聲符部件排序與形聲字發音規則探勘

近年來台灣有相當多的新移民的加入，這些新移民在口語的學習上雖然有地利之便，但是在漢字的認識上則是相當弱勢。由於漢字乃是圖形文字，學習單一字的成本相對的高。如果可以讓漢字教一個字，可以學到十個字，對於漢字教學的成效應有相當的助益。本文從部件教學的概念出發，考慮聲符的發音強度、出現頻率、及筆劃數，做為聲符部件教學順序的準則。我們利用部件發音強度（張嘉惠、林書彥、李淑瑩、蔡孟峰、李淑萍、廖湘美、孫致文、黃鍔，2010），以線性加總、幾合乘積、及調和平均三種方法對部件排序。根據此部件排序學習，前五個部件便可延伸學習多達140個相似發音的漢字。進一步，我們應用中研院文獻處理實驗室所建立的「漢字構形資料庫」，以及標記所得之形聲字，拆解形聲字組成的部件，挖掘串連漢字之間關係的形音關聯規則。我們從600萬條發音規則中篩選與分群出3組高信賴度與5組高支持度的規則，並藉由這些規則來輔助漢語發音的學習，提高學習效率。

關鍵字

形聲字；聲符強度；部件教學；學習曲線；關聯規則

並列摘要

In recent years, there are a considerable number of new immigrants in Taiwan. Although these people are in the good position to learn Chinese, the advantages are limited to speaking and listening. Recognizing Chinese characters is a tough task since one has to memorize the shape, meaning and pronunciation at the same time. Therefore, the cost of learning a single character is relatively high compared with other languages in alphabet system. The goal of this study is to make the 80% pictophonetic characters to be organized more systematically such that the pronunciation of most pictophonetic characters can be inferred automatically. We evaluate the importance of Chinese components by considering the pronunciation strength, occurring frequency, and number of strokes using linear sum, product, and harmonic mean, respectively. Furthermore, we discover pronunciation rules by association mining with priority grouping. Three groups of high reliability rules and five groups of high support rules are demonstrated in this paper to show the effectiveness of pronunciation rule discovery.