CEREX Task @ FIRE 2020
CEREX (Cause-Effect Relation EXtraction from Text) Task
The concept of causality can be informally introduced as a relationship between two events e1 and e2 such that occurrence of e1 results in the occurrence of e2 . Expression of causality can be manifested within a text documents in arbitrarily complex ways:
For example, in the sentence:
“Aircel files for bankruptcy over mounting financial troubles” - the event “mounting financial troubles” is causing the event “Aircel filed for bankruptcy.”
In a more complicated scenario, like:
“Company recalled some vehicles to fix loose bolts that could lead to engine stall” - we can observe nested cause-effect pairs: Here, the effect “company recalled vehicle”is caused by the event “to fix loose bolts is not easy to extract. That the cause “loose bolts” could lead to engine stall”,is even more difficult to detect.
The extraction of causal relations from textual mentions is an important step for the improvement of many Natural Language Processing applications such as question-answering, document-summarization, opinion mining, event analysis - it enables the possibility to reason about the detected events. Many organization, specializing in web intelligence and knowledge graph creation - provide services for the analysis and representation of huge amounts of textual information. Therefore, the extraction of causal information can be used for the creation of new insights.
While there are few explicit lexico-syntactic patterns in exact correspondence with a causal relation, there are a huge number of cases that can evoke a causal relation, but not in a unique way.
For example, the following sentence contains a causal relation where from is the pattern which evokes such relation:
“Pollution from cars is causing serious health problems for Americans.”
In this case, the words (pollution and cars) connected by the cue pattern (from) are in a causal relation.
However, in the following sentence the from pattern doesn’t evoke the same type of relation:
“A man from Oxford with leprosy was cured by the water.”
Therefore, due to the presence of complex grammatical structures in sentences, automatic extraction of causal relations becomes a very difficult task .
The task has two parts
A) Identify whether a given sentence contains a causal event (either cause/effect) :
- Such as the example containing 'from' above
B) Annotate each word in a sentence in terms of the four labels cause (C),effect(E), causal connectives(CC) and None:
- Such as shown in the figure below
We will use datasets from three different domains to generate a set of new training data for this track.
1) Part of the SemEval 2010 Task 8 data set dealing with“Cause-Effect”
2) Adverse drug effect (ADE) dataset
3) BBC News Article dataset
4) An inhouse data from Educational domain
The training data will only be shared with the registered participants upon agreeing to use the data only for academic and research purposes.
Dasgupta, T., Saha, R., Dey, L., & Naskar, A. (2018). Automatic extraction of causal relations from text using linguistically informed deep neural networks. In Proceedings of the 19th Annual SIGdial Meeting on Discourse and Dialogue(pp. 306-316).