注意力模型類神經網路在無監督式學習下的自動歌詞改編生成

An Attention Based Neural Network Model for Unsupervised Lyrics Rewriting

吳肇中(Chao-Chung Wu)

指導教授 ： 林守德

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201800743

摘要

本研究的主要目的為，透過多重編碼器的模型架構，結合語言模型的訓練，達成可以在保留輸出格式的限制下，完成能以無監督式學習的方式對原歌詞下一句的作改寫預測。本研究主要分成二個部分，一部分探討改寫歌詞的品質。透過自動評價以及人工評價的標註，對改寫結果在中心主旨、押韻、可唱度上可以比擬原歌詞甚至好於原歌詞，在自動評價方面亦顯示出押韻的高準確度。另一個部分基於前面部分所建構的模型，觀察模型在學習正確韻律、詞性、情感的改寫上。本研究觀察了改寫條件在情緒、押韻的改動(transfer)結果，以及隨著訓練回數的注意力分布，和每次預測規一化結果的(softmax)分布上前10名在韻律、詞性上的準確率。

關鍵字

自然語言生成 ； 機器學習 ； 創意寫作

並列摘要

Creative writing has become a standard task to showcase the power of artificial intelligence. This work tackles a challenging task in this area, the lyrics rewriting. This task possesses several unique challenges. First, we require the outputs to be not only semantically correlated with the original lyrics, but also coherent in segmentation structure, rhyme as the rewritten lyrics must be performed by the artist with the same music. Second, there is no parallel rewriting lyrics corpus available for supervised training. We propose a deep neural network based model for this task and exploit both general evaluation metrics such as ROUGE and human study to evaluate the effectiveness of the model.

並列關鍵字

Natural language generation ； Machine learning ； Creative Writing

參考文獻

[1]Peter Potash, Alexey Romanov, and Anna Rumshisky. Ghostwriter: Using anlstm for automatic rap lyric generation. In Proceedings of the 2015 Confer-ence on Empirical Methods in Natural Language Processing, pages 1919–1924.Association for Computational Linguistics, 2015.

[2]Dekai Wu, Karteek Addanki, Markus Saers, and Meriem Beloucif. Learning tofreestyle: Hip hop challenge-response induction via transduction rule segmen-tation. In Proceedings of the 2013 Conference on Empirical Methods in NaturalLanguage Processing, pages 102–112, Seattle, Washington, USA, October 2013.Association for Computational Linguistics.

[3]Kento Watanabe, Yuichiroh Matsubayashi, Kentaro Inui, and Masataka Goto.Modeling structural topic transitions for automatic lyrics generation. In Pro-ceedings of the 28th Pacific Asia Conference on Language, Information, andComputation, pages 422–431, Phuket,Thailand, December 2014. Departmentof Linguistics, Chulalongkorn University.

[4]Ananth Ramakrishnan A., Sankar Kuppan, and Sobha Lalitha Devi. Auto-matic generation of tamil lyrics for melodies. In Proceedings of the Workshopon Computational Approaches to Linguistic Creativity, pages 40–46, Boulder,Colorado, June 2009. Association for Computational Linguistics.

[5]Jack Hopkins and Douwe Kiela. Automatically generating rhythmic verse withneural networks. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 168–178.Association for Computational Linguistics, 2017.