Updates from Previous Task


On the 2015 pilot subtask only 1:1 alignments were allowed. We were aware that 1:1 chunk alignment presented some limitations, as we weren't able to explicitly represent all interactions between chunks. A special tag (aligned in context, ALIC) was used to mark chunks participating in N:M relations:

  •     ALIC: This chunk has not any corresponding chunk in the other sentence because of the 1:1 alignment restriction, but otherwise the chunk would be aligned to some other chunk.

In 2016, we are happy to announce that the interpretable task will allow N:M alignments with no restrictions.

For example, lets consider the following sentence pair:

  1. [A smiling woman]1 [with a beer]2 [sitting]3 [outside]4 [with another smiling woman]5
  2. [Two women]1 [sitting]2 [outside]3 [laughing]4

On the previous 2015 pilot subtask we would do the following alignments, limited by the 1:1 requirement:

  • Alignment of chunks:  1 - 1 (EQUI 5), 3 - 2 (EQUI 5), 4 - 3 (EQUI 5),  2 - 0 (NOALI), 5 - 0 (ALIC), 0 - 4 (ALIC)

On the contrary, for the new task the resulting alignmnets would be this way:

  • Alignment of chunks:  1,5 - 1,4 (EQUI 5), 3 - 2 (EQUI 5), 4 - 3 (EQUI 5),  2 - 0 (NOALI)

 

Please check the guidelines for more details

Contact Info

Organizers

  • Eneko Agirre
  • Aitor Gonzalez-Agirre
  • Inigo Lopez-Gazpio
  • Montse Maritxalar
  • German Rigau
  • Larraitz Uria
  • University of the Basque Country (UPV/EHU)

email : ists-semeval@googlegroups.com

group : Interpretable STS SemEval

Other Info