Be the first to post a text review of Thot toolkit for SMT. Rate and review a project by clicking thumbs up or thumbs down in the right column.

• thot 2.2.0Beta file released: Thot-2.2.0Beta.tar.gz

  -------------------------------------------------------------------- Thot 0.90 26th May 2004 => First working code. Models estimated from alignment vectors -------------------------------------------------------------------- Thot 0.91 1st Jun 2004 => Estimation from alignment matrices -------------------------------------------------------------------- Thot 0.92 6th Jul 2004 - Added a tool to perform operations between alignment matrices - First version of the pML estimation - Basic C++ phrase-model library implemented -------------------------------------------------------------------- Thot 0.93 22th Jan 2005 => Several changes from the previous version: - Ported to gcc 3.3.2 - Faster pML estimation - Added bilingual segmentation functionality - Sum between alignment matrices - Estimation of phrase length models - Bug fixes -------------------------------------------------------------------- Thot 0.94 5th May 2005 => Modifications in the pML estimation (normalize by segmentation size) - bug fixes -------------------------------------------------------------------- Thot 0.95 27th Jul 2005 - Code reorganization (modularity increased) - Some documentation added - Changes in command-line options - Output in pharaoh-v123 input format -------------------------------------------------------------------- Thot 0.96 2nd Sep 2005 - Use of GNU tools: automake and autoconf - Ported to Windows using cygwin - Ported to gcc 4.0.0 - bug fixes -------------------------------------------------------------------- Thot 0.97a 28th Sep 2005 - PhraseCounts now implemented as a trie - PhraseTModel implemented as a composition of PhraseCounts and PhraseDic classes - bug fixes -------------------------------------------------------------------- Thot 0.97b 19th Oct 2005 - Deep reorganization of the PhraseTModel class - Changes in PhraseDict class statistics - Added the scripts getModelStats.sh and convGizaAligFile.pl ------------------------------------------------------------------- Thot 0.97c 28th Nov 2005 - PhraseDict reimplemented as a trie - The -pc option now prints the count of (f,e) and the count of e - The -ms option allows to impose the maximum phrase-length constraint over the source phrase -------------------------------------------------------------------- Thot 0.98 17th Jan 2006 - PhraseExtractionCell is reimplemented using a trie structure. This change reduces significantly the memory required by the pseudo_ml estimation process - "-mc" option (used during pml estimation) now is given in millions of segmentations - The maximum sentence length is now equal to 101 words -------------------------------------------------------------------- Thot 0.99 29th Jan 2006 - Both, pml-estimation and the extraction of the best phrase alignments now do not require extra memory to work. Because of this, it is not necessary to prune the segmentations - "-mc" option is no longer used unless the "configure" script is executed with the "--disable-iterators" option -------------------------------------------------------------------- Thot 1.0.0 Beta 25th May 2006 - thot is able to train models of an arbitrary size accessing hard disk - Both "alig_op" and "thot" programs are able to handle large files (>2 GB) - PhraseDict and PhraseCounts are implemented using the TrieOfWords data structure - PhraseTable inherits from the abstract base class BasePhraseTable - thot program is now a shell-script which internally calls the binary "thot_ker" - The name of the "alignManip" tool is changed to "alig_op" - The verbosity of "alig_op" has been increased - The name of the file extensions is changed: .alig -> .A3.final .model -> .ttable .segmLengthTable -> .seglentable - A new class is created in order to load, print and access segmentation-length tables (SegLenTable) - Some functionality implemented by the PhraseModel_From_WBA class is moved to the PhraseModel class - awkInputString class is removed from awkInputStream.h and awkInputStream.cc files - Ported to FreeBsd -------------------------------------------------------------------- Thot 1.1.0 Beta 04th Jul 2006 - Fixed a bug in the Bitset class - Simple tests can be executed by typing "make check" - Portability increased, the package was compiled (but not deeply tested) in the following platforms: * [x86] Linux 2.6 (Debian 3.1) * [x86] NetBSD (2.0.2) * [AMD64] Linux 2.6 (Fedora Core 3 on AMD64 Opteron) * [Alpha] Linux 2.2 (Debian 3.0) * [Power5 - OpenPower 720] SuSE Enterprise Server 9 * [Sparc - R220] Sun Solaris (9) * [x86] OpenBSD 3.8 * [PPC - G5] MacOS X 10.2 SERVER Edition * [x86] Solaris 9 -------------------------------------------------------------------- Thot 2.0.0 Beta 14th Feb 2007 - Implemented a parallel version of Thot - Incorporated a new output format for the translation tables - Added a utility to prune the translation tables - Improvements in the implementation of the data structures that store the bilingual phrases - Incorporated a shell script to filter translation tables given test corpora to be translated - Fixed a bug when reading the alignments in GIZA++ format - The directory for temporaries used by thot can be given as a parameter - The function "extendModel" is added to the PhraseModel_From_WBA class. "extendModel" allows to incrementally add new phrase pairs to the models given an alignment file in .A3.final format - The output of the conv_giza_alig_file script is improved. In addition, the direction of the alignments can be inverted - Solved a bug in the shell script invert_ttable - Added the shell script gen_seglentable_poisson which allows to generate segmentation-length tables by means of Poisson distributions - Incorporated an interface for the PhraseModel class by means of the BasePhraseModel abstract class -------------------------------------------------------------------- Thot 2.0.1 Beta 8th Jun 2007 - Solved a portability issue with awk which could cause execution errors using thot and pbs_thot -------------------------------------------------------------------- Thot 2.0.2 Beta 17th Jun 2007 - Added a short tutorial for the Thot toolkit (including some brief notes about how to combine Thot with Pharaoh and Moses) - Thot incorporates a new parameter: -c, which allows to prune the translation table given a cutoff value for the count of the source phrases - Solved a bug in alig_op when using the -e option - "-no-invert" parameter of alig_op has changed. The new name is "-no-transpose" - Changed the name of the "flip_ef" utility, now its name is "flip_phr" - Changes in the return values of some functions, using constants defined in the file ErrorDefs.h -------------------------------------------------------------------- Thot 2.1.0 - MAJOR CHANGE: NbestTableNode no longer works with probabilities but with log-probabilities. This affects to the functions getNbestTransFor_s_() and getNbestTransFor_t_() - New option for thot and pbs_thot: -lex. The -lex option incorporates two new score components in the phrase table - The library of phrase models provides faster access to the model probabilities - Better implementation of the OrderedVector class - The tool "filter_ttable" has been incorporated to the distribution. "filter_ttable" is a new tool to filter phrase models in order to apply them in machine translation tasks - The tool "score_phr_ibm" has been incorporated to the distribution. "score_phr_ibm" adds lexical scores for the phrase pairs of a given phrase table. The lexical scores are calculated as the score for