diff --git a/langfilter/langfilter.cabal b/langfilter/langfilter.cabal
index 4c18973..3440006 100644
--- a/langfilter/langfilter.cabal
+++ b/langfilter/langfilter.cabal
@@ -11,13 +11,18 @@ executable langfilter
   hs-source-dirs: .
   main-is: langfilter.hs
   ghc-options:
-    -Wall -threaded -rtsopts -with-rtsopts=-N
+    -Wall -Wno-missing-signatures -Wno-missing-pattern-synonym-signatures
+    -Wno-name-shadowing
+    -threaded -rtsopts -with-rtsopts=-N
   default-language: Haskell2010
   default-extensions:
     BlockArguments,
     LambdaCase,
+    OverloadedStrings,
     PatternSynonyms,
     ViewPatterns
   build-depends:
     base ^>= 4.14.0.0,
-    pandoc-types == 1.17.*
+    pandoc-types ^>= 1.22,
+    text,
+    pretty-show ^>= 1.10
diff --git a/langfilter/langfilter.hs b/langfilter/langfilter.hs
index 090c47e..07cf42d 100644
--- a/langfilter/langfilter.hs
+++ b/langfilter/langfilter.hs
@@ -1,27 +1,117 @@
+{-# OPTIONS_GHC -fdefer-typed-holes #-}
 import Text.Pandoc.Definition
 import Text.Pandoc.JSON
 import Text.Pandoc.Builder
+import Text.Pandoc.Walk
+import Data.Char (isUpper, isDigit)
+import Data.Maybe
+import Data.Text (Text)
+import qualified Data.Text as Text
 
 main :: IO ()
-main = toJSONFilter \case
-  Code _ txt
-    | Just _ <- enclosed '/' '/' txt ->
-        Span (cls ["ipa", "ipa-broad"]) $ text' txt
-    | Just _ <- enclosed '[' ']' txt ->
-        Span (cls ["ipa", "ipa-narrow"]) $ text' txt
-    | Just txt' <- enclosed '{' '}' txt ->
-        Span (cls ["lang"]) $ text' txt'
+main = toJSONFilter filter where
+  filter :: Pandoc -> Pandoc
+  filter =
+    walk spans .
+    walk (concatMap makeFigures) .
+    walk (concatMap glosses)
+
+spans :: Inline -> Inline
+spans = \case
+  Code attrs txt
+    | Just ('\\', txt') <- Text.uncons txt      -> Code attrs txt'
+    | Just _            <- enclosed '/' '/' txt -> ipaB txt
+    | Just _            <- enclosed '[' ']' txt -> ipaN txt
+    | Just txt'         <- enclosed '{' '}' txt -> lang txt'
+    | Just txt'         <- enclosed '!' '!' txt -> abbr txt'
   i -> i
 
-cls :: [String] -> Attr
-cls cs = ("", cs, [])
+ipaB, ipaN, lang, abbr :: Text -> Inline
+ipaB = Span (cls ["ipa", "ipa-broad"])  . text'
+ipaN = Span (cls ["ipa", "ipa-narrow"]) . text'
+lang = Span (cls ["lang"])              . text'
+abbr = Span (cls ["abbr"])              . text' . endash
 
-text' :: String -> [Inline]
+text' :: Text -> [Inline]
 text' = toList . text
 
+cls :: [Text] -> Attr
+cls cs = ("", cs, [])
 
-enclosed :: Char -> Char -> String -> Maybe String
+glosses :: Block -> [Block]
+glosses = \case
+  Div (_, cs, _) blocks | "glosses" `elem` cs ->
+    [RawBlock (Format "html") "<figure class=glosses>"] ++
+    mapMaybe glossTable blocks ++
+    [RawBlock (Format "html") "</figure>"]
+  b -> [b]
+
+glossTable :: Block -> Maybe Block
+glossTable = \case
+  Gloss  l   s g t -> Just $ make l Nothing  s g t
+  PGloss l p s g t -> Just $ make l (Just p) s g t
+  HorizontalRule   -> Nothing
+  b                -> error $ "found " ++ show b ++ " in gloss section"
+ where
+  make l p s g t =
+    let n = length $ splitInlines s
+        colspecs = replicate n (AlignDefault, ColWidthDefault)
+        l' = cell1 n l; p' = cell1 n <$> p
+        ss = cells s; gs = cells g; t' = cell1 n t
+    in
+    Table ("", ["gloss"], []) (Caption Nothing []) colspecs
+      (TableHead mempty [])
+      [TableBody mempty (RowHeadColumns 0) [] $ concat
+        [[row "gloss-lang" [l']],
+         maybe [] (pure . row "gloss-pron" . pure) p',
+         [row "gloss-split" ss],
+         [row "gloss-gloss" gs],
+         [row "gloss-trans" [t']]]]
+      (TableFoot mempty [])
+  cell is = Cell mempty AlignDefault (RowSpan 1) (ColSpan 1) [Plain is]
+  cell1 n is = Cell mempty AlignDefault (RowSpan 1) (ColSpan n) [Plain is]
+  cells = map (cell . concatMap abbrs) . splitInlines
+  row c = Row ("", [c], [])
+
+endash :: Text -> Text
+endash = Text.map \case '-' -> '–'; '_' -> ' '; c -> c
+
+abbrs :: Inline -> [Inline]
+abbrs (Str txt) = go $ endash txt where
+  go "" = []
+  go txt
+    | (l, r) <- Text.span isAbbr txt,
+      not $ Text.null l
+    = abbr' l : go r
+    | (l, r) <- Text.break isAbbr txt
+    = Str l : go r
+  abbr' txt = if Text.length txt == 1 then Str txt else abbr txt
+  isAbbr c = isUpper c || isDigit c || c `elem` (",.;\\[]" :: String)
+abbrs i = [i]
+
+splitInlines :: [Inline] -> [Inlines]
+splitInlines is = filter (not . null) $ go is where
+  go [] = []
+  go is =
+    let (is1, is') = break (== Space) is in
+    fromList is1 : splitInlines (dropWhile (== Space) is')
+
+pattern Gloss l g w t = BulletList [[Plain l], [Plain g], [Plain w], [Plain t]]
+pattern PGloss l p g w t =
+  BulletList [[Plain l], [Plain p], [Plain g], [Plain w], [Plain t]]
+
+
+makeFigures :: Block -> [Block]
+makeFigures (Div ("", ["figure"], []) blks) =
+  [html "<figure>"] ++ blks ++ [html "</figure>"]
+  where html = RawBlock (Format "html")
+makeFigures b = [b]
+
+
+enclosed :: Char -> Char -> Text -> Maybe Text
 enclosed o c txt
-  | length txt >= 2, head txt == o, last txt == c
-  = Just $ init $ tail txt
+  | Text.length txt >= 2,
+    Text.head txt == o,
+    Text.last txt == c
+  = Just $ Text.init $ Text.tail txt
 enclosed _ _ _ = Nothing