Japanese has effectively only two accent patterns: A quantitative and information-theoretic approach to lexical predictability

<p>Japanese pitch accent is traditionally described as a free accent system allowing up to n + 1 possible patterns for an n-mora word. However, Japanese accentuation is heavily constrained by factors such as word length, prosodic structure, and lexical stratum, leading to the hypothesis that the system fundamentally reduces to only two dominant patterns: unaccented and thematic (antepenultimate). Evaluating this claim requires moving beyond raw counts of empirically attested patterns to quantify the skewness of their distribution. In this study, we introduce information-theoretic metrics, specifically Shannon's conditional entropy and ecological effective numbers of diversity, to measure the uncertainty an diversity of accent patterns in a representative sample of 8,173 Japanese common nouns. Our findings show that while overall pattern richness reaches the theoretical maximum (7 patterns), global effective diversity remains low (3.12 by type frequency, 3.43 by token frequency). Crucially, with knowledge of word length, prosodic structure, and lexical stratum the number of accent patterns is reduced to effectively two (2.1) dominant patterns, providing a precise, quantitative validation of the two-pattern hypothesis. We further demonstrate that prosodic structure is the single most informative factor, reducing accentuation uncertainty by 23%, and that native nouns exhibit greater residual uncertainty than Sino-Japanese and foreign words. Finally, the strong alignment between type and token results confirms that dictionary-based word lists closely mirror natural speech distributions in phonological diversity.</p>