the viterbi alignment of IBM 1 model. - "make check" is replaced by "make installcheck" since the tool checks are to be executed after installing the package - Source and target sentences or phrases are no longer noted using the letters 'e' and 'f' respectively, which are replaced by 's' and 't'. The modifications affects the name of variables and functions Functions that have changed their names: a) The scoring functions defined in BasePhraseModel: pf_e_, pe_f_, have changed its names: ps_t_, pt_s_ ... b) Functions to obtain lists of translations The old functions cited above are maintained for backward compatibility Functions that have altered the order in which the source and target sentences are given: a) The functions defined in the "print_func" module b) Some function members declared in the "PhraseModel_From_WBA" and the "PhraseExtractionTable" classes. Three functions which were modified are declared as public functions: PhraseModel_From_WBA::obtainBestAlignment PhraseExtractionTable::extractConsistentPhrases PhraseExtractionTable::obtainPossibleSegmentations - Some data types and functions names have been changed in order to simplify and clarify the code: InvNbestTransTable -> PhraseNbestTransTable TransTableNodeData and InvTransTableNodeData -> PhraseTableNodeData BasePhraseModel::getNbestInvTransFor_t_() -> BasePhraseModel::getNbestTransFor_t_() PhraseDict::getInvTranslationsFor_t_() -> PhraseDict::getTranslationsFor_t_() - Solved a bug in ps_t_ function defined in "PhraseModel.cc" - Solved a bug in the alig_op tool. The bug could appear when using the -f option, specifically when the prefix for the output file contained directory components - The robustness of the alig_op tool has been increased due to a better management of the temporary files that are used internally -------------------------------------------------------------------- Thot 2.1.1 - Solved a portability issue with i/o functions - Bug fixed in the pbs_thot tool (reported by Soren Harder) - Faster compilation - Thot package now splits the library libphrase_model.a into two files: * libphrase_model.a * libnlp_common.a libphrase_model.a depends on libnlp_common.a, so both libraries are needed to access the functionality provided by the package. -------------------------------------------------------------------- Thot 2.2.0 - Solved a memory management bug in the PhraseExtractionTable class - Solved portability issues with 64 bit processors - Solved possible portability problems with the locales - Added the utility "train_giza", which allows to easily train IBM models using the GIZA and the mkcls tools - Added a simple class diagram created with the program "dia" (see doc/class_diag.dia) - The hierarchy of phrase-based model classes has been modified. Two new classes have been added: BaseCountPhraseModel and BaseIncrPhraseModel. The following classes have changed their names: * PhraseModel -> _incrPhraseModel * PhraseModel_From_WBA -> WbaIncrPhraseModel * PhraseModelWBAParameters -> WbaIncrPhraseModelParams - Some function names in BasePhraseModel.h and BaseIncrPhraseModel.h have changed: * pt_s_(const Vector<string> ...) -> strPt_s_(.) * logpt_s_(const Vector<string> ...) -> strLogpt_s_(.) * ps_t_(const Vector<string> ...) -> strPs_t_(.) * logps_t_(const Vector<string> ...) -> strLogps_t_(.) * getTransFor_s_(const Vector<string> ...) -> strGetTransFor_s_(.) * getTransFor_t_(const Vector<string> ...) -> strGetTransFor_t_(.) * getNbestTransFor_s_(const Vector<string> ...) -> strGetNbestTransFor_s_(.) * getNbestTransFor_t_(const Vector<string> ...) -> strGetNbestTransFor_t_(.) * addTableEntry(const Vector<string> ...) -> strAddTableEntry(.) * incrCountsOfEntry(const Vector<string> ...) -> strIncrCountsOfEntry(.) -------------------------------------------------------------------- $Date: 2008-09-06 11:33:46 $

