Statistical properties of probabilistic context-sensitive grammars

Kai Nakaishi, Koji Hukushima
Condensed Matter, Disordered Systems and Neural Networks, Disordered Systems and Neural Networks (cond-mat.dis-nn), Statistical Mechanics (cond-mat.stat-mech), Physics and Society (physics.soc-ph)
2024-02-11 00:00:00
Probabilistic context-free grammars, commonly used to randomly generate trees, have been well analyzed theoretically and have found applications in various domains. Despite their utility, the distributions that the grammar can express are limited to those in which the distribution of a subtree depends only on its root and not on its context. This limitation becomes a challenge for modeling various real-world phenomena such as natural language. To overcome this limitation, a probabilistic context-sensitive grammar is introduced, where a subtree's distribution depends on its context, and its statistical properties are explored. Numerical analysis reveals that the distribution of a symbol does not exhibit a qualitative difference from that in the context-free case, but mutual information does. Furthermore, a metric is introduced to directly quantify the breaking of this limitation. This metric is zero in the context-free case, and is applicable to an arbitrary distribution of a tree. Measuring this metric enables the detection of a distinct difference between probabilistic context-free and context-sensitive grammars.
PDF: Statistical properties of probabilistic context-sensitive grammars.pdf
Empowered by ChatGPT