Evaluating Commonsense in Pre-trained Language Models

By Xuhui Zhou, Yue Zhang, Leyang Cui, Dandan Huang. 2019

199 citations recorded by the source corpus.

Contextualized representations trained over large raw text data have given remarkable improvements for NLP tasks including question answering and reading comprehension. There have been works showing that syntactic, semantic and word sense knowledge are contained in such representations, which explains why they benefit such tasks. However, relatively little work has been done investigating commonsense knowledge contained in contextualized representations, which is crucial for human question answering and reading comprehension. We study the commonsense ability of GPT, BERT, XLNet, and RoBERTa by testing them on seven challenging benchmarks, finding that language modeling and its variants are effective objectives for promoting models' commonsense ability while bi-directional context and larger training set are bonuses. We additionally find that current models do poorly on tasks require more necessary inference steps. Finally, we test the robustness of models by making dual test cases, which are correlated so that the correct prediction of one sample should lead to correct prediction of the other. Interestingly, the models show confusion on these test cases, which suggests that they learn commonsense at the surface rather than the deep level. We release a test set, named CATs publicly, for future research.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Large Language Models, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

In Large Language Models

In Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

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.”
Pre-Trained Models: Past, Present and Future (2021, methodology): “For example, VLP (Zhou et al., 2020a) and X-GPT (Xia et al., 2020) employ seq2seq MLM as their pre-training tasks.”
Explaining Question Answering Models through Text Generation (2020, background): “This has been demonstrated by their ability to answer questions that require common sense and world knowledge, without retrieving information from an external source (Trinh and Le, 2018; Zhou et al., 2019; Roberts et al., 2020; Ling et al., 2020).”
Behind the Scene: Revealing the Secrets of Pre-trained Vision-and-Language Models (2020, background): “[26,5] and [42] demonstrate that BERT has surprisingly strong ability to recall factual relational knowledge and perform commonsense reasoning.”
Birds Have Four Legs?! NumerSense: Probing Numerical Commonsense Knowledge of Pre-trained Language Models (2020, background): “Besides the works that we have discussed in Section 1, Zhou et al. (2020) and Talmor et al. (2019a) also proposed to probe the commonsense knowledge of pre-trained language models.”
LIREx: Augmenting Language Inference with Relevant Explanation (2020, background): “This happens because while deep learning-based text generators are powerful enough to generate readable sentences, they often lack commonsense reasoning ability (Zhou et al. 2020).”
LIFT: Language-Interfaced Fine-Tuning for Non-Language Machine Learning Tasks (2022, background): “They have shown great performances for various language-based tasks [20] such as question answering [21, 22], or commonsense reasoning [23].”
A Review on Language Models as Knowledge Bases (2022, background): “LMs are shown to possess a certain amount of commonsense knowledge in their parameters (Petroni et al., 2019; Davison et al., 2019; Zhou et al., 2020; Cui et al., 2021”
Relational World Knowledge Representation in Contextual Language Models: A Review (2021, methodology): “The single-LM approach pools the pretrained LM’s token scores over a candidate set of sequences, then takes the highest-scoring sequence as the LM’s “prediction” or choice (Tamborrino et al., 2020; Bouraoui et al., 2020; Zhou et al., 2020; Brown et al., 2020).”
Factuality Enhanced Language Models for Open-Ended Text Generation (2022, background): “Large LMs can be good at recalling factual knowledge given substantial pre-training corpus, suggested by previous studies on LM’s parameteric knowledge [33], but there still remains room for improvement for reasoning conceptual knowledge [34, 35].”
A Systematic Investigation of Commonsense Understanding in Large Language Models (2021, methodology): “Moreover, we show that given the existing (and sometimes inevitable) biases in some benchmarks, it is important to com-3 We use the 1B model mentioned in §5. pare the models with strong baselines which are sometimes omitted from recent work ( e.g. , Zhou et al., 2020; Brown et al., 2020).”
Linguistically-Informed Transformations (LIT): A Method for Automatically Generating Contrast Sets (2020, methodology): “Consistency In addition to using accuracy to measure models’ performance, recent research pays attention to consistency, which provides another perspective to probe models’ competence in the real world (Trichelair et al., 2018; Zhou et al., 2019; Gardner et al., 2020).”
Reasoning with Transformer-based Models: Deep Learning, but Shallow Reasoning (2021, background): “However, several studies [Forbes et al., 2019, Zhou et al., 2020b, Lin et al., 2020, Boratko et al., 2020, Singh et al., 2021] show that when the datasets are speciﬁcally changed to target the weaknesses of transformer-based models (for example, by adversarial instances), the models fail.”