packages feed

GenI 0.24.2 → 0.24.3

raw patch · 11 files changed

+329/−254 lines, 11 files

Files

GenI.cabal view
@@ -1,5 +1,5 @@ Name:           GenI-Version:        0.24.2+Version:        0.24.3 License:        GPL License-file:   LICENSE Author:         Carlos Areces and Eric Kow@@ -10,7 +10,7 @@ Maintainer:     geni-users@loria.fr Build-Type:     Simple Cabal-Version: >=1.14-data-files: AUTHORS, README.markdown, NEWS, GenI.cabal,+data-files: AUTHORS, README.markdown, changelog, GenI.cabal,             examples/artificial/lexicon,             examples/artificial/trees,             examples/artificial/suite,
− NEWS 
@@ -1,220 +0,0 @@-GenI 0.24, 18 Sep 2013------------------------Note that the changes between this release and 0.20 are considered to be under-the BSD3 license.  GenI overall remains dual-licensed under the GPL and any-commercial-use terms you may wish to negotiate with INRIA. Having secured the-rights to use GenI 0.20, you may freely use GenI 0.24 in a similar fashion.--CHANGES TO WATCH OUT FOR--* Server/GUI changes: Semantic input must be full input.-  - Before: `foo(x) bar(y z)`-  - After:  `semantics:[foo(x) bar(y z)]`-* Customisable semantics : now you can accept any arbitrary-  input for which you can define a mapping to a GenI-style test case-* API change: the custom semantics argument is mandatory-* More extensive documentation--GUI--* Visual distinction between explicit null adjunction constraints-  (marked NA) and constraints inserted by GenI (marked INA)-* Bugfixes (now retains show-feature state across debugger steps)-* More informative top/bot unification failure diagnostics in the-  trash--GENERAL--* New "guided realisation" optimisation (focus on one polarity-  path at a time)-* Fixed errors where top/bottom unification wasn't being done-* Builds with haskell-platform 2013.2.0.0--GenI 0.22, 14 Apr 2012------------------------Note that the changes between this release and 0.20 are considered to be under-the BSD3 license.  GenI overall remains dual-licensed under the GPL and any-commercial-use terms you may wish to negotiate with INRIA. Having secured the-rights to use GenI 0.20, you may freely use GenI 0.22 in a similar fashion.--CHANGES TO WATCH OUT FOR--* Separation of geni and geni-gui into two packages;-  simpler installation-* Now using the top feature for lemanchors instead of bot-* Empty disjunctions forbidden in unification variables-* JSON input/output to morph processors have changed-* Lexical selection unifies semantics of tree schema + lemma with-  input semantics.--GUI--* New results tab with reminder of inputs-* Add gui element to display unanchored lexical selection.-* Jump to new tab in debugger panel when created.-* Update gui to use Pretty instead of Show-* Rename $algo-session tab to "tree assembly"-* Make warnings resizable.-* Fix display of path equations with 'interface' in them.-* Fix #50 - load button has no effect.-* Fix #49 - correct order that leaf nodes are returned.-* Show GenI version in About box.-* Add detect polarities and root feature editor to main gui.--DOCUMENTATION--* Literate GenI and genimanual replaced with-  http://projects.haskell.org/GenI/manual--GENERAL--* Builds with GHC 7-* Core behaviour-     * Early null-adjunction and semfilter obligatory-     * Handling of zero-literal semantic items restored-     * Unconstrained polarities in automaton construction.-       This makes it possible to do polarity detection with only a-       weakly specified root feature like [cat:_], and without needing-       the relevant feature to be implemented everywhere.--* Flags:-     * --trees (used to be --macros; latter is deprecated)-     * --batchdir now optional (defaults to a new temp dir)-     * --maxsteps (stop after N steps)-     * --extrapol REMOVED (use root feature instead)-* Logging facility (not used much in GenI yet)-  Can be configured in ~/.geni/config.yaml, for example-  logging:-   --     name     : NLP.GenI.Console-     level    : WARN-     format   : simple "$msg"-   --     name     : NLP.Geni.LexicalSelection-     level    : DEBUG-     format   : simple "$utcTime $loggername $msg"-     handler  : stderr-* Less repetition in warnings (hopefully easier to read)-* More hiding of constructors for safety-* New enrichment option for setting co-anchors (foo.lex=hello)-* Lexical selection performance enhancements-* Variables that only occur once are converted to anonymous,-  should make unification a bit more efficient-* Fix #39: reject malformed root feature input.