Downloadable software

• A scorer for the English all-words task of SENSEVAL-2 written in Perl, the official scorer created by Senseval allowed repeated answers to count towards coverage and this bug is removed in this scorer. I have used this software in my PhD thesis and in Probabilistic Word Sense Disambiguation.

Software available to NLP students in Cambridge

• Lappin and Leass anaphora resolution algorithm

  An implementation in C of the Lappin and Leass anaphora resolution algorithm. This program does not use any training data, so pretty much depends on the accuracy of the RASP parser on your domain. Some of the filters can be changed if this is relevant to your domain, but it's unlikely to produce a massive change in performance. The input format includes a RASP parse (and GRs) for each sentence, the output format contains equivalences between the heads of noun phrases and pronouns.

  This anaphora resolution algorithm is compared to others in A Comparison Of Probabilistic and Non-Probabilistic Anaphora Resolution Algorithms and the filters are described in Intermediate Parsing for Anaphora Resolution? Implementing the Lappin and Leass non-coreference filters.

• Kennedy and Boguraev anaphora resolution algorithm

  An implementation in C of the Kennedy and Boguraev anaphora resolution algorithm. Again, no training data is used so accuracy is strongly linked to the performance of the RASP GRs. The input format includes a RASP parse (and GRs) for each sentence, the output format contains equivalences between the heads of noun phrases and pronouns.

  This anaphora resolution algorithm is compared to others in A Comparison Of Probabilistic and Non-Probabilistic Anaphora Resolution Algorithms.

• Ge and Charniak anaphora resolution algorithm

  An implementation in C of the Ge and Charniak probabilistic anaphora resolution algorithm. This algorithm requires training data, and its performance is strongly dependent on the amount of relevant data. I have some training data that I've created, but it might not be relevant to your domain and so the algorithm won't be quite so accurate as if you provided your own training data. The input format includes a RASP parse for each sentence, the output format contains equivalences between the heads of noun phrases and pronouns.

  This anaphora resolution algorithm is compared to others in A Comparison Of Probabilistic and Non-Probabilistic Anaphora Resolution Algorithms.