Predictive Modelling of Type Valency in Word Formation Grammar

ABSTRACT This paper explores different regression models for predicting the type valency of Persian suffixes within a usage-based approach. Usage-based models treat the type frequency of a suffix as a key predictor for its type valency revealing that an increase in the type frequency leads to a greater combining power between a construction’s paradigmatic elements. However, this effect is limited to a certain degree by the potential productivity of a suffix, as inferred from the statistically distinguishable negative correlation between the type valency and the potential productivity, as well as from the statistical significance of the variable of the number of hapaxes and the potential productivity in the regression models of conditional inference trees. Moreover, polyvalency as a distinct feature of Persian derivation implies a number of other characteristics, namely greater morphological diversity of patterns, parsability, semantic transparency and larger conversion power of morphemes. This is contrasted with English whose morphemes are predominantly type-monovalent.