-* New feature:-  * Variables with constraints on their possible values (?X/foo|bar)-  * Fancy disjunctions: tree schemata now allow disjunctions of-    unification variables (but not recursive). These must flatten down-    to plain old atomic disjunction once converted to elementary trees--* Time metric in statistics (Laura Perez)-* Do top/bottom unification on na-constrained nodes during initialisation.-* Polarity detection relaxed to work with constrained variables too.-  (?X/foo|bar gives the result as foo|bar)-* Bugfix in interpretation of root feature (it should be sorted).-* Flags:-  * --rootfeat always assumed, but defaults to '[cat:_]'-  * '-r' a short command line option for '--rootfeat'.--LIBRARY--* Somewhat safer and easier to understand API-  (still no stability yet)-* Replace String with Text in GeniVal-* Replace Show abuse with custom Pretty class-* Slightly less stateful (no target semantics in state)-* Much rearranging and renaming (sorry!)-* New: ability to provide custom lexical selection action-* New: ability to provide custom morph postprocessing action-* de-haskell98 (Gwern Branwen)-* No more null builder.--TOOLS (darcs get --lazy http://code.haskell.org/GenI)--* new tool, genireport: summarise batch generation results-* geniserver now talks over HTTP using a JSON format-* Use Sylvain Salvati's xmg2geni.xsl instead of geniconvert.-    * xmg2geni: fix treatment of missing lexemes.-    * xmg2geni: treat type=coanchor as other nodes in GenI.-* geniwrapper: illustrates C wrapper of GenI API--GenI 0.20, 29 Sep 2009------------------------- NON BACKWARD-COMPATIBLE CHANGES--  * GHC 6.8 or higher now required (GHC 6.10 + Haskell Platform preferred)-  * Input files now *assumed* to be UTF-8 encoded-    (output still ISO8859-1; hopefully locale-based in GHC 6.12)-  * Interface between GenI and morphological realiser now uses a JSON format.-  * Rootfeat optimisation now mandatory (flag no longer recognised)--- NEW FUNCTIONALITY--  * new flag --version now works and reports version from Cabal file-  * new flag --dump dumps derivation output to stdout in JSON format-  * new flag --ranking foo ranks outputs according to OT-style constraints expressed in foo-  * --batchdir now works with --testcase-  * --batchdir now dumps derivations and statistics in JSON format-  * --batchdir now works with instructions files-  * --metrics always expands 'default' to the default metrics-    (in addition to any other statistics you request)-  * new optimisation: early null-adjunction detection--- QUALITY ASSURANCE--  * Reference manual in docs/genimanual.pdf (make docs)-  * cabal install -fgui and cabal install produce the same underlying library-    only the executable is different-  * Easier GUI installation for MacOS X (application bundles now built-    automatically; no manual intervention required)-  * Adopted odd/even convention to distinguish devel builds from stable ones-    (eg. GenI 0.21.x is devel whereas GenI 0.20.x is stable)--GenI 0.17.4, 6 Apr 2009-------------------------* GHC 6.6 support--* GUI disabled by default--GenI 0.17.3, 3 Apr 2009-------------------------* Simplified build method-  * one single cabal package instead of two-  * cabal configure -f-gui to disable GUI-  * makefile stripped out--* Lexical selection on empty semantics now allowed-  * This is so that the zero-literal semantics mechanism can work again--* Better help text--* Baked-in unit testing (geni --unit-test)--GenI 0.10, 11 Dec 2006------------------------* Performance much improved after Haskell profiling-* Pre-compiled grammars option, geniserver (buggy)--GenI 0.9, 8 Dec 2006----------------------* Switched to new syntax: variables now must be prefixed with '?'