Sri International

Ordinate Corporation

Other

Jared C Bernstein

Hy Murveit

Michael H Cohen

Mitchel Weintraub

Patti J Price

Brent Townshend

Dimitry Rtischev

John W Butzberger

George T Chen

A method for adaptive scoring of responses to constructed response questions is disclosed. Adaptive scoring may be used to apply evaluator time in such a way that a predetermined reliability level is reached with the least possible use of evaluator time, including adjusting the number of response graded and/or the number of evaluators grading each response. A score may be calculated after grading a subset of a test taker's responses to the constructed response questions. A probability or an error estimate is calculated and compared to a threshold value. Grading may be discontinued based on the comparison. A score may be calculated based on a predetermined number of ratings for the test taker's response to a constructed response. A probability that the score is within a predetermined range of what the score would be if all the responses are graded is calculated. If the probability is less than a threshold value, the number of ratings is increased.

Adaptive scoring of responses to constructed response questions

Linguistic and/or extra-linguistic information is extracted from speech signals to provide measures that may then be compared to expected norms, individual baselines or other nominal or numeric criteria (according to particular psychomotor, perceptual, cognitive or emotional constructs) that are required for satisfactory performance of particular tasks, or that indicate a user's psychological or physical state. The user produces the speech signals in the context of a constrained voice-interactive dialog that utilizes prompts chosen such that the expected range of responses will exhibit low linguistic entropy. For example, the prompts may be interpreted by the user as requests for information, requests to read or repeat or paraphrase a word, sentence, or larger linguistic unit, requests to draw an inference, requests to complete, or identify elements in graphic or verbal aggregates (e.g., pictures or discourses), as examples to imitate, or any similar graphical or verbal presentation that conventionally serves as a prompt to speak. The display is presented though a device either integral or peripheral to a computer system, such as a local or remote video display terminal or telephone.

Method and apparatus for estimating fitness to perform tasks based on

A computer system with a speech recognition component provides a method and apparatus for instructing and evaluating the proficiency of human users in skills that can be exhibited through speaking. The computer system tracks linguistic, indexical and paralinguistic characteristics of the spoken input of users, and implements games, data access, instructional systems, and tests. The computer system combines characteristics of the spoken input automatically to select appropriate material and present it in a manner suitable for the user. In one embodiment, the computer system measures the response latency and speaking rate of the user and presents its next spoken display at an appropriate speaking rate. In other embodiments, the computer system identifies the gender and native language of the user, and combines that information with the relative accuracy of the linguistic content of the user's utterance to select and display material that may be easier or more challenging for speakers with these characteristics.

Method and apparatus for combining information from speech signals for

Spoken-language instruction method and apparatus employ context-based speech recognition for instruction and evaluation, particularly language instruction and language fluency evaluation. A system can administer a lesson, and particularly a language lesson, and evaluate performance in a natural interactive manner while tolerating strong foreign accents, and produce as an output a reading quality score. A finite state grammar set corresponding to the range of word sequence patterns in the lesson is employed as a constraint on a hidden Markov model (HMM) search apparatus in an HMM speech recognizer which includes a set of hidden Markov models of target-language narrations produced by native speakers of the target language. The invention is preferably based on use of a linguistic context-sensitive speech recognizer. The invention includes a system with an interactive decision mechanism which employs at least three levels of error tolerance to simulate a natural level of patience in human-based interactive instruction. A system for a reading phase is implemented through a finite state machine having at least four states which recognizes reading error at any position in a script and which employs a first set of actions. A related system for an interactive question phase is implemented through a finite state machine, but which recognizes reading errors as well as incorrect answers while invoking a second set of actions. A linguistically-sensitive utterance endpoint detector is provided for judging termination of a spoken utterance to simulate human turn-taking in conversational speech.

Method and apparatus for voice-interactive language instruction

An automatic speech recognition methodology, wherein words are modeled as probabilistic networks of allophones, collects nodes in the probabilistic network into equivalence classes when those nodes have the same allophonic choices governed by the same phonological rules. The allophonic choices allow for representation of dialectic pronunciation variations between different speakers. Training data is shared among nodes in an equivalence class so that accurate pronunciation probabilities may be determined even for words for which there is only a limited amount of training data. A method is used to determine probabilities for each of a multitude of pronunciation models for each word in the vocabulary, based on automatic extraction of linguistic knowledge from sets of phonological rules, in order to robustly and accurately model dialectal variation.

Method for recognizing speech using linguistically-motivated hidden

An automatic speech recognition methodology takes advantage of linguistic constraints wherein words are modeled as probabilistic networks of phonetic segments (herein phones), and each phone is represented as a context-independent hidden Markov phone model mixed with a number of context-dependent phone models. Recognition is based on use of methods to design phonological rule sets based on measures of coverage and overgeneration of pronunciations which achieves high coverage of pronunciations with compact representations. Further, a method estimates probabilities of the different possible pronunciations of words. A further method models cross-word coarticulatory effects. In a specific embodiment of the system, a specific method determines the single most-likely pronunciation of words. In further specific embodiments of the system, methods generate speaker-dependent pronunciation networks.

Growing community of inventors

Palo Alto, CA, United States of America

Jared C Bernstein