Reducing Search by Partitioning the Word Network
نویسنده
چکیده
Information is passed between the two systems using the Word Network (or Word Lattice) which is a set of word-score pairs together with the start and end points for each word. The word network is organized as a directed acyclic graph, whose arcs are labeled as word-score pairs, and whose nodes are moments in time. The recognition problem is to find the best scoring grammatical sequence of words from the graph's begln-point to its end. In our experments, we analyzed word networks from the TIMIT domain and the Resource Management domain, constructed without using any language model. A spoken language system requires the coupling of speech recognition capabilities with a capability for understanding the meaning of the utterance as provided by a natural language understanding module. The architecture of such a system is an active topic of research. One approach is to simply couple the two components sequentially: the speech recognizer converts the speech signal into a sequence of words, and then the natural language system attempts to understand those words. The criticism of this approach is that natural language system has no opportunity to correct the errors made by the recog-nizer; nor can it help in controlling the search space. A variant to the this approach is to have the speech recognizer produce a ranked set of candidate sentences. These sentences are then be processed by the natural language system, and the best scoring grammatical (or meaningful) sentence accepted. Alternatively, the two systems can be integrated at a much deeper level, allowing natural language constraints to provide assistance to the recognizer by scoring sentence prefixes as they are produced by the recognizer; thls would have the effect of using the grammar to prune the search space, while also producing a representation of the meaning of the utterance. This latter approach-close interleaving of the speech and natural language components-is our architectural goal. However, in order to test the effectiveness of these ideas, we have experimented wlth a sequential coupling of the two systems where the speech recognizer produces a word network, from which candidate sentences are extracted and processed by the natural language component. Our initial problem in coupling the MIT speech recognition system with the Unlsys PUNDIT natural language system was to define an appropriate interface for experimentation. The MIT speech recognizer uses a method called Viterbi Search to find the highest scoring sentence. This search strategy is able …
منابع مشابه
Solving the tandem AGV network design problem using tabu search: Cases of maximum workload and workload balance with fixed and non-fixed number of loops
A tandem AGV configuration connects all cells of a manufacturing area by means of non-overlapping, sin-gle-vehicle closed loops. Each loop has at least one additional P/D station, provided as an interface between adjacent loops. This study describes the development of three tabu search algorithms for the design of tandem AGV systems. The first algorithm was developed based on the basic definiti...
متن کاملOptimal reconfiguration of radial distribution system with the aim of reducing losses and improving voltage profiles using the improved lightning search algorithm
In this paper, a modified version of the lightning search algorithm is proposed in order to find the optimal reconfiguration of the switches and locate and determine the optimal capacity of distributed generation sources in the distribution feeder. The main optimization goals are to reduce ohmic losses and voltage deviations in the standard 33-bus and 94-node IEEE feeders. The simulation result...
متن کاملENERGY AWARE DISTRIBUTED PARTITIONING DETECTION AND CONNECTIVITY RESTORATION ALGORITHM IN WIRELESS SENSOR NETWORKS
Mobile sensor networks rely heavily on inter-sensor connectivity for collection of data. Nodes in these networks monitor different regions of an area of interest and collectively present a global overview of some monitored activities or phenomena. A failure of a sensor leads to loss of connectivity and may cause partitioning of the network into disjoint segments. A number of approaches have be...
متن کاملA new memetic algorithm for mitigating tandem automated guided vehicle system partitioning problem
Automated Guided Vehicle System (AGVS) provides the flexibility and automation demanded by Flexible Manufacturing System (FMS). However, with the growing concern on responsible management of resource use, it is crucial to manage these vehicles in an efficient way in order reduces travel time and controls conflicts and congestions. This paper presents the development process of a new Memetic Alg...
متن کاملIntellectual Structure of Knowledge in Information Behavior: A Co-Word Analysis
Background and Aim: The intellectual structure of knowledge and its research front can be identified by co-word analysis. This research attempts to reveal the intellectual structure of knowledge in information behavior inquiries, via co-word, network analysis, and science visualization tools. Methods: Bibliometric methodology and social network analysis are used. Population comprises 2146 recor...
متن کامل