SRG generation: merge categories with identical set of productions. The LC_LR algorithm produces lots of those, especially when there is little inflection.

2026-05-11 04:02:52 -06:00 · 2007-03-24 23:29:38 +00:00
parent 436ddd5ebf
commit ed1e7f4485
2 changed files with 14 additions and 1 deletions
--- a/src/GF/Speech/SRG.hs
+++ b/src/GF/Speech/SRG.hs
@@ -95,7 +95,8 @@ makeSimpleSRG opt s =
    probs = stateProbs s
    l = fmap (replace '_' '-') $ getOptVal opts speechLanguage
    (cats,cfgRules) = unzip $ preprocess $ cfgToCFRules s
-    preprocess = removeLeftRecursion origStart 
+    preprocess = mergeIdentical
+                 . removeLeftRecursion origStart 
                 . fix (topDownFilter origStart . bottomUpFilter)
                 . removeCycles
    names = mkCatNames name cats