batchalign 0.8.2.post2__tar.gz → 0.8.2.post4__tar.gz

{batchalign-0.8.2.post2/batchalign.egg-info → batchalign-0.8.2.post4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post2
+Version: 0.8.2.post4
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.8.2.post2 → batchalign-0.8.2.post4}/batchalign/cli/cli.py

@@ -415,10 +415,11 @@ def benchmark(ctx, in_dir, out_dir, lang, num_speakers, whisper, whisper_oai, **
 def compare(ctx, in_dir, out_dir, lang, **kwargs):
     """Compare transcripts against gold-standard references.
 
-    For each FILE.cha in IN_DIR, expects a companion FILE.gold.cha in the
-    same directory.  Runs morphosyntax analysis on the main transcript, then
-    produces a word-level diff stored as %%xsrep / %%xsmor tiers and writes
-    error metrics to a .compare.csv file in OUT_DIR.
+    For each FILE.cha in IN_DIR, looks for a companion FILE.gold.cha in the
+    same directory.  If not found, falls back to template.gold.cha as the
+    default gold for all files.  Runs morphosyntax analysis on the main
+    transcript, then produces a word-level diff stored as %%xsrep / %%xsmor
+    tiers and writes error metrics to a .compare.csv file in OUT_DIR.
     """
 
     _dispatch("compare", lang, 1, ["cha"], ctx,

{batchalign-0.8.2.post2 → batchalign-0.8.2.post4}/batchalign/cli/dispatch.py

@@ -204,19 +204,23 @@ def _run_pipeline_for_file(command, pipeline, file, output, loader_info, writer_
         if file.endswith(".gold.cha"):
             return
 
-        # Find companion gold file
+        # Find companion gold file: FILE.gold.cha overrides, then
+        # template.gold.cha serves as the default for the whole directory
         p = P(file)
         gold_path = p.parent / (p.stem + ".gold.cha")
+        if not gold_path.exists():
+            gold_path = p.parent / "template.gold.cha"
         if not gold_path.exists():
             raise FileNotFoundError(
                 f"No gold .cha file found for comparison. "
-                f"main: {p.name}, expected: {gold_path.name}, looked in: {str(gold_path)}"
+                f"main: {p.name}, expected: {p.stem}.gold.cha or template.gold.cha, "
+                f"looked in: {str(p.parent)}"
             )
 
         main_doc = CHATFile(path=str(p)).doc
         gold_doc = CHATFile(path=str(gold_path), special_mor_=True).doc
 
-        # Pipeline: morphosyntax(main) -> compare -> compare_analysis
+        # Pipeline: compare (runs morphosyntax internally) -> compare_analysis
         result = pipeline(main_doc, callback=progress_callback, gold=gold_doc)
 
         # Write annotated CHAT
@@ -384,7 +388,7 @@ Cmd2Task = {
     "coref": "coref",
     "translate": "translate",
     "opensmile": "opensmile",
-    "compare": "morphosyntax,compare,compare_analysis",
+    "compare": "compare,compare_analysis",
 }
 
 # this is the main runner used by all functions

{batchalign-0.8.2.post2 → batchalign-0.8.2.post4}/batchalign/formats/chat/generator.py

@@ -131,6 +131,9 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
             prefix = "+" if tok.status == "extra_main" else ("-" if tok.status == "extra_gold" else "")
             xsrep_parts.append(f"{prefix}{tok.text}")
             xsmor_parts.append(f"{prefix}{tok.pos or '?'}")
+        # Strip PUNCT POS from final element, keep the delimiter char (same as %mor)
+        if xsmor_parts and xsrep_parts and xsmor_parts[-1] == "PUNCT":
+            xsmor_parts[-1] = xsrep_parts[-1]
         result.append(f"%xsrep:\t" + " ".join(xsrep_parts))
         result.append(f"%xsmor:\t" + " ".join(xsmor_parts))
 

{batchalign-0.8.2.post2 → batchalign-0.8.2.post4}/batchalign/pipelines/analysis/compare.py

@@ -4,7 +4,7 @@ Engines for transcript comparison against gold-standard references.
 