-* Switched from Alex/Happy to Parsec-* Added API documentation via haddock--GenI 0.8, 16 Sep 2005-----------------------* Support for latin-1 characters using Unicode wxhaskell-* Added a configure script and an install target in Makefile-* Changed from use of configuration files to use of command line switches-* Full configuration GUI-* Implementation of atomic disjunction-* --ignore-semantics mode (Jackie Lai)--GenI 0.7, 8 June 2005-----------------------* Upgraded to ghc 6.4, wxwidgets 2.6, wxhaskell 0.9.4--GenI 0.6, 3 May 2005-----------------------* User interface fixes (Window size bugs)-* Interface with Yannick Parmientier's selection module
README.markdown view
@@ -22,13 +22,13 @@ 2. Install the basic version first. If you just want the    last stable release of GenI, -           cabal install GenI geniserver+           cabal install GenI geniserver geni-util         However, if you want to install the unstable version    from source, change to the directory this README is in    and -           cabal install . ./geniserver+           cabal install . ./geniserver ./geni-util  ## Graphical interface
+ changelog view
@@ -0,0 +1,236 @@+GenI 0.24.3 15 Nov 2013+-----------------------+BUGFIXES:++* Lexical selection: trees that fail enrichment due to unification failure are+  now discarded++  Previously, there was some confusion over whether the expected behaviour was+  for this to be an error or warning. The behaviour after this fix is that+  missing nodes trigger a warning, an unification failures trigger an error+  (discarded tree)++* Realisation: When anonymising singleton variables, GenI preserves their+  constraints. This should fix some subtle overgeneration problems that may+  arise in automatically generated grammars++GenI 0.24, 18 Sep 2013+----------------------+Note that the changes between this release and 0.20 are considered to be under+the BSD3 license.  GenI overall remains dual-licensed under the GPL and any+commercial-use terms you may wish to negotiate with INRIA. Having secured the+rights to use GenI 0.20, you may freely use GenI 0.24 in a similar fashion.++CHANGES TO WATCH OUT FOR++* Server/GUI changes: Semantic input must be full input.+  - Before: `foo(x) bar(y z)`+  - After:  `semantics:[foo(x) bar(y z)]`+* Customisable semantics : now you can accept any arbitrary+  input for which you can define a mapping to a GenI-style test case+* API change: the custom semantics argument is mandatory+* More extensive documentation++GUI++* Visual distinction between explicit null adjunction constraints+  (marked NA) and constraints inserted by GenI (marked INA)+* Bugfixes (now retains show-feature state across debugger steps)+* More informative top/bot unification failure diagnostics in the+  trash++GENERAL++* New "guided realisation" optimisation (focus on one polarity+  path at a time)+* Fixed errors where top/bottom unification wasn't being done+* Builds with haskell-platform 2013.2.0.0++GenI 0.22, 14 Apr 2012+----------------------+Note that the changes between this release and 0.20 are considered to be under+the BSD3 license.  GenI overall remains dual-licensed under the GPL and any+commercial-use terms you may wish to negotiate with INRIA. Having secured the+rights to use GenI 0.20, you may freely use GenI 0.22 in a similar fashion.++CHANGES TO WATCH OUT FOR++* Separation of geni and geni-gui into two packages;+  simpler installation+* Now using the top feature for lemanchors instead of bot+* Empty disjunctions forbidden in unification variables+* JSON input/output to morph processors have changed+* Lexical selection unifies semantics of tree schema + lemma with+  input semantics.++GUI++* New results tab with reminder of inputs+* Add gui element to display unanchored lexical selection.+* Jump to new tab in debugger panel when created.+* Update gui to use Pretty instead of Show+* Rename $algo-session tab to "tree assembly"+* Make warnings resizable.+* Fix display of path equations with 'interface' in them.+* Fix #50 - load button has no effect.