Studies of Russian, Polish, and Lithuanian language learners converge on the finding that morphological features of nouns are first generalized to word clusters of high morpho-phonological similarities such as diminutives, that grammatical categorisation is are more easily applied to novel words that fall into these clusters. The present thesis explores whether the facilitating effect of diminutives on the acquisition of complex noun morphology can be extended to Serbian, a south Slavic language, morphologically similar to Russian and Polish. Specifically, the thesis explores the role of parameters responsible for the obtained diminutive advantage: high frequency of a particular cluster of words in child-directed speech (CDS) and morpho-phonological homogeneity within this cluster. A corpus analysis of the distribution of diminutives in Serbian CDS indicated a rather unexpected difference in frequency relative to Russian and Polish CDS, despite the high similarity of the diminutive derivation across these three Slavic languages. Out of the total number of nouns in Serbian CDS only 7% were diminutives, compared to 20-30% in Polish and 45% in Russian. Two experimental studies explored whether the low frequency of diminutives in Serbian CDS attenuates the diminutive advantage in morphology learning compared to Russian and Polish. In the first two experiments, Serbian children exhibited a strong diminutive advantage for both gender agreement and case marking in the same range as Russian children, indicating that morpho-phonological homogeneity within the cluster of diminutives may play as important a role as their frequency for grammatical categorisation of novel nouns. A third study investigated in more detail the effects of morpho-phonological homogeneity on the emergence of the diminutive advantage using a gender-agreement task with novel nouns in simplex and pseudo-diminutive form over four sessions with Serbian children. The results showed a pseudo-diminutive advantage for gender agreement by Session 2, suggesting that the categorisation of nouns into grammatical categories is based on morpho-phonological homogeneity of the word cluster, emerges relatively fast, and can occur despite the much lower frequency of diminutives in Serbian CDS. Finally, a series of neural network simulations designed to capture the pattern of results from the third experimental study was used to examine to what extent a simple associative learning mechanism, relying on morpho-phonological similarity of the noun endings, can explain the findings. The performance of three models, a whole-word feed-forward network, a Simple Recurrent Network (SRN) and a last-syllable feed-forward network, was compared to the experimental data. The superior fit of the SRN suggests that gender learning is based on a very fast sequential build-up of representations of the entire word, allowing the system to exploit the predictive power of word stems to anticipate regularised endings. Overall, the findings of this thesis contribute to our general understanding of mechanisms responsible for the acquisition of complex inflectional noun morphology in two ways. First, by extending experimental studies and neural network simulations to Serbian, the results underline the universality of the idea that noun morphology is learned and processed through a single-route associative mechanism based on the frequency and morpho-phonological structure of nouns. More specifically, the results from experimental studies and neural network simulations demonstrate that for diminutives, the low-level grammatical categorisation is based mainly on the morpho-phonological similarity of word endings, and can emerge after just a few exposures. And second, the neural network simulations suggest that during the process of categorisation of nouns into gender categories, learners rely not only on predictable information from the noun endings, but also on phonological regularities in the stems of nouns. Taken together, these findings contribute also to a better understanding of the facilitating role of CDS in morphology acquisition.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:513682 |
Date | January 2006 |
Creators | Seva, Nada |
Contributors | Kempe, Vera |
Publisher | University of Stirling |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://hdl.handle.net/1893/114 |
Page generated in 0.0145 seconds