Exploring Lavender Tongue from Social Media Texts[In Chinese]

Exploring Lavender Tongue from Social Media Texts 吳小涵 Hsiao-Han Wu 國立臺灣大學語言學研究所 Graduate Institute of Linguistics National Taiwan University xiaohanwu.hanna@gmail.com 謝舒凱 Shu-Kai Hsieh 國立臺灣大學語言學研究所 Graduate Institute of Linguistics National Taiwan University shukaihsieh@ntu.edu.tw 摘要 在性別與自然語言處理的脈絡下，大多數研究僅專注於生理性別的討論，對於性別文本 的分類，更僅建立於異性戀男女的文本上。針對此一現象，本研究為中文性別與自然語 言處理領域中，第一個由性取向的觀點出發，討論同性文本偵測的研究。首先本研究由 網路論壇-PTT 收集同性戀文本並討論同性戀男女的語言學特徵。其次，藉由觀察到的 語言學現象，利用 5 折交叉驗證支持向量機器(Support Vector Machine)與樸素貝葉斯分 類器(Naive Bayes)模型，以機器訓練的方式，利用不同的語言學特徵組來偵測男同性戀 與女同性戀的網路文本。機器訓練結果顯示，在同性文本的預測上，由於本研究使用了 傳統性別與自然語言處理研究未考量到的同性戀特有詞彙特徵，而在同性文本偵測上達 到了較佳的正確率。 Abstract Under the issue of gender and Natural Language Processing (NLP), most papers aim at gender- norm language that spoken by biologically males and females with opposite-sex desires. However, from the point of view of sexual orientation, this study presents the first work in the task of Chinese homosexual identification. Firstly, we collect homosexual texts from social media, and secondly examine linguistic behavior found in gay and lesbian texts. In addition, we also provide sets of linguistic features to automatically predict homosexual language with the adoption of 5-fold cross-validation Support Vector Machine (SVM) and Naive Bayes (NB) models. Training procedure in the study resulted in promising f-score around 70% with the use 68 The 2017 Conference on Computational Linguistics and Speech Processing ROCLING 2017, pp. 68-80  The Association for Computational Linguistics and Chinese Language Processing of particular lexicon-based feature set. 關鍵詞：同性文本偵測，薰衣草語言學，中文自然語言處理，支持向量機器，樸素貝葉 斯分類器 Keywords: homosexual identification, lavender linguistics, Chinese NLP, Support Vector Machine, Naive Bayes 1. Introduction Lavender Linguistics has been emerging as a linguistic sub-field which analyzes language used by gay, lesbian, bisexual, transgender, and queer (LGBTQ) speakers [1]. It is suggested that there is still considerable room for linguistic research based on fine-grained sexual orientation [2]. Previous studies of gender and NLP mainly focused on dichotomous genders in biological sense without considering the gender complexity of human beings in real world. When it comes to gender, a general but complicated term with various dimensions involving both biology and psychology, anthropologists have divided it into three major classes: 1) Sex refers to physical or biological differences between males and females. 2) Opposite to physical characteristics, gender is characterized by self-identity, namely, whether one see himself/herself as male or female. 3) Sexuality is about one's sexual attraction and orientation. People who have opposite-sex desires are regarded as heterosexuals. Conversely, people who have same-sex desires are, therefore, regarded as homosexuals. In the field of gender and NLP (abbreviated as GenderNLP), gender is usually considered with the norm that subjects are biologically males and females with heterosexual desires. However, based on the perspective towards sexuality, the present paper discusses lavender speakers and NLP (abbreviated as LavenderNLP) with the hypothesis that previous study on gender identification cannot correctly identify gender in a more complex dimension and that GenderNLP has failed to consider the complexity of sexuality. 69 Since GenderNLP only aims at biological gender, LavenderNLP, as a subclass of GenderNLP, targets not only at biological gender but also at psychological gender. Therefore, referring to Table 1, subjects in the current study are regarded as homosexual males (gays) and females (lesbians) who have same-sex desires regardless of whether he or she self-identifies as male or female; in other words, only sex and sexuality are taken into account …

Content truncated to 500 words