Tim and Casey are best buddies since high school. One day, after years, they meet up in a bar for some drinks, sharing their work life stories, cursing at their bosses ;). They’re working at a company named McRels Inc. It turns out that their jobs are very similar. Both of them are working with text, news to be exact. Tim’s main responsibility is reading news, then ordering every events happening in the news in a timeline, guessing whether an event comes after or before another event. Casey started working on his current job just recently. His job is to decide whether there is a causality between two events.
Tim: “So, when I am given a text Typhoon Haiyan struck the eastern Philippines on Friday, killing thousands of people, I should be able to guess that the struck happens before the killing.”
Casey: “Aaah, I see… For me, I must decide whether the struck caused the killing or not.”
Casey’s job seems to be easier since the decision is binary: yes or no (well, also to decide which one is the cause and which one is the effect), but it’s actually much more difficult than Tim’s. One reason is that, unlike Tim, Casey doesn’t have enough resources to learn how to decide on the causality. Moreover, the concept of causality is more abstract than temporal ordering.
Tim is very lucky, because he could participate in a challenge on guessing the event ordering. As we know, competition can lead people to perform their best—that is, it can improve their quality of performance. Unfortunately, Casey doesn’t get that chance.
First of all, Casey needs to build resources for learning on deciding the causality between events, so he hires minions to do that. He can only afford two minions since he’s low on budget.
The minions are not so smart, so he needs to set up guidelines for them to annotate causal information in text. Tim offers him to use his available resources, which is a collection of text already annotated with all events available in the text. Having the resources annotated with causal information, Casey can finally learn how to identify causality between events in text.
Since in theory, causality has a temporal constraint, that the cause happens before the effect, Tim and Casey have an idea to cooperate in order to improve their learning abilities. There still need to be some discussions for this idea, involving more meetings in a bar… with a lot of drinks, I suppose :).
P.S.: in case you don’t get the metaphors, this story is basically my PhD topic, where Tim and Casey are automatic systems for extracting temporal and causal relations, respectively. And the two minions are actually me and my advisor :D.
- Paramita Mirza and Sara Tonelli. 2014. Classifying Temporal Relations with Simple Features. In Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 308–317, Gothenburg, Sweden, April. Association for Computational Linguistics.↩
- Paramita Mirza, Rachele Sprugnoli, Sara Tonelli and Manuela Speranza. 2014. Annotating causality in the TempEval-3 corpus. In Proceedings of the EACL 2014 Workshop on Computational Approaches to Causality in Language (CAtoCL), pages 10–19, Gothenburg, Sweden, April. Association for Computational Linguistics.↩
- Paramita Mirza and Sara Tonelli. 2014. An Analysis of Causality between Events and its Relation to Temporal Information. (to appear) in Proceedings of the 25th International Conference on Computational Linguistics, Dublin, Ireland.↩
- Paramita Mirza. 2014. Extracting Temporal and Causal Relations between Events. In Proceedings of the ACL 2014 Student Research Workshop, pages 10–17, Baltimore, MD, United States, June.↩