  posted 429 days ago

• File released: /thot/2.2.0Beta/Thot-2.2.0Beta.tar.gz

  posted 429 days ago

• Thot toolkit for SMT: New Release 2.2.0Beta

  New release 2.2.0Beta of the Thot toolkit with some improvements. Some portability issues have been solved.

  posted by danulino 429 days ago

• Thot toolkit for SMT: New Release 2.1.1Beta

  The 2.1.1Beta version of the Thot toolkit has been released. The new version fixes a minor bug and incorporates some improvements.

  posted by danulino 581 days ago

• thot 2.1.1Beta file released: Thot-2.1.1Beta.tar.gz

  -------------------------------------------------------------------- Thot 0.90 26th May 2004 => First working code. Models estimated from alignment vectors -------------------------------------------------------------------- Thot 0.91 1st Jun 2004 => Estimation from alignment matrices -------------------------------------------------------------------- Thot 0.92 6th Jul 2004 - Added a tool to perform operations between alignment matrices - First version of the pML estimation - Basic C++ phrase-model library implemented -------------------------------------------------------------------- Thot 0.93 22th Jan 2005 => Several changes from the previous version: - Ported to gcc 3.3.2 - Faster pML estimation - Added bilingual segmentation functionality - Sum between alignment matrices - Estimation of phrase length models - Bug fixes -------------------------------------------------------------------- Thot 0.94 5th May 2005 => Modifications in the pML estimation (normalize by segmentation size) - bug fixes -------------------------------------------------------------------- Thot 0.95 27th Jul 2005 - Code reorganization (modularity increased) - Some documentation added - Changes in command-line options - Output in pharaoh-v123 input format -------------------------------------------------------------------- Thot 0.96 2nd Sep 2005 - Use of GNU tools: automake and autoconf - Ported to Windows using cygwin - Ported to gcc 4.0.0 - bug fixes -------------------------------------------------------------------- Thot 0.97a 28th Sep 2005 - PhraseCounts now implemented as a trie - PhraseTModel implemented as a composition of PhraseCounts and PhraseDic classes - bug fixes -------------------------------------------------------------------- Thot 0.97b 19th Oct 2005 - Deep reorganization of the PhraseTModel class - Changes in PhraseDict class statistics - Added the scripts getModelStats.sh and convGizaAligFile.pl ------------------------------------------------------------------- Thot 0.97c 28th Nov 2005 - PhraseDict reimplemented as a trie - The -pc option now prints the count of (f,e) and the count of e - The -ms option allows to impose the maximum phrase-length constraint over the source phrase -------------------------------------------------------------------- Thot 0.98 17th Jan 2006 - PhraseExtractionCell is reimplemented using a trie structure. This change reduces significantly the memory required by the pseudo_ml estimation process - "-mc" option (used during pml estimation) now is given in millions of segmentations - The maximum sentence length is now equal to 101 words -------------------------------------------------------------------- Thot 0.99 29th Jan 2006 - Both, pml-estimation and the extraction of the best phrase alignments now do not require extra memory to work. Because of this, it is not necessary to prune the segmentations - "-mc" option is no longer used unless the "configure" script is executed with the "--disable-iterators" option -------------------------------------------------------------------- Thot 1.0.0 Beta 25th May 2006 - thot is able to train models of an arbitrary size accessing hard disk - Both "alig_op" and "thot" programs are able to handle large files (>2 GB) - PhraseDict and PhraseCounts are implemented using the TrieOfWords data structure - PhraseTable inherits from the abstract base class BasePhraseTable - thot program is now a shell-script which internally calls the binary "thot_ker" - The name of the "alignManip" tool is changed to "alig_op" - The verbosity of "alig_op" has been increased - The name of the file extensions is changed: .alig -> .A3.final .model -> .ttable .segmLengthTable -> .seglentable - A new class is created in order to load, print and access segmentation-length tables (SegLenTable) - Some functionality implemented by the PhraseModel_From_WBA class is moved to the PhraseModel class - awkInputString class is removed from awkInputStream.h and awkInputStream.cc files - Ported to FreeBsd -------------------------------------------------------------------- Thot 1.1.0 Beta 04th Jul 2006 - Fixed a bug in the Bitset class - Simple tests can be executed by typing "make check" - Portability increased, the package was compiled (but not deeply tested) in the following platforms: * [x86] Linux 2.6 (Debian 3.1) * [x86] NetBSD (2.0.2) * [AMD64] Linux 2.6 (Fedora Core 3 on AMD64 Opteron) * [Alpha] Linux 2.2 (Debian 3.0) * [Power5 - OpenPower 720] SuSE Enterprise Server 9 * [Sparc - R220] Sun Solaris (9) * [x86] OpenBSD 3.8 * [PPC - G5] MacOS X 10.2 SERVER Edition * [x86] Solaris 9 -------------------------------------------------------------------- Thot 