A group of researchers from IIIT Delhi, NSUT Delhi, IIIT Hyderabad, and IIT Bombay have developed a multilingual tool called IndIE for extracting valuable facts from natural language text in open domains. They have also established a benchmark called Hindi-BenchIE to evaluate the tool’s effectiveness in the Hindi language.
IndIE has been evaluated against golden triples extracted from 112 Hindi sentences and has shown the ability to generalize across different languages. It is believed that IndIE can generate meaningful triples for languages like Urdu, Tamil, and Telugu, which share common dependency relations with Hindi.
Professor PonnurangamKumaraguru from IIIT Hyderabad explained that IndIE supports languages based on the intersection of languages supported by the stanza library and languages with significant dependency overlap with Hindi. This includes Tamil, Telugu, and Urdu. IndIE is not recommended for Marathi due to training data limitations.
The most common dependency relations used by IndIE are subject and object relations, which help identify the head and tail of a triple. The potential applications of IndIE are similar to other OIE tools, as it can be used to build knowledge graphs from unstructured text.
The researchers believe that IndIE fills the gap in publicly available tools for Indic languages and can serve as a one-stop solution for multilingual OIE methods. They are optimistic that IndIE will inspire further work in the field of multilingual OIE.
