How Can We Accelerate Progress Towards Human-like Linguistic Generalization?

By Tal Linzen. 2020

215 citations recorded by the source corpus.

This position paper describes and critiques the Pretraining-Agnostic Identically Distributed (PAID) evaluation paradigm, which has become a central tool for measuring progress in natural language understanding. This paradigm consists of three stages: (1) pre-training of a word prediction model on a corpus of arbitrary size; (2) fine-tuning (transfer learning) on a training set representing a classification task; (3) evaluation on a test set drawn from the same distribution as that training set. This paradigm favors simple, low-bias architectures, which, first, can be scaled to process vast amounts of data, and second, can capture the fine-grained statistical properties of a particular data set, regardless of whether those properties are likely to generalize to examples of the task outside the data set. This contrasts with humans, who learn language from several orders of magnitude less data than the systems favored by this evaluation paradigm, and generalize to new tasks in a consistent way. We advocate for supplementing or replacing PAID with paradigms that reward architectures that generalize as quickly and robustly as humans.

Subfields: Artificial Intelligence, 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 Natural Language Processing

Citations in the corpus

Experience Grounds Language (2020, neutral): “ability to measure generalization (Linzen, 2020).”
Are NLP Models really able to Solve Simple Math Word Problems? (2021, methodology): “Inspired by the notion of Normative evaluation (Linzen, 2020), our goal is to create a dataset of simple problems that any system designed to solve MWPs should be expected to solve.”
The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics (2021, background): “For example, we may improve the training sets, make the test sets more challenging, or probe for specific skills a model must exhibit with testonly datasets (Perez-Beltrachini and Gardent, 2017; Linzen, 2020; Ribeiro et al., 2020; Schlegel et al., 2020).”
Underspecification Presents Challenges for Credibility in Modern Machine Learning (2020, background): “There is increasing concern about whether natural language processing systems are learning general linguistic principles, or whether they are simply learning to use surface-level shortcuts (e.g., Bender and Koller, 2020; Linzen, 2020).”
Measuring and Improving Consistency in Pretrained Language Models (2021, background): “However, accuracy is just one measure of PLM performance (Linzen, 2020).”
Dynabench: Rethinking Benchmarking in NLP (2021, background): “train/test splits and human language (Linzen, 2020; Stiennon et al., 2020).”
CrossFit: A Few-shot Learning Challenge for Cross-task Generalization in NLP (2021, background): “In retrospect, researchers have advocated for building more human-like, general linguistic intelligence that can “reuse previously acquired knowledge about a language and adapt to a new task quickly” (Yogatama et al., 2019; Linzen, 2020).”
Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics (2020, background): “(OOD) generalization remains a challenge (Yogatama et al., 2019; Linzen, 2020); and, while recent large pretrained language models help, they fail to close this gap (Hendrycks et al., 2020).”
DynaSent: A Dynamic Benchmark for Sentiment Analysis (2020, background): “Challenge and adversarial datasets (Winograd, 1972; Levesque, 2013) have risen to prominence in response to the sense that benchmark results are over-stating the quality of the models we are developing (Linzen, 2020).”
Teaching Machine Comprehension with Compositional Explanations (2020, background): “This often leads to a large gap between methods in the research settings and practical use cases, as large amounts of annotated data rarely exist for a new task or a low-resource domain (Linzen, 2020).”
Learning from others' mistakes: Avoiding dataset biases without modeling them (2020, background): “However, in practice, that remains a challenge in natural language processing (Linzen, 2020; Yogatama et al., 2019) and our work aims at out-of-distribution robustness without signiﬁcantly compromising in-distribution performance.”
Predicting Inductive Biases of Pre-Trained Models (2021, background): “Evidence that similar generalization behavior can result from pre-trained representations has exciting implications for those interested in sample efﬁciency and cognitively-plausible language learning (Warstadt & Bowman, 2020; Linzen, 2020).”
PROST: Physical Reasoning about Objects through Space and Time (2021, methodology): “…is a pretrained QA model, built off T5, and finetuned on SQuad 1.1, SQuaD 2.0, Narra-tiveQA, RACE, ARC, OpenBookQA, MCTest, and BoolQ (Rajpurkar et al., 2016, 2018; Kočiský et al., 2018; Lai et al., 2017; Clark et al., 2018; Mihaylov et al., 2018; Richardson et al., 2013; Clark et al., 2019).”
The debate over understanding in AI’s large language models (2022, background): “Thus benchmarks or assessments that would be appropriate for measuring human understanding might not be appropriate for assessing such machines [10, 18, 31].”
Evaluation Examples are not Equally Informative: How should that change NLP Leaderboards? (2021, background): “An unfortunate side effect of leaderboard popularity is SOTA-chasing, often at the expense of carefully inspecting data and models (Linzen, 2020).”
Why AI is harder than we think (2021, background): “…test only very limited versions of these abilities; moreover, many of these benchmarks allow machines to learn shortcuts, as I described above—statistical correlations that machines can exploit to achieve high performance on the test without learning the actual skill being tested [57, 58].”