differences in accuracy with respect to these random baselines, previous work had to constrain either the amount of probe training data or its model size. Instead, we propose an alternative to the standard probes, information-theoretic probing with minimum description length (MDL). With MDL probing, training a probe to predict labels is recast as teaching it to effectively transmit the data. Therefore, the measure of interest changes from probe accuracy to the description length of labels given representations. In addition to probe quality, the description length evaluates "the amount of effort" needed to achieve the quality. This amount of effort characterizes either (i) size of a probing model, or (ii) the amount of data needed to achieve the high quality. We consider two methods for estimating MDL which can be easily implemented on top of the standard probing pipelines: variational coding and online coding. We show that these methods agree in results and are more informative and stable than the standard probes.

Subfields: Natural Language Processing, 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 Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

True Few-Shot Learning with Language Models (2021, methodology): “We use NLL following prior work in MDL [32, 60, 10], to retain MDL’s property as a measure of label compression.”
Decoupling the Role of Data, Attention, and Losses in Multimodal Transformers (2021, background): “Previous work in NLP has considered probing classifiers to evaluate representations; however, this approach can be misleading as the performance of probing classifiers does not solely depend on the quality of representations (e.g., Hewitt and Liang, 2019; Voita and Titov, 2020).”
Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model Beliefs (2021, background): “…work has explored how much knowledge is stored in LMs (Heinzerling and Inui, 2021), approaches to querying models for knowledge (Hewitt and Liang, 2019; Jiang et al., 2020; Voita and Titov, 2020; West et al., 2021), and meth-ods for learning more knowledge during pretraining (Wang et al., 2021b,a).”
Factual Probing Is [MASK]: Learning vs. Learning to Recall (2021, background): “These include random baselines (Hewitt and Liang, 2019) and information-theoretic measurements (Voita and Titov, 2020).”
Probing Classifiers: Promises, Shortcomings, and Advances (2021, background): “In a more practical view, Voita and Titov (2020) propose to measure both the performance of the probe g and its complexity, by estimating the minimum description length of the code required to transmit property z knowing the representations f l ( x ) : MDL ( g, f, D O , D P ) .”
Amnesic Probing: Behavioral Explanation with Amnesic Counterfactuals (2020, background): “Indeed, there is an ongoing debate on the assertion that certain information is ‘‘encoded’’ in the representation (Voita and Titov, 2020; Pimentel et al., 2020).”
Eliciting Knowledge from Language Models Using Automatically Generated Prompts (2020, background): “However, probing classiﬁers require additional learned parameters and are thus susceptible to false positives; high probing accuracy is not a sufﬁcient condition to conclude that an LM contains a certain piece of knowledge (Hewitt and Liang, 2019; Voita and Titov, 2020).”
When Do You Need Billions of Words of Pretraining Data? (2020, methodology): “Second, we apply minimum description length probing (Voita and Titov, 2020) to the edge probing suite, with the goal of quantifying the accessibility of these features.”
What Context Features Can Transformer Language Models Use? (2021, positive): “Our approach also shares similar motivations to information-theoretic work on probing (Voita and Titov, 2020; Pimentel et al., 2020), which uses related tools to interpret linguistic structure in LM representations rather than characterizing their effect on LM predictions.”
Information-Theoretic Probing for Linguistic Structure (2020, background): ““Better” representations should make their respective probes easily learnable—and consequently their encoded information is more accessible (Voita and Titov, 2020).”
Acquisition of chess knowledge in AlphaZero (2021, methodology): “More sophisticated regression methods such as data-dependent sparsity using Gated Linear Networks [84], information-theoretic regularisation using the information bottleneck [85], minimum description length probing [86] or Bayesian probing [87] may provide ways to mitigate this problem.”
Analyzing Redundancy in Pretrained Transformer Models (2020, background): “Recently, Voita and Titov (2020); Pimentel et al. (2020) proposed probing methods based on information theoretic measures.”
Asking without Telling: Exploring Latent Ontologies in Contextual Representations (2020, methodology): “measures to estimate and trade off capacity with accuracy (Hewitt and Liang, 2019; Voita and Titov, 2020).”
Investigating Gender Bias in BERT (2020, methodology): “Equity Evaluation Corpus (EEC) [23] The dataset contains template-based sentences such as “<Name> feels angry”.”
Conditional probing: measuring usable information beyond a baseline (2021, background): “Voita and Titov (2020) propose a new probing-like methodology, minimum description length (MDL) probing , to measure the number of bits required to transmit both the speciﬁcation of the probe and the speciﬁcation of labels.”