Natural Language Statistical Features of LSTM-generated Texts

TL;DR

This paper analyzes the statistical properties of texts generated by LSTM networks, comparing them to natural language and Markov models, revealing that LSTM-generated texts best mimic natural long-range correlations at an optimal temperature setting.

Contribution

It provides a comprehensive quantitative analysis of LSTM-generated language, highlighting its ability to replicate long-range correlations similar to natural language, and identifies an optimal generation parameter.

Findings

LSTM texts reproduce long-range correlations similar to natural language.

Word-frequency and entropy measures of LSTM texts are comparable to real language.

An optimal temperature parameter exists that makes LSTM texts most similar to natural language.

Abstract

Long Short-Term Memory (LSTM) networks have recently shown remarkable performance in several tasks dealing with natural language generation, such as image captioning or poetry composition. Yet, only few works have analyzed text generated by LSTMs in order to quantitatively evaluate to which extent such artificial texts resemble those generated by humans. We compared the statistical structure of LSTM-generated language to that of written natural language, and to those produced by Markov models of various orders. In particular, we characterized the statistical structure of language by assessing word-frequency statistics, long-range correlations, and entropy measures. Our main finding is that while both LSTM and Markov-generated texts can exhibit features similar to real ones in their word-frequency statistics and entropy measures, LSTM-texts are shown to reproduce long-range correlations at scales comparable to those found in natural language. Moreover, for LSTM networks a temperature-like parameter controlling the generation process shows an optimal value---for which the produced texts are closest to real language---consistent across all the different statistical features investigated.