parajain / structscribe

Resources and code for: Scalable Micro-planned Generation of Discourse from Structured Data

Major dependencies

• Python
• Pytorch
• Spacy

Dataset

• Link to download dataset: link
• Above link contains: Dataset 1, 2 and 3 as describes in the paper

Setup DBpedia Lookup

• You can use online API (http://lookup.dbpedia.org/api/search/KeywordSearch?QueryString=Albert%20Einstein), but it will be faster (and recommended) to host the API locally.
• Follow instruction: https://github.com/dbpedia/lookup
• Change api link. code/lookupner/dbpedia_ner.py

For experimentation purpose models are directly loaded in the memory. But there is an option to create an rpc server to create an API (this is slower, but to try our different inputs it is useful). Not needed to run experiment.py

Generation server

• To run: python S2SServer.py configurations/s2s.config.json

Gender predictor module server

• Gender server code is present in code\anaphora_replacer
• To run: python GenderServer.py