Showing papers on "Sketch recognition published in 1983"
TL;DR: The formal context-free parsing method of Earley is examined and shown to suggest a useful control structure model for integrating top-down and botton-up search in schemata representations.
Abstract: This paper is concerned with generalizing formal recognition methods from parsing theory to schemata knowledge representations. Within Artificial Intelligence, recognition tasks include aspects of natural language understanding, computer vision, episode understanding, speech recognition, and others. The notion of schemata as a suitable knowledge representation for these tasks is discussed. A number of problems with current schemata-based recognition systems are presented. To gain insight into alternative approaches, the formal context-free parsing method of Earley is examined. It is shown to suggest a useful control structure model for integrating top-down and botton-up search in schemata representations.
7 citations
TL;DR: Although many speech recognition systems have been demonstrated and several commercial products are currently being used in selected environments, computer speech recognition has still very limited capabilities when compared with human performance.
Abstract: This paper provides a review of the state of the art in speech recognition by machine. Although many speech recognition systems have been demonstrated and several commercial products are currently being used in selected environments, computer speech recognition has still very limited capabilities when compared with human performance. Typical component subsystems at the acoustic, phonetic, syntactic, and semantic level are described. The major problems in developing recognition systems and their performance under various conditions are presented.
3 citations
01 Apr 1983
TL;DR: The architecture of a machine that exploits this characteristic of speech recognition systems and some Artificial Intelligence programs is presented together with the results of the simulation of one possible implementation.
Abstract: The behavior of many speech recognition systems, and of some Artificial Intelligence programs, can he modeled as a data driven computation in which the "instructions" are complex but inexpensive operations (e.g. the evaluation of the likelihood of a partial sentence hypothesis), and the "data-flow graph" is derived directly from the knowledge representation (e.g. the phoneme-level network in a Harpy system). The architecture of a machine that exploits this characteristic is presented together with the results of the simulation of one possible implementation.