Probing Natural Language Inference Models through Semantic Fragments

By Kyle Richardson, Hai Hu, Larry Moss, Ashish Sabharwal. 2019

152 citations recorded by the source corpus.

Do state-of-the-art models for language understanding already have, or can they easily learn, abilities such as boolean coordination, quantification, conditionals, comparatives, and monotonicity reasoning (i.e., reasoning about word substitutions in sentential contexts)? While such phenomena are involved in natural language inference (NLI) and go beyond basic linguistic understanding, it is unclear the extent to which they are captured in existing NLI benchmarks and effectively learned by models. To investigate this, we propose the use of semantic fragments—systematically generated datasets that each target a different semantic phenomenon—for probing, and efficiently improving, such capabilities of linguistic models. This approach to creating challenge datasets allows direct control over the semantic diversity and complexity of the targeted linguistic phenomena, and results in a more precise characterization of a model's linguistic behavior. Our experiments, using a library of 8 such semantic fragments, reveal two remarkable findings: (a) State-of-the-art models, including BERT, that are pre-trained on existing NLI benchmark datasets perform poorly on these new fragments, even though the phenomena probed here are central to the NLI task; (b) On the other hand, with only a few minutes of additional fine-tuning—with a carefully selected learning rate and a novel variation of “inoculation”—a BERT-based model can master all of these logic and monotonicity fragments while retaining its performance on established NLI benchmarks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

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

Citations in the corpus

oLMpics-On What Language Model Pre-training Captures (2019, background): “Richardson et al. (2019) showed that LMs are able to successfully apply knowledge of antonyms in a natural language inference setup.”
Transformers as Soft Reasoners over Language (2020, background): “The synthetic, conditional probes in [Richardson et al. , 2020] test single rule application, e.g., “If Joe has visited Potsdam then Anne has visited Tampa.”
CLUE: A Chinese Language Understanding Evaluation Benchmark (2020, background): “We believe that for a model to make robust inferences, it needs to understand such unique Chinese phenomena, and also has other important linguistic abilities, such as handling anaphora resolution (Webster et al., 2018) and monotonicity reasoning (Yanaka et al., 2019; Richardson et al., 2020).”
A Primer in BERTology: What We Know About How BERT Works (2020, background): “Zhou et al. (2020) and Richardson and Sabharwal (2019) also show that the performance drops with the number of necessary inference steps.”
Thinking Aloud: Dynamic Context Generation Improves Zero-Shot Reasoning Performance of GPT-2 (2021, background): “Pre-trained neural language models, whether uni-or bidirectional, display a poor zero-shot performance on reasoning tasks [e.g., Yanaka et al., 2019, Clark et al., 2020, Richardson et al., 2020].”
PRover: Proof Generation for Interpretable Reasoning over Rules (2020, methodology): “These include tasks in the bAbI dataset (Weston et al., 2015), synthetically generated probe tasks (Richardson et al., 2020) or reading comprehension tasks in datasets such as QuaRTz (Tafjord et al., 2019) and ROPES (Lin et al., 2019).”
Unsupervised Distillation of Syntactic Information from Contextualized Word Representations (2020, background): “A large body of works have shown sensitivity to both semantic (Tenney et al., 2019a; Richardson et al., 2019) and syntactic (Tenney et al., 2019b; Lin et al., 2019; Reif et al., 2019; Hewitt and Manning, 2019; Liu et al., 2019a) information.”
Quantifying Adaptability in Pre-trained Language Models with 500 Tasks (2021, background): “A great deal of past work has focused on improving evaluation using synthetic evaluation sets (Jia and Liang, 2017; Naik et al., 2018; Lake and Baroni, 2018; Richardson et al., 2020).”
MonaLog: a Lightweight System for Natural Language Inference Based on Monotonicity (2019, methodology): “We will look into options for artiﬁcially creating neutral examples, e.g. by ﬁnding reverse entail-ments 10 , as illustrated by Richardson et al. (2019).”
multiPRover: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning (2021, methodology): “The task of rule reasoning (Clark et al., 2020) is related to other recently proposed tasks on QA (We-ston et al., 2015; Yang et al., 2018; Lin et al., 2019; Tafjord et al., 2019; Richardson et al., 2020) and NLI (MacCartney and Manning, 2014).”
Are Pretrained Language Models Symbolic Reasoners over Knowledge? (2020, methodology): “Richardson et al. (2020) introduce synthetic corpora testing logic and monotonicity reasoning.”
OCNLI: Original Chinese Natural Language Inference (2020, methodology): “…NLI, future work will also focus on using our dataset for doing Chinese model probing (e.g., building on work such as Warstadt et al. (2019); Richardson et al. (2020); Jeretic et al. (2020)) and sentence representation learning (Reimers and Gurevych, 2019), as well as for investigating…”
What Does My QA Model Know? Devising Controlled Probes Using Expert Knowledge (2019, methodology): “There has also been work on using synthetic data of the type we consider in this paper (Poliak et al., 2018a; Geiger et al., 2019; Richardson et al., 2020).”
Think Before You Speak: Using Self-talk to Generate Implicit Commonsense Knowledge for Response Generation (2021, background): “Recent work also aims to probe models in these tasks to see if reasoning is actually achieved (Richardson and Sabharwal, 2020; Richardson et al., 2020; Zhou et al., 2020; Lin et al., 2021b).”
RICA: Evaluating Robust Inference Capabilities Based on Commonsense Axioms (2020, background): “Recent work also aims to probe models in these tasks to see if reasoning is actually achieved (Richardson and Sabharwal, 2020; Richardson et al., 2020).”
ANLIzing the Adversarial Natural Language Inference Dataset (2020, background): “There has also been work in probing NLI models to see what they learn (Richardson et al., 2019), as well as specifically on the collection of NLI annotations (Bowman et al., 2020) and analyzing inherent disagreements between human annotators (Pavlick and Kwiatkowski, 2019).”
Commonsense-Focused Dialogues for Response Generation: An Empirical Study (2021, background): “Recent work also aims to probe models in these tasks to see if reasoning is actually achieved (Richardson and Sabharwal, 2020; Richardson et al., 2020; Zhou et al., 2020).”
Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge (2020, background): “Models show high accuracy on skills that they are not explicitly trained for, but have been shown to be acquired during pre-training, such as size comparison [11], and meronyms [18].”
Exploring End-to-End Differentiable Natural Logic Modeling (2020, methodology): “The monotonicity subset of Semantic Fragments is automatically generated with a controlled set of rules and lexicons, which contains around 2,000 pairs.”