+* Fix #49 - correct order that leaf nodes are returned.+* Show GenI version in About box.+* Add detect polarities and root feature editor to main gui.++DOCUMENTATION++* Literate GenI and genimanual replaced with+  http://projects.haskell.org/GenI/manual++GENERAL++* Builds with GHC 7+* Core behaviour+     * Early null-adjunction and semfilter obligatory+     * Handling of zero-literal semantic items restored+     * Unconstrained polarities in automaton construction.+       This makes it possible to do polarity detection with only a+       weakly specified root feature like [cat:_], and without needing+       the relevant feature to be implemented everywhere.++* Flags:+     * --trees (used to be --macros; latter is deprecated)+     * --batchdir now optional (defaults to a new temp dir)+     * --maxsteps (stop after N steps)+     * --extrapol REMOVED (use root feature instead)+* Logging facility (not used much in GenI yet)+  Can be configured in ~/.geni/config.yaml, for example+  logging:+   -+     name     : NLP.GenI.Console+     level    : WARN+     format   : simple "$msg"+   -+     name     : NLP.Geni.LexicalSelection+     level    : DEBUG+     format   : simple "$utcTime $loggername $msg"+     handler  : stderr+* Less repetition in warnings (hopefully easier to read)+* More hiding of constructors for safety+* New enrichment option for setting co-anchors (foo.lex=hello)+* Lexical selection performance enhancements+* Variables that only occur once are converted to anonymous,+  should make unification a bit more efficient+* Fix #39: reject malformed root feature input.+* New feature:+  * Variables with constraints on their possible values (?X/foo|bar)+  * Fancy disjunctions: tree schemata now allow disjunctions of+    unification variables (but not recursive). These must flatten down+    to plain old atomic disjunction once converted to elementary trees++* Time metric in statistics (Laura Perez)+* Do top/bottom unification on na-constrained nodes during initialisation.+* Polarity detection relaxed to work with constrained variables too.+  (?X/foo|bar gives the result as foo|bar)+* Bugfix in interpretation of root feature (it should be sorted).+* Flags:+  * --rootfeat always assumed, but defaults to '[cat:_]'+  * '-r' a short command line option for '--rootfeat'.++LIBRARY++* Somewhat safer and easier to understand API+  (still no stability yet)+* Replace String with Text in GeniVal+* Replace Show abuse with custom Pretty class+* Slightly less stateful (no target semantics in state)+* Much rearranging and renaming (sorry!)+* New: ability to provide custom lexical selection action+* New: ability to provide custom morph postprocessing action+* de-haskell98 (Gwern Branwen)+* No more null builder.++TOOLS (darcs get --lazy http://code.haskell.org/GenI)++* new tool, genireport: summarise batch generation results+* geniserver now talks over HTTP using a JSON format+* Use Sylvain Salvati's xmg2geni.xsl instead of geniconvert.+    * xmg2geni: fix treatment of missing lexemes.+    * xmg2geni: treat type=coanchor as other nodes in GenI.+* geniwrapper: illustrates C wrapper of GenI API++GenI 0.20, 29 Sep 2009+----------------------+- NON BACKWARD-COMPATIBLE CHANGES++  * GHC 6.8 or higher now required (GHC 6.10 + Haskell Platform preferred)+  * Input files now *assumed* to be UTF-8 encoded+    (output still ISO8859-1; hopefully locale-based in GHC 6.12)+  * Interface between GenI and morphological realiser now uses a JSON format.+  * Rootfeat optimisation now mandatory (flag no longer recognised)++- NEW FUNCTIONALITY++  * new flag --version now works and reports version from Cabal file+  * new flag --dump dumps derivation output to stdout in JSON format+  * new flag --ranking foo ranks outputs according to OT-style constraints expressed in foo+  * --batchdir now works with --testcase+  * --batchdir now dumps derivations and statistics in JSON format+  * --batchdir now works with instructions files+  * --metrics always expands 'default' to the default metrics+    (in addition to any other statistics you request)+  * new optimisation: