Robust Interactive Dialogue Interpretation

Created by W.Langdon from gp-bibliography.bib Revision:1.8081

@PhdThesis{rose:thesis,

• author = "Carolyn Penstein Rose",
• title = "Robust Interactive Dialogue Interpretation",
• school = "Language Technologies Insititute, Carnegie Mellon University",
• year = "1997",
• address = "USA",
• note = "Tech. Rept. CMU-LTI-97-151",
• keywords = "genetic algorithms, genetic programming",
• URL = "http://www.lti.cs.cmu.edu/Research/Thesis/CarolynRose97.pdf",
• size = "273 pages",
• abstract = "ROSE is pronounced Rose like the wine

  The first essential task of a natural language interface is to map the users utterance onto some meaning representation which can then be used for further processing The three biggest challenges which continue to stand in the way of accomplishing even this basic task are extragrammaticality ambiguity and recognition errors In this dissertation I address the issue of how to handle the problem of extragrammaticality efficiently where extragrammaticality is defined as any deviation of an input string from the coverage of a given systems parsing grammar A useful analogy can be made between humancomputer interaction through a natural language interface and language interaction between speakers of different languages with a small shared language base Humans who share a very small language base are able to communicate when the need arises by simplifying their speech patterns and negotiating until they manage to transmit their ideas to one another Hatch As the speaker is speaking the listener throws his net in order to catch those fragments of speech which are comprehensible to him which he then attempts to fit together semantically His subsequent negotiation with the speaker builds upon this partial understanding

  The approach presented here is based on this same model. The ROSE approach RObustness with Structural Evolution repairs extra-grammatical input in two stages The first stage Repair Hypothesis Formation is responsible for assembling a set of hypotheses about the meaning of the ungrammatical utterance This stage is itself divided into two steps Partial Parsing and Combination The Partial Parsing step is similar to the concept of the listener casting his net for comprehensible fragments of speech Lavies GLR* parser Lavie Lavie and Tomita is used to obtain an analysis of islands of the speakers sentence in cases where it is not possible to obtain an analysis for the entire sentence In the Combination step the fragments from the partial parse are assembled into a set of alternative meaning representation hypotheses A genetic programming approach is used to search for different ways to combine the fragments in order to avoid requiring any handcrafted repair rules In ROSEs second stage Interaction with the user the system generates a set of queries negotiating with the speaker in order to narrow down to a single best meaning representation hypothesis

  The primary objective of the ROSE approach is to handle the problem of extra grammaticality in an effective and efficient way The most straightforward way to evaluate different approaches to handling extragrammaticality is by comparing them based on im provement in terms of percentage of sentences handled correctly or improvement of overall accuracy on a particular corpus However it is misleading to compare instantiations of different approaches this way since in theory many of these approaches have the potential for yielding the same amount of improvement given sufficient resources in terms of space both static and dynamic time both development time and run time and interactional effort The real question is which approach can use these resources most economically

  I argue that the ROSE approach of separating the Partial Parsing and Combination steps is more efficient than placing the full burden of robustness on a single parsing algorithm An analogous tradeoff in human-human communication would be the casting and combining model versus one in which the listener tries to construct a complete syntac tic analysis for a sentence outside of his language competence Though humans are known to make a mental note of grammatical features that they are not able to process correctly most of them are regarded mainly as noise Hatch",

• abstract = "Another goal of this work is to demonstrate that it is more efficient to separate Repair Hypothesis Formation from User Interaction rather than interleaving them In other words a set of alternative ways of fitting the whole set of fragments from the partial parse Global Repair Hypotheses is constructed before any queries are generated rather than generating a query to verify each repair step Local Repair Hypotheses Besides being more efficient this approach is arguably more effective When humans collaborate for the purpose of understanding they direct their questions towards information which is necessary for accomplishing their task Clark and Schaefer Clark and Wilkes-Gibbs When questions are directed at clarifying the speakers meaning rather than furthering the shared task speakers become agitated Garfinkel By delaying the interaction until hypotheses about the speakers whole meaning are formed it is possible to focus the interaction on the task level rather than on the language level

  Therefore it will be shown that the ROSE approach robustly extracts the meaning from the users extragrammatical utterance efficiently and without placing an undue burden on the user in terms of interactional effort Finally because the ROSE approach does not rely on any hand crafted repair rules or additional knowledge sources it is a completely general and portable solution",

• notes = "Thu, 26 Feb 1998 16:03:59 EST Eventually when I get around to clearing off some space in my unix account, I plan to put my dissertation on the WWW, but in the mean time, you can order one from CMU. You can email Debra Germany at debra@cs.cmu.edu.

  ",

}

Genetic Programming entries for Carolyn Penstein Rose

Citations