以語言知識庫爲基礎的智慧型作詩填詞輔助系統

舊有的作詩填詞格律自動檢測索引系統，以詩詞譜、韻書、單字讀音資料、工具書爲基礎，「能」將一首不合格律的詩詞修改爲符合格律的詩詞，但是因爲電腦不具備人的知識體系，無法識別情感，因此「不能」修改對仗聯的格律，「不能」修改用得不好的字詞，無法幫助使用者解決創作時最容易遇到的詞窮問題，當然更「不能」將一首寫得不好的詩詞，改成修辭與心境、情境、意境等形式與內容皆美的作品。本研究探討如何在包含：詞彙讀音資料、平仄聲調、韻書、詞彙字數、詞性、語意、語意概念分類等信息的韻文語言知識庫及情感計算的研究成果基礎上，建構一套「能」輔助使用者修改舊有系統「不能」檢測修改的多音字之平仄聲調與押韻，對仗聯之平仄聲調、詞性、句型句法，以及詞窮時如何循著符合人類知識體系的語意概念結構，搜尋符合格律需要的同近義詞詞彙，達到修改用詞的功能，以及未來如何在情感計算及整合韻文、語體文、現代漢語等三種語言知識庫的基礎上，再進一步強化作詩填詞系統的功能，使電腦具備更接近人工智慧以及具有情感識別與表達的能力，輔助使用者寫出形式與內容皆美的詩詞。

關鍵字

作詩填詞系統；語言知識庫；情感計算；數位學習

並列摘要

Existing automatic prosodic indexing systems for poetry and lyrics writings are mainly built upon poetry and lyric scores, rhyme books, pronunciation guides for characters and tool books. While these computer tools are ”capable” of editing prosodically non-conforming poems and lyrics to conform to prosodic principles, they do not have the knowledge structure humans possess particularly with respect to distinguishing feelings and emotions, and therefore are ”incapable” of editing the prosodic errors in couplets or assisting in polishing inappropriate choice of words/characters. All in all, they fail completely in terms of helping users to overcome their greatest challenge in breaking the bottleneck of vocabulary. Needless to say, they are helpless in terms of lifting a poorly written poem or lyric up another level to become a composition sound in rhetoric, depicting mindset, circumstances as well as images in corresponding forms, achieving the ultimate goal of form-content balance. This present study explores the feasibility of establishing a system upon research findings in areas as verse language knowledge database as well as affective computing covering a spectrum of subjects including pronunciation information, tone patterns, rhyming, word length, parts of speech, word senses, and semantic concept categories. It is hoped that such a system, surpassing the limitations listed above, would be able to ”empower” users in poetry and lyric composition. The ideal system would be able to match tone patterns and rhymes for characters with multiple pronunciations and identify prosody in couplets, parts of speech, sentence patterns among others. When users run out of inspiration, the system would present synonym choices that conform to prosodic principles and in accordance with conceptual meanings in human knowledge system. On this basis, we make projection on an advanced poetry and lyric composing system supported by affecting computing and drawing on knowledge database of verses, colloquial writing style, and modern Mandarin. An expressive computing system capable of recognizing feelings and emotions bring us one notch higher in artificial intelligence, fulfilling the purpose of helping users to produce poems/lyrics high in aesthetic values in both form and content.