Evaluating Computational Language Models with Scaling Properties of Natural Language
This work addresses the problem of evaluating language models for researchers in computational linguistics and AI, offering a method to assess model fidelity to natural language statistics, though it is incremental in applying existing scaling analyses to models.
The study evaluated computational language models by testing their ability to reproduce five universal scaling properties of natural language, such as Zipf's law and long-range correlations. It found that only RNN-based models with gating mechanisms (e.g., LSTM, GRU) could replicate the long memory behavior, and identified Taylor's law exponent as a good indicator of model quality.
In this article, we evaluate computational models of natural language with respect to the universal statistical behaviors of natural language. Statistical mechanical analyses have revealed that natural language text is characterized by scaling properties, which quantify the global structure in the vocabulary population and the long memory of a text. We study whether five scaling properties (given by Zipf's law, Heaps' law, Ebeling's method, Taylor's law, and long-range correlation analysis) can serve for evaluation of computational models. Specifically, we test $n$-gram language models, a probabilistic context-free grammar (PCFG), language models based on Simon/Pitman-Yor processes, neural language models, and generative adversarial networks (GANs) for text generation. Our analysis reveals that language models based on recurrent neural networks (RNNs) with a gating mechanism (i.e., long short-term memory, LSTM; a gated recurrent unit, GRU; and quasi-recurrent neural networks, QRNNs) are the only computational models that can reproduce the long memory behavior of natural language. Furthermore, through comparison with recently proposed model-based evaluation methods, we find that the exponent of Taylor's law is a good indicator of model quality.