Language Model Evaluation Beyond Perplexity

By Clara Meister, Ryan Cotterell. 2021

124 citations recorded by the source corpus.

We propose an alternate approach to quantifying how well language models learn natural language: we ask how well they match the statistical tendencies of natural language. To answer this question, we analyze whether text generated from language models exhibits the statistical tendencies present in the human-generated text on which they were trained. We provide a framework–paired with significance tests–for evaluating the fit of language models to these trends. We find that neural language models appear to learn only a subset of the tendencies considered, but align much more closely with empirical trends than proposed theoretical distributions (when present). Further, the fit to different distributions is highly-dependent on both model architecture and generation strategy. As concrete examples, text generated under the nucleus sampling scheme adheres more closely to the type–token relationship of natural language than text produced using standard ancestral sampling; text from LSTMs reflects the natural language distributions over length, stopwords, and symbols surprisingly well.

Subfields: Artificial Intelligence, Large Language Models, 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 Large Language Models

In Evaluation & Benchmarks

Citations in the corpus

Measuring Data (2022, background): “This means that these measurements are only directly comparable across different datasets when the same models are used – this is a known issue with perplexity, whose limitations in terms of evaluating language models have been pointed out (Meister and Cotterell, 2021b).”
How Much Do Language Models Copy From Their Training Data? Evaluating Linguistic Novelty in Text Generation Using RAVEN (2021, unknown): “Pannitto and Herbelot (2020), Meister and Cotterell (2021), and Yamakoshi et al. (2022) also analyzed models’ linguistic abilities by comparing model-generated text to human-generated text, but none of these focused on novelty.”
LMentry: A Language Model Benchmark of Elementary Language Tasks (2022, background): “Perplexity, while associated with general model quality, does not provide specific or interpretable insights into model capabilities (Meister and Cotterell, 2021).”