early null-adjunction detection++- QUALITY ASSURANCE++  * Reference manual in docs/genimanual.pdf (make docs)+  * cabal install -fgui and cabal install produce the same underlying library+    only the executable is different+  * Easier GUI installation for MacOS X (application bundles now built+    automatically; no manual intervention required)+  * Adopted odd/even convention to distinguish devel builds from stable ones+    (eg. GenI 0.21.x is devel whereas GenI 0.20.x is stable)++GenI 0.17.4, 6 Apr 2009+-----------------------+* GHC 6.6 support++* GUI disabled by default++GenI 0.17.3, 3 Apr 2009+-----------------------+* Simplified build method+  * one single cabal package instead of two+  * cabal configure -f-gui to disable GUI+  * makefile stripped out++* Lexical selection on empty semantics now allowed+  * This is so that the zero-literal semantics mechanism can work again++* Better help text++* Baked-in unit testing (geni --unit-test)++GenI 0.10, 11 Dec 2006+----------------------+* Performance much improved after Haskell profiling+* Pre-compiled grammars option, geniserver (buggy)++GenI 0.9, 8 Dec 2006+--------------------+* Switched to new syntax: variables now must be prefixed with '?'+* Switched from Alex/Happy to Parsec+* Added API documentation via haddock++GenI 0.8, 16 Sep 2005+---------------------+* Support for latin-1 characters using Unicode wxhaskell+* Added a configure script and an install target in Makefile+* Changed from use of configuration files to use of command line switches+* Full configuration GUI+* Implementation of atomic disjunction+* --ignore-semantics mode (Jackie Lai)++GenI 0.7, 8 June 2005+---------------------+* Upgraded to ghc 6.4, wxwidgets 2.6, wxhaskell 0.9.4++GenI 0.6, 3 May 2005+---------------------+* User interface fixes (Window size bugs)+* Interface with Yannick Parmientier's selection module
examples/artificial/lexicon view
@@ -67,3 +67,22 @@  trace-good tr-good () semantics:[trace()] trace-bad  tr-bad  () semantics:[trace()]++enrichment-good enrichment ()+equations:[anchor.bot.foo:+]+semantics:[enrichment-good()]++enrichment-bad  enrichment ()+equations:[anchor.bot.foo:-]+semantics:[enrichment-bad()]++enrichment-restraint-good enrichment-restraint ()+equations:[anchor.bot.foo:+]+semantics:[enrichment-restraint-good()]++enrichment-missing-node-good enrichment ()+equations:[foo.bot.foo:+]+semantics:[enrichment-missing-node-good()]++single-user-var single-use-var ()+semantics:[single-use-var()]
examples/artificial/suite view
@@ -34,3 +34,12 @@ variablecat semantics:[variablecat()]  trace-good semantics:[trace()[T1]]++enrichment-good+semantics:[enrichment-good()]++enrichment-restraint-good+semantics:[enrichment-restraint-good()]++enrichment-missing-node-good+semantics:[enrichment-missing-node-good()]
examples/artificial/suite-bad view
@@ -14,3 +14,8 @@  interface-unification semantics:[interface()] trace-bad  semantics:[trace()[T2]]++enrichment-bad semantics:[enrichment-bad()]++single-use-var+semantics:[single-use-var()]
examples/artificial/trees view
@@ -158,3 +158,16 @@ tr-bad() initial X anchor [cat:x]![cat:y] trace:[T2]++enrichment() initial+X anchor [cat:x]![foo:+]++enrichment-restraint:e1() initial % meant to fail+X anchor [cat:x]![foo:-]++enrichment-restraint:e2() initial % meant to be left behind+X anchor [cat:x]![]++% should not be anonymised away+single-use-var() initial+X anchor [cat:x]![cat:?X/y]
src/NLP/GenI/GeniVal/Internal.hs view
@@ -386,12 +386,18 @@ -- | Anonymise any variable that occurs only once in the object anonymiseSingletons :: (Collectable a, DescendGeniVal a) => a -> a anonymiseSingletons x =-   replace subst x+   descendGeniVal anonymiseIfLonely x  where-   subst = Map.map (const mkGAnon) . Map.filter (== 1)-           -- merge counts for same var, different constraints-         . Map.fromListWith (+) . map (first fst) . Map.toList-         $ collect x Map.empty+   anonymiseIfLonely v =+       case gLabel v of+           Just l | l `elem` lonely -> anonymise v+           _                        -> v+   anonymise v = v { gLabel = Nothing }+   lonely = Map.keys+          . Map.filter (== 1)+          -- merge counts for same var, different constraints+          . Map.fromListWith (+) . map (first fst) . Map.toList+          $ collect x Map.empty  -- | 'finaliseVarsById' appends a unique suffix to all variables in -- an object.  This avoids us having to alpha convert all the time
