《Emotion Cause Events: Corpus Construction and Analysis》

1.Key Points

---

1. Construct a corpus for emotion cause detection task

2. Analysis of the linguistic phenomena of cause events

3. Exam the correlations between emotions and cause events in terms of linguistic cues

2. Related Theoretical Issues

---

1. Although there is no agreement among different theories on the emotion classes, a small number of primarys are commonly assumed whereas other emotions are secondary emotions which are the mixture of the primary emotions.

   five primary emotions: happiness, sadness, fear, anger, surprise

   Researchers have attempted to prospose the list of primary emotions, varying from two to ten basic emotions. Fear and anger appear on every list, whereas happiness sadness. Other less common primary emotions are surprise, disgust, shame, distress, guilt, interest, pain and acceptance

2. Many basic issues remains unresolved, for instance, the relationship among emotions, emotion class selection, etc.

3. The lack of research on cause emotion restricted current emotion analysis to simple classificatory work without exploring the potential of the rich applications of putting emotion ‘in context’

3. Linguistic Phenomena of Cause Events

1. Cause event refers to the immediate cause of the emotion, which can be the actual trigger event or the perception of the trigger event

   The emotion Fear is more likely to be linked to the perception of the trigger event, for instance a snake is not the direct cause of Fear in a person, but through the cognitive awareness of the possible danger that the snake may bring to the person.

2. In Chinese, cause events are categorized into two types:
   2.1 verbal events
   2.2 nominal event

3. In the corpus, there are many cases requiring coreference.

   eg. " 台湾当局以政治理由拒接当事人返台。当事人对 此 愤懑"。在这句话中，“此”被识别为触发事件，而实际上它是一个代词，指向 “台湾当局以政治理由拒接当事人返台”。

4. When the antecedent is out of context, only the anaphoric expression is marked.

   夫妻两昨天都对 此喜讯 兴奋不已

5. Since the existence of covert subjects and covert objects is a common phenomenon in Chinese syntactic structure, it is not surprising that a single verb alone is the cause event.

   …救了我的性命。人鱼公主 听了 很伤心

6. The actual cause event is usually too long and complicated, which involves several events.

   妈妈为了鼓励我考试得满分，便告诉我说：如果你考一张一百分，妈妈就给你五元做奖励。我 听了 好高兴

7. There are also cases where the actual cause events are not expressed in the context.

4. Construct the Corpus

---

1. Firstly, extract entries of sentences by emotion keywords matching from the Sinica data in our emotion corpus.

2. Secondly, remove irrelevant entries that do not need

   2.1 are non-emotional

   Happiness is a wonderful song

   2.2 contain highly-ambigous emotion keywords

   如意, 害羞, 为难

3. Entries without causes explicitly expressed are mainly due to following resons:

   3.1 there is not enough contextual information

   the previous or the suffix sentence are interjection, e.g. 嗯哼！“aha”

   3.2 when the focus sentence is the beginning or the ending of a paragraph, no prefix sentence or suffix sentence can be extracted as the context

   the cause may be beyond the context

   3.3 when the cause is obscure, which can be something very abstract or even with unknown reasons

5. Corpus Analysis - Linguistic Cues

- Causative verbs 原因动词
- Perception verbs 感知动词
- Epistemic markers 认知标志
- Prepositions 介词
- Conjunctions 连词
- Others 
	- is 的是
	- say 的说
	- at 于
	- can 能