A Contribution to the Theory of Word Length Distribution Based on a Stochastic Word Length Distribution Model

Abstract

We derive a stochastic word length distribution model based on the concept of compound distributions and show its relationships with and implications for Wimmer et al. ’s (1994) synergetic word length distribution model.

Notes

¹This might consequently be considered a Bayesian approach, cf. Lee (2009).

²Approaches that have used these more general alignment models include Galescu & Allen (2001), Jiampojamarn et al. (2007), and Bisani & Ney (2008).

³Or similar sizes of other segmental writing systems such as the Greek, Cyrillic, etc.

⁴Taken from the Pascal G2P challenge, http://www.pascal-network.org/Challenges/PRONASYL/

⁵If M also takes the value M = 0, then M might be perceived of as having a peak at M = 1.

⁶Note that compound distributions can be considered a special case of a mixed distribution.

⁷Here, by we denote polynomial coefficients (cf. Comtet, 1974; Caiado, 2007).

⁸The question is what the scope of such recursive schemes as (4) is. Certainly, to sketch an extreme, if they contained all possible probability distributions, they would be useless as theoretical models.

⁹Plus something termed “local modifications” in Wimmer & Altmann (1996) that mainly consists of marginal “corrections” to fit the data at hand, such as displacing probability distributions by 1, etc.