Thesis Progress Report

# Thesis Progress Report #### Yi-Ting Tsai --- ## March #### 2023.03.03 ---- ### The Importance of Metaphor Detection - metaphors play a fundamental communicative role in verbal and written interactions (Lakoff & Johnson, 1980) - the portion of words used metaphorically ranges from 5% to 20% (Steen et al., 2010) - a choice of metaphors affects decision making (Thibodeau & Boroditsky, 2011) ---- ### Challenges for Metaphor Detection - metaphors cannot be interpreted at face value - metaphors are productive constructions - repeated use of a metaphor eventually solidifies it into a fixed expression (Hovy et al., 2013) - there is no clear-cut semantic distinction between figurative and metaphorical language (Shutova et al., 2010) - metaphors can be domain- and context-dependent (Tsvetkov et al., 2014) ---- ### Previous Studies on Metaphor Detection - WordNet (Neuman, 2013) - knowledge graph embedding (Song et al., 2021) - syntactic feature + lexical resource (Stowe, 2019) - word embedding (Su et al., 2020) - tree kernel (Hovy, 2013) ---- ### Metaphor Topics and Sources - **pain:** United Daily News + Chinese literary works (Lee, 2016) - **love:** 金曲龍虎榜 + G-music 風雲榜 + HitFm 年度百首單曲榜 (Wang, 2010) ---- ### To-do List - choose the scope and topic of metaphors - collect the dataset - develop research questions ---- --- ## April #### 2023.04.21 ---- ### To-do List in March - choose the scope and topic of metaphors - collect the dataset - develop research questions ---- ### Conceptual Metaphor (Kovecses, 2010) - a set of systematic correspondences between the source and the target (**mapping** the preexisting similarities). - source domain provides the target concept a particular structure or a set of elements - e.g., progress and destination of love ---- ### Love Metaphor (Kovecses, 2010) - the domain of emotion (e.g., anger, fear, love, happiness, sadness, shame, pride...) is primarily understood by means of conceptual metaphors - human relationships include such concepts as friendship, love, and marriage - love metaphor covers the **psychological and mental states and events** and **social groups and processes**. ---- ### Research Procedure 1. collect metaphor keywords 2. expand keyword list with collocations 3. find features in metaphoric expressions 4. train models to label metaphor source words ---- ### 1. collect metaphor keywords - PTT 小生 (mrp) - filter out articles that were not on Boy-Girl, marriage, WomenTalk, MenTalk, and sex Board - split with sentence - clean data ![](https://i.imgur.com/ilwNe4Q.png =500x250) ---- ### 2. expand keyword list with collocations - use articles on Boy-Girl, marriage, WomenTalk, MenTalk, and sex Board from 2014/07/24 to 2022/06/26 ---- ### 3. find features in metaphoric expressions - observe from the annotation process - refer to the literature - e.g., semantic similarity, word abstractness... ---- ### 4. train models to label metaphor source words ![](https://i.imgur.com/EyB6VVk.png) ---- ### To-do List - finish labeling the metaphor keywords - crawl the articles for finding collocations ---- --- ## May #### 2023.05.12 ---- ### Brief Introduction - Topic: Metaphor Detection in Love Topic - Research Purpose: identify metaphorical expressions to aid in translation and disambiguation ---- ### To-do List in April - finish labeling the metaphor keywords - crawl the articles for finding collocations ---- ### Metaphor Keywords Labeling - word references: 教育部《重編國語辭典修訂本》 - POS references: CKIP Parser - labeled sentences: 2035 - annotated keyword types: 235 ---- ### Observation of Metaphor Keywords 1. marriage breakdown: (1) military strategy e.g., 以拖待變、黑暗兵法 (2) chess game e.g., 死局、布局 2. sex: contest e.g., 決鬥、防具 3. free love: writing e.g., 小說家、結局 4. cohabitation: math e.g., 公約數、最佳解 5. marriage: drama e.g., 劇本、角色 6. pursuit: war e.g., 戰略價值、進退失據 ---- ### Observation of Metaphor Keywords 7. break up: train e.g., 火車、月台 8. facing choices: game e.g., 關卡、難關 9. taking care of children: logistics e.g., 支援、後援 10. making mistakes: crime e.g., 絞刑、犯案 11. relationship issue: schoolwork e.g., 課題、功課 12. betrayal: cooperative game e.g., 破關、豬隊友 ---- ### Collocation Pool - time period: 2014/07/24 to 2022/06/26 - total data: 349804 - sex board: 67126 - WomenTalk board: 105043 - MenTalk board: 70514 - Boy-Girl board: 76011 - marriage board: 31110 ---- ### Problem - "a number of fixed