Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

PHRED: A Generator for Natural Language Interfaces

Paul S. Jacobs

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-84-198
September 1984

http://www.eecs.berkeley.edu/Pubs/TechRpts/1984/CSD-84-198.pdf

PHRED (PHRasal English Diction) is a natural language generator designed for use in a variety of domains. It was constructed to share a knowledge base with PHRAN (PHRasal ANalyzer) as part of a real-time user-friendly interface. The knowledge base consists of pattern-concept pairs, i.e., associations between linguistic structures and conceptual templates. Using this knowledge base, PHRED produces appropriate and grammatical natural language output from a conceptual representation.

PHRED and PHRAN are currently used as central components of the user interface to the UNIX Consultant System (UC). This system answers questions and solves problems related to the UNIX operating system. UC passes the conceptual form of its responses, usually either questions or answers to questions, to the PHRED generator, which expresses them in the user's language. Currently the consultant can answer questions and produce its responses in either English or Spanish.

There are a number of practical advantages to PHRED as the generation component of a natural language system. Having knowledge base shared between analyzer and generator eliminates the inevitable redundancy of having separate grammars and lexicons for input and output. It avoids possibly awkward inconsistencies caused by such a separation, and allows for interchangeable interfaces, such as the English and Spanish versions of the UC interface.

The phrasal approach to language processing realized in PHRED has proven helpful in generation as in analysis. PHRED commands the use of idioms, grammatical constructions, and canned phrases without a specialized mechanism or data structure. It does so without restricting its ability to utilize more general linguistic knowledge.

While PHRED affords extensibility, simplicity, and processing speed, its design incorporates a cognitive motivation as well. It diverges from the traditional computational approach to language as a symbol manipulation process, and treats it more as an associative process among knowledge structures. The phrasal approach minimizes the autonomy of the individual word, the bane of some Artificial Intelligence approaches to language. The treatment of most linguistic knowledge as declarative bears cognitive as well as practical significance. The two-stage process used by PHRED to select appropriate linguistic structures also fits well with cognitive theories of language and memory.


BibTeX citation:

@techreport{Jacobs:CSD-84-198,
    Author = {Jacobs, Paul S.},
    Title = {PHRED: A Generator for Natural Language Interfaces},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {1984},
    Month = {Sep},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/1984/5920.html},
    Number = {UCB/CSD-84-198},
    Abstract = {PHRED (PHRasal English Diction) is a natural language generator designed for use in a variety of domains.  It was constructed to share a knowledge base with PHRAN (PHRasal ANalyzer) as part of a real-time user-friendly interface.  The knowledge base consists of pattern-concept pairs, i.e., associations between linguistic structures and conceptual templates.  Using this knowledge base, PHRED produces appropriate and grammatical natural language output from a conceptual representation.  <p>  PHRED and PHRAN are currently used as central components of the user interface to the UNIX Consultant System (UC). This system answers questions and solves problems related to the UNIX operating system. UC passes the conceptual form of its responses, usually either questions or answers to questions, to the PHRED generator, which expresses them in the user's language.  Currently the consultant can answer questions and produce its responses in either English or Spanish.  <p>  There are a number of practical advantages to PHRED as the generation component of a natural language system.  Having knowledge base shared between analyzer and generator eliminates the inevitable redundancy of having separate grammars and lexicons for input and output.  It avoids possibly awkward inconsistencies caused by such a separation, and allows for interchangeable interfaces, such as the English and Spanish versions of the UC interface.  <p>  The phrasal approach to language processing realized in PHRED has proven helpful in generation as in analysis. PHRED commands the use of idioms, grammatical constructions, and canned phrases without a specialized mechanism or data structure.  It does so without restricting its ability to utilize more general linguistic knowledge.  <p>  While PHRED affords extensibility, simplicity, and processing speed, its design incorporates a cognitive motivation as well. It diverges from the traditional computational approach to language as a symbol manipulation process, and treats it more as an associative process among knowledge structures.  The phrasal approach minimizes the autonomy of the individual word, the bane of some Artificial Intelligence approaches to language.  The treatment of most linguistic knowledge as declarative bears cognitive as well as practical significance.  The two-stage process used by PHRED to select appropriate linguistic structures also fits well with cognitive theories of language and memory.}
}

EndNote citation:

%0 Report
%A Jacobs, Paul S.
%T PHRED: A Generator for Natural Language Interfaces
%I EECS Department, University of California, Berkeley
%D 1984
%@ UCB/CSD-84-198
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/1984/5920.html
%F Jacobs:CSD-84-198