1 / 8

Release Progress Report

min. Release Progress Report. XML. ABNF. IHD. BNF. JSGF. BNF. Daniel May Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering. Overview. Progress Since Last Presentation LanguageModelJSGF LanguageModel Utilities

nunezm
Télécharger la présentation

Release Progress Report

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. min Release Progress Report XML ABNF IHD BNF JSGF BNF Daniel May Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering

  2. Overview • Progress Since Last Presentation • LanguageModelJSGF • LanguageModel Utilities • HiddenMarkovModel • Outstanding Issues • Updated Release Plan

  3. LanguageModeJSGF • Changes took a lot longer than expected due to the poor condition of some of the related classes. • JSGFParser and JSGFToken required substantial changes. • The previous implementation did not convert IHD to JSGF tokens before writing the JSGF language model to a file. • Instead, the conversion sequence was • JSGF → JSGF Tokens → IHD • IHD → JSGF • In order to support the new LanguageModel classes, we needed the conversion sequence • JSGF → JSGF Tokens → ABNF → BNF → IHD • IHD → BNF → JSGF Tokens → JSGF • The highlighted conversions were not previously implemented. • These additional changes delayed completion of the LanguageModelJSGF class.

  4. LanguageModel Utilities • isip_network_converter • Finished. The only changes for the release were the addition of support for XML, ABNF, and BNF Language Models, and a minimize option. • isip_network_builder • Finished. Support for loading and opening BNF, ABNF, and XML language models has been added. • isip_lm_tester • Won has implemented the language model parsing capability and has tested it with a set of test cases. • Seungchan corrected a few sentence generation problems. These include removing the dummy symbols from generated sentences and only counting symbols at the top level of the grammar. He has also tested his changes with a set of test cases. • All low-level code has been moved to the HierarchicalSearch class. Previously, the low-level, pointer intensive sentence generation code was part of the utility source.

  5. HiddenMarkovModel • Initial changes to work with the new LanguageModel classes included the addition of a HierarchicalDigraph object to the member data. • Previously, the HierarchicalDigraph object was accessed via the general LanguageModel object. • In the new LanguageModel classes, the general LanguageModel class does not have a HierarchicalDigraph as member data, but instead stores the grammar information in a RuleModel object. • When a HierarchicalDigraph object was added to the member data of HiddenMarkovModel, we started having strange memory problems that resulted in segmentation faults.

  6. HiddenMarkovModel • To work around this, a LanguageModelIHD object was added to the member data of HiddenMarkovModel and the HierarchicalDigraph object was removed. • When the HierarchicalDigraph object is accessed via LanguageModelIHD, the problems disappeared. • The root of the problem is undoubtedly related to the pointer intensive nature of the search classes. • The recognizer has been tested using a set of word model test cases I put together a long time ago. The hypotheses files for IHD and JSGF match, but there are slight differences in the XML hypothesis scores.

  7. Outstanding Issues • Temporary file support for isip_transform. Sridhar has a working solution and is currently working to ensure that the temporary files that are generated will be 100% unique between processors. • Changes to the front end to allow MatrixFloat objects as input. Sundar is working on this. • XML weights issue related to the weights on the arcs extending from the start node. I’ve asked Wesley to take a look at this. • Updating diagnose methods for LanguageModel classes. Won and I are working on this. • Running Purify on all our utilities. Initial tries have been problematic, but Seungchan is investigating. • Regression testing.

  8. Updated Release Schedule • April 14th • All code checked in. • All documentation updated. • All diagnose methods updated. • Everything passes ‘make test’. • April 14th – May 1st • Regression testing. • Arbitrary Grammar Tests • ‘Purification’ of IFCs. • May 1st • Release Date

More Related