multi-word expressions or polywords are analyzed as one lexical unit" (Lu & Wang, 2017) e.g., 關公面前耍大刀、山海連成一線、蒙在鼓裡、投之以桃報之以李、夭鬼假小心、見笑轉生氣、軟土深掘、薑還是老的辣、做牛做馬 ---- ### To-do List in May - find collocations of metaphor keywords --- ## June #### 2023.06.02 ---- ### Integrate the Observation of Metaphor Keywords Stage Theory of Relationships (Levinger, 1980): - Attraction - Building a relationship - Continuance - Deterioration - Ending  ---- ### To-do List in May - find collocations of metaphor keywords ---- ### Procedure for Finding Collocations **colloction pool (2014/07/24 ~ 2022/06/26): 349804** 1. conduct word segmentation with CKIP WS 2. remove non-Chinese elements 3. calculate IDF to construct a custom stopword list (Ousirimaneechai & Sinthupinyo, 2018) - threshold: IDF value < 3.0 (Argaw et al., 2015) 4. filter stopwords 5. use NLTK package for finding collocations ---- ### Example for Finding Collocations ``` [ConcordanceLine( left=['的', '資訊', '兒童', '及', '少年', '性', '剝削', '依法', '指', '以下', '情況', '一', '使', '兒童', '或', '少年', '為', '有', '對'], query='價', right=['之', '性交', '或', '猥褻', '行為', '二', '利用', '兒童', '或', '少年', '為', '性交', '猥褻', '之', '行為', '以', '供', '人'] ``` - query (labeled metaphor keyword)：價 - collocations (after stopwords removal)： - left：少年 - right：性交 ---- ### Collocation Results - total stopwords: 468 - collocation types: 150926 ---- ### Problems - How to set the IDF threshold level for the stopword list? - Are these collocations all applicable? ---- ### To-do list in June - Check the applicability of the collocations --- ## October #### 2023.10.13 ---- ### Introduction - **Motivation:** manpower consuming and neglect of novel expressions on metaphor annotation - **Purpose:** proposing a semi-automatic method that uses collocations for annotation ---- ### Procedure 1. label metaphorical seed words in love topic 2. expand metaphorical expressions with collocations 3. inspect the collocations from an ontological perspective 4. use the ontological structure to examine Levinger's Relationship Stage Theory ---- ### Procedure 5. input metaphorical expressions into BERT as training data for metaphor recognition 6. demonstrate zero-shot on LLaMA to label metaphorical expressions 7. compare the performance with or without metaphorical expressions as training data ---- ### Research Questions 1. In the metaphor of love, what vocabulary terms are used in the source domain? 2. Do the collocations of the metaphorical seed words also express metaphorically? ---- ### Research Questions 3. Whether to train a pre-trained language model with a metaphorical expression list or not, which one can better capture metaphorical vocabulary terms in sentences? 4. After clustering the metaphorical vocabulary terms, can the clusters well represent the development processes in the Relationship Stage Theory by Levinger (1980)? ---- ### Current Progress - modify input data and metrics for BERT - slowly draw up the draft of the proposal ---- ### To-do List in October - finish the proposal draft - retrain BERT --- ## November #### 2023.11.23 ---- ### To-do List in October - **⍻** finish the proposal draft - ✗ retrain BERT ---- ### Mapping Love Metaphors to Ontological Structure - 10 love metaphors for observation：出軌、難關、抗爭、隊友、羈絆、後援、阻力、重擔、粉紅泡泡、冷戰 - Relationship Stage Theory (Levinger, 1980)： - Attraction - Building a relationship - Continuation - Deterioration - Ending ---- ### Examples for Annotation：重擔 |Collocation|Sentence|Annotation| |:---:|:---:|:---:| |經濟|重點是妳是否願意與他一起存錢達成財務目標，為婚姻家庭的經濟重擔預作準備。|Building → Continuation| |扛|段交往談結婚，麵包和財力證明的重擔恐怕也是全丟給男方扛。而他無法也不覺得能|Building → Continuation| |落在|據了解，因陳女的丈夫是軟爛男，家中經濟重擔落在陳女身上，原本在朋友介紹的按摩店上班，但仍...|Continuation → Deterioration| ---- ### Ontological Structure for Love Metaphors - Protégé ![observe_love_metaphor_ontology](https://hackmd.io/_uploads/Skzf222E6.png) ---- ### Problems <table> <tr> <th>Love Metaphor</th> <th>Collocation</th> <th>Sentence</th> <th>Annotation</th> </tr > <tr > <td rowspan="3">抗爭</td> <td>香港</td> <td> </td> <td> </td> </tr> <tr> <td>經過</td> <td>她就說到，她的前任是家人朋友都反對，但是她們經過種種抗爭，好不容易在一起</td> <td>Building → Continuation</td> </tr> <tr> <td>抗爭</td> <td>此時不論是選擇與家長抗爭到底，還是分手，對於雙方都是一場煎熬。</td> <td>Building → Continuation Continuation → Deterioration </td> </tr> ---- ### To-do List in November - mapping all of the 466 collocations to the ontological structure - revise and finish the proposal