2.0.0 Beta 14th Feb 2007 - Implemented a parallel version of Thot - Incorporated a new output format for the translation tables - Added a utility to prune the translation tables - Improvements in the implementation of the data structures that store the bilingual phrases - Incorporated a shell script to filter translation tables given test corpora to be translated - Fixed a bug when reading the alignments in GIZA++ format - The directory for temporaries used by thot can be given as a parameter - The function "extendModel" is added to the PhraseModel_From_WBA class. "extendModel" allows to incrementally add new phrase pairs to the models given an alignment file in .A3.final format - The output of the conv_giza_alig_file script is improved. In addition, the direction of the alignments can be inverted - Solved a bug in the shell script invert_ttable - Added the shell script gen_seglentable_poisson which allows to generate segmentation-length tables by means of Poisson distributions - Incorporated an interface for the PhraseModel class by means of the BasePhraseModel abstract class -------------------------------------------------------------------- Thot 2.0.1 Beta 8th Jun 2007 - Solved a portability issue with awk which could cause execution errors using thot and pbs_thot -------------------------------------------------------------------- Thot 2.0.2 Beta 17th Jun 2007 - Added a short tutorial for the Thot toolkit (including some brief notes about how to combine Thot with Pharaoh and Moses) - Thot incorporates a new parameter: -c, which allows to prune the translation table given a cutoff value for the count of the source phrases - Solved a bug in alig_op when using the -e option - "-no-invert" parameter of alig_op has changed. The new name is "-no-transpose" - Changed the name of the "flip_ef" utility, now its name is "flip_phr" - Changes in the return values of some functions, using constants defined in the file ErrorDefs.h -------------------------------------------------------------------- Thot 2.1.0 - MAJOR CHANGE: NbestTableNode no longer works with probabilities but with log-probabilities. This affects to the functions getNbestTransFor_s_() and getNbestTransFor_t_() - New option for thot and pbs_thot: -lex. The -lex option incorporates two new score components in the phrase table - The library of phrase models provides faster access to the model probabilities - Better implementation of the OrderedVector class - The tool "filter_ttable" has been incorporated to the distribution. "filter_ttable" is a new tool to filter phrase models in order to apply them in machine translation tasks - The tool "score_phr_ibm" has been incorporated to the distribution. "score_phr_ibm" adds lexical scores for the phrase pairs of a given phrase table. The lexical scores are calculated as the score for the viterbi alignment of IBM 1 model. - "make check" is replaced by "make installcheck" since the tool checks are to be executed after installing the package - Source and target sentences or phrases are no longer noted using the letters 'e' and 'f' respectively, which are replaced by 's' and 't'. The modifications affects the name of variables and functions Functions that have changed their names: a) The scoring functions defined in BasePhraseModel: pf_e_, pe_f_, have changed its names: ps_t_, pt_s_ ... b) Functions to obtain lists of translations The old functions cited above are maintained for backward compatibility Functions that have altered the order in which the source and target sentences are given: a) The functions defined in the "print_func" module b) Some function members declared in the "PhraseModel_From_WBA" and the "PhraseExtractionTable" classes. Three functions which were modified are declared as public functions: PhraseModel_From_WBA::obtainBestAlignment PhraseExtractionTable::extractConsistentPhrases PhraseExtractionTable::obtainPossibleSegmentations - Some data types and functions names have been changed in order to simplify and clarify the code: InvNbestTransTable -> PhraseNbestTransTable TransTableNodeData and InvTransTableNodeData -> PhraseTableNodeData BasePhraseModel::getNbestInvTransFor_t_() -> BasePhraseModel::getNbestTransFor_t_() PhraseDict::getInvTranslationsFor_t_() -> PhraseDict::getTranslationsFor_t_() - Solved a bug in ps_t_ function defined in "PhraseModel.cc" - Solved a bug in the alig_op tool. The bug could appear when using the -f option, specifically when the prefix for the output file contained directory components - The robustness of the alig_op tool has been increased due to a better management of the temporary files that are used internally -------------------------------------------------------------------- Thot 2.1.1 - Solved a portability issue with i/o functions - Bug fixed in the pbs_thot tool (reported by Soren Harder) - Faster compilation - Thot package now splits the library libphrase_model.a into two files: * libphrase_model.a * libnlp_common.a libphrase_model.a depends on libnlp_common.a, so both libraries are needed to access the functionality provided by the package. -------------------------------------------------------------------- $Date: 2008-03-05 16:50:08 $

