Linguistic Knowledge and Transferability of Contextual Representations

By Nelson F. Liu, Matt Gardner, Yonatan Belinkov, Matthew E. Peters, Noah A. Smith. 2019

833 citations recorded by the source corpus.

Contextual word representations derived from large-scale neural language models are successful across a diverse set of NLP tasks, suggesting that they encode useful and transferable features of language. To shed light on the linguistic knowledge they capture, we study the representations produced by several recent pretrained contextualizers (variants of ELMo, the OpenAI transformer language model, and BERT) with a suite of sixteen diverse probing tasks. We find that linear models trained on top of frozen contextual representations are competitive with state-of-the-art task-specific models in many cases, but fail on tasks requiring fine-grained linguistic knowledge (e.g., conjunct identification). To investigate the transferability of contextual word representations, we quantify differences in the transferability of individual layers within contextualizers, especially between recurrent neural networks (RNNs) and transformers. For instance, higher layers of RNNs are more task-specific, while transformer layers do not exhibit the same monotonic trend. In addition, to better understand what makes contextual word representations transferable, we compare language model pretraining with eleven supervised pretraining tasks. For any given task, pretraining on a closely related task yields better performance than language model pretraining (which is better on average) when the pretraining dataset is fixed. However, language model pretraining on more data gives the best results.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Representation & Self-Supervised Learning

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 Computer Vision and Pattern Recognition

In Natural Language Processing

In Representation & Self-Supervised Learning

Citations in the corpus

SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems (2019, methodology): “The outside results for COPA, MultiRC, and RTE are from Sap et al. (2019), Trivedi et al. (2019), and Liu et al. (2019c) respectively.”
VisualGPT: Data-efficient Adaptation of Pretrained Language Models for Image Captioning (2021, background): “Ideally, we would like to exploit the massive amount of linguistic knowledge stored in the PLM weights (e.g., [44]), while referring to the visual input only when required.”
Causal Analysis of Syntactic Agreement Mechanisms in Neural Language Models (2021, background): “For instance, several studies have mapped LM representations to either independent syntactic dependencies (Belinkov, 2018; Liu et al., 2019; Tenney et al., 2019b) or full dependency parses (Hewitt and Manning, 2019; Chi et al., 2020) as a proxy for discovering latent syntactic knowledge within the…”
Probing Classifiers: Promises, Shortcomings, and Advances (2021, background): “Some prior work acknowledged that conclusions can only be made about the existing trained models, not about general architectures (Liu et al. 2019).”
Transformer Feed-Forward Layers Are Key-Value Memories (2020, positive): “This observation corroborates recent ﬁndings that lower (upper) layers in deep contextualized models encode shallow (semantic) features of the inputs (Peters et al., 2018; Jawahar et al., 2019; Liu et al., 2019).”
Amnesic Probing: Behavioral Explanation with Amnesic Counterfactuals (2020, methodology): “We conclude that dependency information, POS and NER are important for word prediction.”
What Does BERT Look at? An Analysis of BERT’s Attention (2019, neutral): “Other work has drawn a similar conclusions from examining BERT’s predictions on agreement tasks (Goldberg, 2019) or internal vector representations (He-witt and Manning, 2019; Liu et al., 2019).”
BERTScore: Evaluating Text Generation with BERT (2019, methodology): “This is consistent with recent ﬁndings showing that the intermediate BERT layers may lead to more semantically meaningful contextual embeddings than the ﬁnal layer (Reimers and Gurevych, 2019; Liu et al., 2019).”
Evaluating the Underlying Gender Bias in Contextualized Word Embeddings (2019, methodology): “Peters et al. (2019) and Liu et al. (2019) conﬁrmed the viability of using ELMo representations directly as features for downstream tasks without re-training the full model on the target task.”
Designing and Interpreting Probes with Control Tasks (2019, methodology): “Probes have been used to make relative claims between models or components (Adi et al., 2017; Liu et al., 2019; Belinkov et al., 2017) or absolute claims about models above baselines.”
A Primer in BERTology: What We Know About How BERT Works (2020, methodology): “, 2020; Rosa and Mareček, 2019), although probing classifiers could not recover the labels of distant parent nodes in the syntactic tree (Liu et al., 2019a).”
Revealing the Dark Secrets of BERT (2019, background): “Jawahar et al. (2019) 2019) investigated the transferability of contextualized word representations to a number of probing tasks requiring linguistic knowledge.”
Revisiting Few-sample BERT Fine-tuning (2020, methodology): “BERT Representation Transferability BERT pre-trained representations have been widely studied using probing methods showing that the pre-trained features from intermediate layers are more transferable [15, 16, 25, 40, 41] or applicable [55] to new tasks than features from later layers, which change ”
Eliciting Knowledge from Language Models Using Automatically Generated Prompts (2020, methodology): “A common strategy is to use probing classiﬁers—shallow classiﬁers that predict certain attributes using an LMs’ representations as features (Conneau et al., 2018; Liu et al., 2019).”
BERTology Meets Biology: Interpreting Attention in Protein Language Models (2020, background): “At a high level, mechanisms for interpreting BERT can be categorized into three main categories: interpreting the learned embeddings [1, 16, 22, 47, 85], BERT’s learned knowledge of syntax [26, 29, 30, 43, 45, 74], and BERT’s learned knowledge of semantics [23, 74].”
Memory Efficient Continual Learning with Transformers (2022, background): “There are a number of works that explores task transferability in NLP [42, 32, 59, 45, 44]”
Analyzing the Structure of Attention in a Transformer Language Model (2019, methodology): “In a Transformer model, the middle layers were most predictive of dependencies (Liu et al., 2019; Tenney et al., 2019).”