Quantitative studies on the Indonesian prefixes PE- and PEN-

Abstract

The purpose of this dissertation is to conduct a systematic quantitative approach to analyse the Indonesian prefixes PE- and PEN-. The questions addressed in this dissertation are (1) whether the two similar prefixes PE- and PEN- are allomorph or not, (2) to what extent the semantics differences between these two prefixes could be captured by distributional semantics, and (3) how are PE- and PEN- learn by a computational model, namely Linear Discriminative Learning. In order to answer the questions, I compiled a database containing 3090 words with PE- and PEN- from a corpus of written Indonesian. Using the productivity analyses, the data showed that PEN- is apparently more productive than PE-. The difference productivity also occurs in their semantic roles, inflectional variants, as well as frequency ratio between their corresponding verbal prefixes BER- and MEN-. This corpus-based research thus suggests that PE- and PEN- are two independent prefixes. Furthermore, a distributional vector space model was applied to my data to clarify whether PE- and PEN- have discriminable semantics. Cosine similarities mean comparisons revealed that pairs consisting of words with PE- has a higher similarity than words with PEN-. Furthermore, nouns with PE- were more similar to their base words than was the case for words with PEN-. What drives the higher similarity for PE- than for PEN- is that PE- is specialised to create athletes. These findings provide a further quantitative evidence for treating PE- and PEN- as two independent prefixes. Finally, I made use of a computational model, the `discriminative lexicon' (DL) model, to investigate the learnability of PE- and PEN-. As nouns with PEN- is corresponding with verbs with MEN-, this last study focused on if there is a trade-off between the affix substitution PEN- and MEN- regularity in comprehension learning. The findings suggest that PE- is learned more robustly than PEN- for two main reasons. First, PE- words tend to be longer and hence have more discriminative triphones. Second, due to cue competition with MEN-, the prefixal triphones of PEN- are less effective cues than those of PE-. A new measure of functional load is also proposed to shed light the relative importance of the triphones in the prefixes. To sum up, this dissertation clarifies that prefixes, which at first blush look like allomorphs, can have different qualitative and quantitative properties.