src/NLP/GenI/LexicalSelection.hs view
@@ -307,7 +307,7 @@ -- ** Enrichment -- ---------------------------------------------------------------------- --- | See <http://projects.haskell.org/manual/lexical-selection>+-- | See <http://kowey.github.io/GenI/manual/lexical-selection.html> --   on enrichement enrich :: LexEntry -> SchemaTree -> LexCombine SchemaTree enrich l t = do@@ -327,39 +327,46 @@  -- *** 'enrich' helpers +data EnrichmentResult = EnrSuccess SchemaTree Subst+                      | EnrNotFound+                      | EnrFailed+ -- | Helper for 'enrich' (enrich by single path equation) enrichBy :: SchemaTree          -> PathEqPair          -> LexCombine SchemaTree enrichBy t eq@(eqLhs, _) =     case maybeEnrichBy t eq of-        Nothing     -> lexTell enrichErr >> return t-        Just (t2,_) -> return t2+        EnrNotFound     -> return t -- warning generated separately :-(+        EnrFailed       -> lexTell enrichErr   >> mzero+        EnrSuccess t2 _ -> return t2   where-    enrichErr = SchemaError [pidname t] (EnrichError (PeqJust eqLhs))+    enrichErr   = SchemaError [pidname t] $+        EnrichError (PeqJust eqLhs)  -- | Helper for 'enrichBy' maybeEnrichBy :: SchemaTree               -> PathEqPair-              -> Maybe (SchemaTree, Subst)-maybeEnrichBy t (eqLhs, eqVal) = do-    node <- seekCoanchor eqLhs t-    case eqLhs of-        PeqFeat _ eqTop eqAtt -> do-            let (get, set) = case eqTop of-                               Top     -> (gup,   \n x -> n { gup = x })-                               Bottom  -> (gdown, \n x -> n { gdown = x})-            (fs, sub) <- hush $ enrichFeat (AvPair eqAtt eqVal) (get node)-            let t2 = fixNode (set node fs) (replace sub t)-            return (t2, sub)-        PeqLex _ -> do-             vs <- gConstraints eqVal-             let node2 = node { glexeme = FL.fromFL vs }-                 t2    = fixNode node2 t-             return (t2, Map.empty)+              -> EnrichmentResult+maybeEnrichBy t (eqLhs, eqVal) =+    maybe EnrNotFound enrichNode $ seekCoanchor eqLhs t   where     fixNode n mt =         mt { tree = repNodeByNode (matchNodeName eqLhs) n (tree mt) }+    enrichNode node = fromMaybe EnrFailed $+        case eqLhs of+            PeqFeat _ eqTop eqAtt -> do+                let (get, set) = case eqTop of+                                   Top     -> (gup,   \n x -> n { gup = x })+                                   Bottom  -> (gdown, \n x -> n { gdown = x})+                (fs, sub) <- hush $ enrichFeat (AvPair eqAtt eqVal) (get node)+                let t2 = fixNode (set node fs) (replace sub t)+                return $ EnrSuccess t2 sub+            PeqLex _ -> do+                 vs <- gConstraints eqVal+                 let node2 = node { glexeme = FL.fromFL vs }+                     t2    = fixNode node2 t+                 return $ EnrSuccess t2 Map.empty  -- | @enrichFeat av fs@ attempts to unify @av@ with @fs@ --
src/NLP/GenI/LexicalSelection/Types.hs view
@@ -99,7 +99,7 @@      | SchemaError [Text] LexCombineError2  deriving Eq -data LexCombineError2 = EnrichError PathEqLhs+data LexCombineError2 = EnrichError             PathEqLhs                       | StringError Text  deriving (Eq, Ord) @@ -111,10 +111,10 @@  leq _ _ = False  instance Poset LexCombineError2 where- leq (EnrichError e1) (EnrichError e2)  = leq e1 e2- leq (EnrichError _ ) (StringError _ )  = True- leq (StringError s1) (StringError s2)  = leq s1 s2- leq _ _ = False+    leq (EnrichError e1) (EnrichError e2)  = leq e1 e2+    leq (EnrichError _ ) (StringError _ )  = True+    leq (StringError s1) (StringError s2)  = leq s1 s2+    leq _ _ = False  instance Poset PathEqLhs where  leq l1 l2 = leq (showPathEqLhs l1) (showPathEqLhs l2)