Code

NeuralAmr as described in:

  • Ioannis Konstas, Srinivasan Iyer, Mark Yatskar, Yejin Choi, Luke Zettlemoyer. 2017. Neural AMR: Sequence-to-Sequence Models for Parsing and Generation, ACL 2017.

Joint Concept-to-Text Generator as described in:

  • Ioannis Konstas. 2014. Joint Models for Concept-to-Text Generation, PhD Thesis, University of Edinburgh.

PLTAG parser as described in:

  • Vera Demberg, Frank Keller and Alexander Koller, 2013. Incremental, Predictive Parsing with Psycholinguistically Motivated Tree-Adjoining Grammar, Computational Linguistics, Vol. 39, No. 4, pages 1025-1066. [pdf]
  • Ioannis Konstas, Frank Keller, Vera Demberg and Mirella Lapata. 2014. Incremental Semantic Role Labeling with TAG, EMNLP'14.

ATIS Corpus

This is the ATIS version following a record-field-value schema as described in:

  • Ioannis Konstas and Mirella Lapata. 2012. Unsupervised concept-to-text generation with hypergraphs, NAACL'12.
  • Ioannis Konstas and Mirella Lapata. 2012. Concept-to-text Generation via Discriminative Reranking, ACL 2012.
The mapping to the relational schema from the lambda version of the dataset (Zettlemoyer and Collins, ACL'07), has been produced automatically as explained here (Section 3.2.3).
We used the original split into 4962 training and 448 testing examples. You may download the corpus here.

WinHelp Corpus

This is the WinHelp version following a record-field-value schema as described in:

  • Ioannis Konstas and Mirella Lapata. 2013. Inducing Document Plans for Concept-to-Text Generation, EMNLP '13.
The annotation is explained here (Section 3.2.4).
We provide the full dataset (128 documents), and the 10-fold split used in the paper. You may download the corpus here.

WeatherGov Corpus

This is the WeatherGov corpus following a record-field-value schema as described in:

Note that we use the same split with Gabor Angeli, and have collated all individual files from the original dataset to 3 single files. You may download the corpus here.

Robocup Corpus

This is the Robocup corpus following a record-field-value schema as described in:

You may download the corpus here.