 CompareEngine (PROCESSING): Aligns main vs gold transcripts word-by-word
 using the same conform/match_fn logic as WER evaluation, then annotates
-each main utterance with comparison tokens (%xsrep / %xsmor).
+each gold utterance with comparison tokens (%xsrep / %xsmor).
 
 CompareAnalysisEngine (ANALYSIS): Reads the comparison annotations and
 computes error-rate metrics for CSV output.
@@ -208,6 +208,11 @@ def _find_best_segment(gold_tokens, main_tokens, mfn):
                 elif len_delta == best_len_delta and end > best[1]:
                     best = (start, end)
 
+    # If no tokens overlap at all, return an empty window so the caller
+    # doesn't consume main tokens that belong to a later gold utterance.
+    if best_score <= 0:
+        return 0, 0
+
     return best
 
 
@@ -249,6 +254,15 @@ def conform_with_mapping(words, conform_fn):
 class CompareEngine(BatchalignEngine):
     tasks = [Task.COMPARE]
 
+    def __init__(self):
+        self._stanza = None
+
+    def _get_stanza(self):
+        if self._stanza is None:
+            from batchalign.pipelines.morphosyntax import StanzaEngine
+            self._stanza = StanzaEngine()
+        return self._stanza
+
     def process(self, doc, **kwargs):
         gold = kwargs.get("gold")
         if not gold or not isinstance(gold, Document):
@@ -256,6 +270,12 @@ class CompareEngine(BatchalignEngine):
                 f"CompareEngine requires a 'gold' Document kwarg, got '{type(gold)}'"
             )
 
+        # --- 0. Run morphosyntax on both docs ---
+        stanza = self._get_stanza()
+        non_gold_kwargs = {k: v for k, v in kwargs.items() if k != "gold"}
+        doc = stanza.process(doc, **non_gold_kwargs)
+        gold = stanza.process(gold, **non_gold_kwargs)
+
         # --- 1. Extract words from main utterances ---
         main_utterances = [
             u for u in doc.content if isinstance(u, Utterance)
@@ -343,16 +363,14 @@ class CompareEngine(BatchalignEngine):
                     gold_form = gold_info[orig_gold_idx][2]
                     last_gold_form_idx = gold_form_idx
 
+                    # Copy timing from main; keep gold's own morphology
                     if main_form.time is not None:
                         gold_form.time = main_form.time
-                    if main_form.morphology is not None:
-                        gold_form.morphology = main_form.morphology
-                    if main_form.dependency is not None:
-                        gold_form.dependency = main_form.dependency
+                    gold_form.dependency = None
 
                     utt_positioned[utt_idx].append((gold_form_idx, CompareToken(
                         text=item.key,
-                        pos=_get_pos(main_form),
+                        pos=_get_pos(gold_form),
                         status="match"
                     )))
                     local_main_cursor += 1
@@ -365,6 +383,8 @@ class CompareEngine(BatchalignEngine):
                         gold_form = gold_info[orig_gold_idx][2]
                         last_gold_form_idx = gold_form_idx
 
+                        gold_form.dependency = None
+
                         utt_positioned[utt_idx].append((gold_form_idx, CompareToken(
                             text=item.key,
                             pos=_get_pos(gold_form),
@@ -395,6 +415,13 @@ class CompareEngine(BatchalignEngine):
                     pos="PUNCT",
                     status="match"
                 )))
+                # Ensure punctuation forms have morphology so %mor includes
+                # the final delimiter (matches generator's PUNCT|<char> logic)
+                if form.morphology is None:
+                    form.morphology = [Morphology(
+                        pos="PUNCT", lemma=form.text.strip(), feats=""
+                    )]
+                form.dependency = None
             utt_positioned[utt_idx].sort(key=lambda x: x[0])
 
         # --- 7. Set comparison on each gold utterance ---

batchalign-0.8.2.post4/batchalign/version

@@ -0,0 +1,3 @@
+0.8.2-post.4
+March 30 2026
+More compare bug fixes

{batchalign-0.8.2.post2 → batchalign-0.8.2.post4/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post2
+Version: 0.8.2.post4
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

batchalign-0.8.2.post2/batchalign/version

@@ -1,3 +0,0 @@
-0.8.2-post.2
-Feb 27th 2026
-adds new feature regarding compare