US 7,574,362 B2
Method for automated sentence planning in a task classification system
Marilyn A. Walker, Morristown, N.J. (US); Owen Christopher Rambow, Hoboken, N.J. (US); and Monica Rogati, Pittsburgh, Pa. (US)
Assigned to AT&T Intellectual Property II, L.P., New York, N.Y. (US)
Appl. No. 10/258,849
PCT Filed Mar. 11, 2002, PCT No. PCT/US02/07236
§ 371(c)(1), (2), (4) Date Oct. 29, 2002,
PCT Pub. No. WO02/073598, PCT Pub. Date Sep. 19, 2002.
Claims priority of provisional application 60/275653, filed on Mar. 14, 2001.
Prior Publication US 2004/0098245 A1, May 20, 2004
This patent is subject to a terminal disclaimer.
Int. Cl. G10L 11/00 (2006.01); G10L 21/00 (2006.01)
U.S. Cl. 704—275  [704/270] 33 Claims
OG exemplary drawing
 
1. A method for sentence planning in a task classification system that interacts with a user, comprising:
recognizing symbols in a user's single input communication to a task classification system;
determining whether the user's input communication can be understood, wherein if the user's communication can be understood, understanding data is generated;
generating a plurality of communicative goals based on the recognized symbols and understanding data, the generated plurality of communicative goals being related to information needed to be obtained from the user;
in response to information from the user's single input communication:
generating a plurality of sentence plans based on the plurality of generated communicative goals, each sentence plan in the plurality of sentence plans being a realization comprising elementary speech acts each corresponding to a respective communicative goal and combined into at least one complete sentence that accomplishes the plurality of communicative goals, and wherein each sentence plan of the plurality of sentence plans is a viable and potentially usable prompt in response to the user's single input communication;
independent of the user, ranking the plurality of generated sentence plans; and
outputting at least one of the ranked sentence plans to the user as a response to the user's single input communication such that one dialog turn occurs starting with the user's single input communication and ending with the outputted sentence plan.