  posted 581 days ago

• File released: /thot/2.1.1Beta/Thot-2.1.1Beta.tar.gz

  posted 581 days ago

• Thot toolkit for SMT: New Release 2.1.0Beta

  Thot is a toolkit to train phrase-based models for statistical machine translation. Thot allows to estimate phrase-based models and to obtain the best alignments at phrase level for a given set of sentence pairs. Released a new version with several improvements and some extra functionality. See NEWS file to find out more information.

  posted by danulino 662 days ago

• thot 2.1.0Beta file released: Thot-2.1.0Beta.tar.gz

  -------------------------------------------------------------------- Thot 0.90 26th May 2004 => First working code. Models estimated from alignment vectors -------------------------------------------------------------------- Thot 0.91 1st Jun 2004 => Estimation from alignment matrices -------------------------------------------------------------------- Thot 0.92 6th Jul 2004 - Added a tool to perform operations between alignment matrices - First version of the pML estimation - Basic C++ phrase-model library implemented -------------------------------------------------------------------- Thot 0.93 22th Jan 2005 => Several changes from the previous version: - Ported to gcc 3.3.2 - Faster pML estimation - Added bilingual segmentation functionality - Sum between alignment matrices - Estimation of phrase length models - Bug fixes -------------------------------------------------------------------- Thot 0.94 5th May 2005 => Modifications in the pML estimation (normalize by segmentation size) - bug fixes -------------------------------------------------------------------- Thot 0.95 27th Jul 2005 - Code reorganization (modularity increased) - Some documentation added - Changes in command-line options - Output in pharaoh-v123 input format -------------------------------------------------------------------- Thot 0.96 2nd Sep 2005 - Use of GNU tools: automake and autoconf - Ported to Windows using cygwin - Ported to gcc 4.0.0 - bug fixes -------------------------------------------------------------------- Thot 0.97a 28th Sep 2005 - PhraseCounts now implemented as a trie - PhraseTModel implemented as a composition of PhraseCounts and PhraseDic classes - bug fixes -------------------------------------------------------------------- Thot 0.97b 19th Oct 2005 - Deep reorganization of the PhraseTModel class - Changes in PhraseDict class statistics - Added the scripts getModelStats.sh and convGizaAligFile.pl ------------------------------------------------------------------- Thot 0.97c 28th Nov 2005 - PhraseDict reimplemented as a trie - The -pc option now prints the count of (f,e) and the count of e - The -ms option allows to impose the maximum phrase-length constraint over the source phrase -------------------------------------------------------------------- Thot 0.98 17th Jan 2006 - PhraseExtractionCell is reimplemented using a trie structure. This change reduces significantly the memory required by the pseudo_ml estimation process - "-mc" option (used during pml estimation) now is given in millions of segmentations - The maximum sentence length is now equal to 101 words -------------------------------------------------------------------- Thot 0.99 29th Jan 2006 - Both, pml-estimation and the extraction of the best phrase alignments now do not require extra memory to work. Because of this, it is not necessary to prune the segmentations - "-mc" option is no longer used unless the "configure" script is executed with the "--disable-iterators" option -------------------------------------------------------------------- Thot 1.0.0 Beta 25th May 2006 - thot is able to train models of an arbitrary size accessing hard disk - Both "alig_op" and "thot" programs are able to handle large files (>2 GB) - PhraseDict and PhraseCounts are implemented using the TrieOfWords data structure - PhraseTable inherits from the abstract base class BasePhraseTable - thot program is now a shell-script which internally calls the binary "thot_ker" - The name of the "alignManip" tool is changed to "alig_op" - The verbosity of "alig_op" has been increased - The name of the file extensions is changed: .alig -> .A3.final .model -> .ttable .segmLengthTable -> .seglentable - A new class is created in order to load, print and access segmentation-length tables (SegLenTable) - Some functionality implemented by the PhraseModel_From_WBA class is moved to the PhraseModel class - awkInputString class is removed from awkInputStream.h and awkInputStream.cc files - Ported to FreeBsd -------------------------------------------------------------------- Thot 1.1.0 Beta 04th Jul 2006 - Fixed a bug in the Bitset class - Simple tests can be executed by typing "make check" - Portability increased, the package was compiled (but not deeply tested) in the following platforms: * [x86] Linux 2.6 (Debian 3.1) * [x86] NetBSD (2.0.2) * [AMD64] Linux 2.6 (Fedora Core 3 on AMD64 Opteron) * [Alpha] Linux 2.2 (Debian 3.0) * [Power5 - OpenPower 720] SuSE Enterprise Server 9 * [Sparc - R220] Sun Solaris (9) * [x86] OpenBSD 3.8 * [PPC - G5] MacOS X 10.2 SERVER Edition * [x86] Solaris 9 -------------------------------------------------------------------- Thot 2.0.0 Beta 14th Feb 2007 - Implemented a parallel version of Thot - Incorporated a new output format for the translation tables - Added a utility to prune the translation tables - Improvements in the implementation of the data structures that store the bilingual phrases - Incorporated a shell script to filter translation tables given test corpora to be translated - Fixed a bug when reading the alignments in GIZA++ format - The directory for temporaries used by thot can be given as a parameter - The function "extendModel" is added to the PhraseModel_From_WBA class. "extendModel" allows to incrementally add new phrase pairs to the models given an alignment file in .A3.final format - The output of the conv_giza_alig_file script is improved. In addition, the direction of the alignments can be inverted - Solved a bug in the shell script invert_ttable - Added the shell script gen_seglentable_poisson which allows to generate segmentation-length tables by means of Poisson distributions - Incorporated an interface for the PhraseModel class by means of the BasePhraseModel abstract class -------------------------------------------------------------------- Thot 2.0.1 Beta 8th Jun 2007 - Solved a portability issue with awk which could cause execution errors using thot and pbs_thot -------------------------------------------------------------------- Thot 2.0.2 Beta 17th Jun 2007 - Added a short tutorial for the Thot toolkit (including some brief notes about how to combine Thot with Pharaoh and Moses) - Thot incorporates a new parameter: -c, which allows to prune the translation table given a cutoff value for the count of the source phrases - Solved a bug in alig_op when using the -e option - "-no-invert" parameter of alig_op has changed. The new name is "-no-transpose" - Changed the name of the "flip_ef" utility, now its name is "flip_phr" - Changes in the return values of some functions, using constants defined in the file ErrorDefs.h -------------------------------------------------------------------- Thot 2.1.0 - MAJOR CHANGE: NbestTableNode no longer works with probabilities but with log-probabilities. This affects to the functions getNbestTransFor_s_() and getNbestTransFor_t_() - New option for thot and pbs_thot: -lex. The -lex option incorporates two new score components in the phrase table - The library of phrase models provides faster access to the model probabilities - Better implementation of the OrderedVector class - The tool "filter_ttable" has been incorporated to the distribution. "filter_ttable" is a new tool to filter phrase models in order to apply them in machine translation tasks - The tool "score_phr_ibm" has been incorporated to the distribution. "score_phr_ibm" adds lexical scores for the phrase pairs of a given phrase table. The lexical scores are calculated as the score for the viterbi alignment of IBM 1 model. - "make check" is replaced by "make installcheck" since the tool checks are to be executed after installing the package - Source and target sentences or phrases are no longer noted using the letters 'e' and 'f' respectively, which are replaced by 's' and 't'. The modifications affects the name of variables and functions Functions that have changed their names: a) The scoring functions defined in BasePhraseModel: pf_e_, pe_f_, have changed its names: ps_t_, pt_s_ ... b) Functions to obtain lists of translations The old functions cited above are maintained for backward compatibility Functions that have altered the order in which the source and target sentences are given: a) The functions defined in the "print_func" module b) Some function members declared in the "PhraseModel_From_WBA" and the "PhraseExtractionTable" classes. Three functions which were modified are declared as public functions: PhraseModel_From_WBA::obtainBestAlignment PhraseExtractionTable::extractConsistentPhrases PhraseExtractionTable::obtainPossibleSegmentations - Some data types and functions names have been changed in order to simplify and clarify the code: InvNbestTransTable -> PhraseNbestTransTable TransTableNodeData and InvTransTableNodeData -> PhraseTableNodeData BasePhraseModel::getNbestInvTransFor_t_() -> BasePhraseModel::getNbestTransFor_t_() PhraseDict::getInvTranslationsFor_t_() -> PhraseDict::getTranslationsFor_t_() - Solved a bug in ps_t_ function defined in "PhraseModel.cc" - Solved a bug in the alig_op tool. The bug could appear when using the -f option, specifically when the prefix for the output file contained directory components - The robustness of the alig_op tool has been increased due to a better management of the temporary files that are used internally -------------------------------------------------------------------- $Date: 2008-01-16 19:24:41 $

  posted 662 days ago

• File released: /thot/2.1.0Beta/Thot-2.1.0Beta.tar.gz

  posted 662 days ago

• Released a better documented version of the toolkit

  A better documented version of the toolkit has been released. The new version incorporates a tutorial of the toolkit in pdf format.

  posted by danulino 873 days ago