Paper Reading #24: A Natural Language Interface of Thorough Coverage by Concordance with Knowledge Bases

Commentary

See what I have to say about ___'s and ___'s work.

References

Han, Y., et al. (2010). A natural language interface of thorough coverage by concordance with knowledge bases. Proceeding of the Acm conference on intelligent user interfaces. Hong Kong: http://www.iuiconf.org/

Article Summary

In this paper, Han et al. discuss a novel approach to solving a common problem with natural language interface (NLI) systems. As illustrated below, of the total set of expressions a user might possibly input, there is a partial disparity between those expressions that a given system can interpret and those which a given knowledge base can answer.

Image courtesy of the above-cited article.

Whereas most NLI systems try to make up the difference by expanding the number of expressions interpretable by the system, this team proposed to generate the interpretable expressions from the expressions answerable by the knowledge base instead. They identify several levels of classification based on a graph representation of the knowledge base, and generate all queries for a given level that are answerable. They then cast user expressions as one of the evaluable expressions by a similarity measure.

Discussion

I don't know how many any NLI systems work, but honestly, this approach, while (purportedly) novel, does not seem like a great feat of science and research. Maybe it is, I don't know: like I said, I know nothing about NLI systems. But I feel like I might possibly have stumbled upon this concept if you gave me the image above and described the problem to me. I'm just saying. That being said, was 2010 really the first time that anyone has thought to cast the problem like this, or was that just the first time anyone had thought to publish a paper on the topic? It just really seems like this specific problem is one that computer science as an all-encompassing entity would have solved a long time ago. Please, rebuke me, correct me, enlighten me if you can.