---------------------------------------------------------------------- -- | -- Module : SRG -- Maintainer : BB -- Stability : (stable) -- Portability : (portable) -- -- > CVS $Date: 2005/11/01 20:09:04 $ -- > CVS $Author: bringert $ -- > CVS $Revision: 1.20 $ -- -- Representation of, conversion to, and utilities for -- printing of a general Speech Recognition Grammar. -- -- FIXME: remove \/ warn \/ fail if there are int \/ string literal -- categories in the grammar ----------------------------------------------------------------------------- module GF.Speech.SRG (SRG(..), SRGRule(..), SRGAlt(..), SRGCat, SRGNT, CFTerm, makeSimpleSRG , lookupFM_, prtS , cfgCatToGFCat, srgTopCats , EBnfSRGAlt(..), EBnfSRGItem , ebnfSRGAlts ) where import GF.Data.Operations import GF.Data.Utilities import GF.Infra.Ident import GF.Formalism.CFG import GF.Formalism.Utilities (Symbol(..), NameProfile(..) , Profile(..), SyntaxForest , filterCats, mapSymbol, symbol) import GF.Conversion.Types import GF.Infra.Print import GF.Speech.TransformCFG import GF.Speech.Relation import GF.Speech.FiniteState import GF.Speech.RegExp import GF.Infra.Option import GF.Probabilistic.Probabilistic (Probs) import GF.Compile.ShellState (StateGrammar, stateProbs, stateOptions, cncId) import Data.List import Data.Maybe (fromMaybe, maybeToList) import Data.Map (Map) import qualified Data.Map as Map import Data.Set (Set) import qualified Data.Set as Set data SRG = SRG { grammarName :: String -- ^ grammar name , startCat :: String -- ^ start category name , origStartCat :: String -- ^ original start category name , grammarLanguage :: Maybe String -- ^ The language for which the grammar -- is intended, e.g. en-UK , rules :: [SRGRule] } deriving (Eq,Show) data SRGRule = SRGRule SRGCat String [SRGAlt] -- ^ SRG category name, original category name -- and productions deriving (Eq,Show) -- | maybe a probability, a rule name and a list of symbols data SRGAlt = SRGAlt (Maybe Double) CFTerm [Symbol SRGNT Token] deriving (Eq,Show) type SRGCat = String -- | An SRG non-terminal. Category name and its number in the profile. type SRGNT = (SRGCat, Int) -- | SRG category name and original name type CatName = (SRGCat,String) type CatNames = Map String String -- | Create a non-left-recursive SRG. -- FIXME: the probabilities in the returned -- grammar may be meaningless. makeSimpleSRG :: Options -- ^ Grammar options -> StateGrammar -> SRG makeSimpleSRG opt s = SRG { grammarName = name, startCat = lookupFM_ names origStart, origStartCat = origStart, grammarLanguage = l, rules = rs } where opts = addOptions opt (stateOptions s) name = prIdent (cncId s) origStart = getStartCatCF opts s probs = stateProbs s l = fmap (replace '_' '-') $ getOptVal opts speechLanguage (cats,cfgRules) = unzip $ preprocess $ cfgToCFRules s preprocess = removeLeftRecursion origStart . removeEmptyCats . topDownFilter origStart . removeIdenticalRules . removeCycles names = mkCatNames name cats rs = map (cfgRulesToSRGRule names probs) cfgRules -- FIXME: merge alternatives with same rhs and profile but different probabilities cfgRulesToSRGRule :: Map String String -> Probs -> [CFRule_] -> SRGRule cfgRulesToSRGRule names probs rs@(r:_) = SRGRule cat origCat rhs where origCat = lhsCat r cat = lookupFM_ names origCat rhs = nub $ map ruleToAlt rs ruleToAlt r@(CFRule c ss n) = SRGAlt (ruleProb probs r) n (mkSRGSymbols 0 ss) where mkSRGSymbols _ [] = [] mkSRGSymbols i (Cat c:ss) = Cat (renameCat c,i) : mkSRGSymbols (i+1) ss mkSRGSymbols i (Tok t:ss) = Tok t : mkSRGSymbols i ss renameCat = lookupFM_ names ruleProb :: Probs -> CFRule_ -> Maybe Double ruleProb probs r = lookupProb probs (ruleFun r) -- FIXME: move to GF.Probabilistic.Probabilistic? lookupProb :: Probs -> Ident -> Maybe Double lookupProb probs i = lookupTree prIdent i probs mkCatNames :: String -- ^ Category name prefix -> [String] -- ^ Original category names -> Map String String -- ^ Maps original names to SRG names mkCatNames prefix origNames = Map.fromList (zip origNames names) where names = [prefix ++ "_" ++ show x | x <- [0..]] allSRGCats :: SRG -> [String] allSRGCats SRG { rules = rs } = [c | SRGRule c _ _ <- rs] cfgCatToGFCat :: SRGCat -> Maybe String cfgCatToGFCat c -- categories introduced by removeLeftRecursion contain dashes | '-' `elem` c = Nothing -- some categories introduced by -conversion=finite have the form -- "{fun:cat}..." | "{" `isPrefixOf` c = case dropWhile (/=':') $ takeWhile (/='}') $ tail c of ':':c' -> Just c' _ -> error $ "cfgCatToGFCat: Strange category " ++ show c | otherwise = Just $ takeWhile (/='{') c srgTopCats :: SRG -> [(String,[SRGCat])] srgTopCats srg = buildMultiMap [(oc, cat) | SRGRule cat origCat _ <- rules srg, oc <- maybeToList $ cfgCatToGFCat origCat] -- -- * Size-optimized EBNF SRGs -- data EBnfSRGAlt = EBnfSRGAlt (Maybe Double) CFTerm EBnfSRGItem deriving (Eq,Show) type EBnfSRGItem = RE (Symbol SRGNT Token) ebnfSRGAlts :: [SRGAlt] -> [EBnfSRGAlt] ebnfSRGAlts alts = [EBnfSRGAlt p n (ebnfSRGItem sss) | ((n,p),sss) <- buildMultiMap [((n,p),ss) | SRGAlt p n ss <- alts]] ebnfSRGItem :: [[Symbol SRGNT Token]] -> EBnfSRGItem ebnfSRGItem = unionRE . map mergeItems . sortGroupBy (compareBy filterCats) -- | Merges a list of right-hand sides which all have the same -- sequence of non-terminals. mergeItems :: [[Symbol SRGNT Token]] -> EBnfSRGItem mergeItems = minimizeRE . ungroupTokens . minimizeRE . unionRE . map seqRE . map groupTokens groupTokens :: [Symbol SRGNT Token] -> [Symbol SRGNT [Token]] groupTokens [] = [] groupTokens (Tok t:ss) = case groupTokens ss of Tok ts:ss' -> Tok (t:ts):ss' ss' -> Tok [t]:ss' groupTokens (Cat c:ss) = Cat c : groupTokens ss ungroupTokens :: RE (Symbol SRGNT [Token]) -> RE (Symbol SRGNT Token) ungroupTokens = joinRE . mapRE (symbol (RESymbol . Cat) (REConcat . map (RESymbol . Tok))) -- -- * Utilities for building and printing SRGs -- lookupFM_ :: (Ord key, Show key) => Map key elt -> key -> elt lookupFM_ fm k = Map.findWithDefault (error $ "Key not found: " ++ show k) k fm prtS :: Print a => a -> ShowS prtS = showString . prt