Scalable Micro-planned Generation of Discourse from Structured Data
We present a framework for generating natural language description from structured data such as tables. Motivated by the need to approach this problem in a manner that is scalable and easily adaptable to newer domains, unlike existing related systems, our system does not require parallel data; it rather relies on monolingual corpora and basic NLP tools which are easily accessible. The system employs a 3-staged pipeline that: (i) converts entries in the structured data to canonical form, (ii) generates simple sentences for each atomic entry in the canonicalized representation, and (iii) combines the sentences to produce a coherent, fluent and adequate paragraph description through sentence compounding and co-reference replacement modules. Experiments on a benchmark mixed-domain dataset curated for paragraph description from tables reveals the superiority of our system over existing data-to-text approaches. We also demonstrate the robustness of our system in accepting other data types such as Knowledge-Graphs and Key-Value dictionaries.
READ FULL TEXT