Native Language Processing using Exegy Text Miner
Lawrence Livermore National Laboratory's New Architectures Testbed recently evaluated Exegy's Text Miner appliance to assess its applicability to high-performance, automated native language analysis. The evaluation was performed with support from the Computing Applications and Research Department in close collaboration with Global Security programs, and institutional activities in native language analysis. The Exegy Text Miner is a special-purpose device for detecting and flagging user-supplied patterns of characters, whether in streaming text or in collections of documents at very high rates. Patterns may consist of simple lists of words or complex expressions with sub-patterns linked by logical operators. These searches are accomplished through a combination of specialized hardware (i.e., one or more field-programmable gates arrays in addition to general-purpose processors) and proprietary software that exploits these individual components in an optimal manner (through parallelism and pipelining). For this application the Text Miner has performed accurately and reproducibly at high speeds approaching those documented by Exegy in its technical specifications. The Exegy Text Miner is primarily intended for the single-byte ASCII characters used in English, but at a technical level its capabilities are language-neutral and can be applied to multi-byte character sets such as those found in Arabic and Chinese. The system is used for searching databases or tracking streaming text with respect to one or more lexicons. In a real operational environment it is likely that data would need to be processed separately for each lexicon or search technique. However, the searches would be so fast that multiple passes should not be considered as a limitation a priori. Indeed, it is conceivable that large databases could be searched as often as necessary if new queries were deemed worthwhile. This project is concerned with evaluating the Exegy Text Miner installed in the New Architectures Testbed running under software version 2.0. The concrete goals of the evaluation were to test the speed and accuracy of the Exegy and explore ways that it could be employed in current or future text-processing projects at Lawrence Livermore National Laboratory (LLNL). This study extended beyond this to evaluate its suitability for processing foreign language sources. The scope of this study was limited to the capabilities of the Exegy Text Miner in the file search mode and does not attempt simulating the streaming mode. Since the capabilities of the machine are invariant to the choice of input mode and since timing should not depend on this choice, it was felt that the added effort was not necessary for this restricted study.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 922325
- Report Number(s):
- UCRL-TR-235781; TRN: US200806%%227
- Country of Publication:
- United States
- Language:
- English
Similar Records
STAR: Software Toolkit for Analysis Research
Integrated computer control system CORBA-based simulator FY98 LDRD project final summary report