batchalign 0.7.5a7__tar.gz → 0.7.6a0__tar.gz

{batchalign-0.7.5a7/batchalign.egg-info → batchalign-0.7.6a0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a7
+Version: 0.7.6a0
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -22,7 +22,7 @@ Requires-Dist: plotly>=5.18.0
 Requires-Dist: transformers>=4.37
 Requires-Dist: tokenizers>=0.14.1
 Requires-Dist: pycountry>=22.3
-Requires-Dist: stanza>=1.7
+Requires-Dist: stanza[transformers]>=1.9.1
 Requires-Dist: scipy~=1.11
 Requires-Dist: rev_ai>=2.18.0
 Requires-Dist: rich~=13.6

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/cli/cli.py

@@ -217,6 +217,27 @@ def morphotag(ctx, in_dir, out_dir, **kwargs):
               loader, writer, C)
 
 
+#################### MORPHOTAG ################################
+
+@batchalign.command(hidden=True)
+@common_options
+@click.pass_context
+def coref(ctx, in_dir, out_dir, **kwargs):
+    """Perform coreference analysis on transcripts."""
+
+    def loader(file):
+        cf = CHATFile(path=os.path.abspath(file))
+        doc = cf.doc
+        return doc, {}
+
+    def writer(doc, output):
+        CHATFile(doc=doc).write(output)
+
+    _dispatch("coref", "eng", 1, ["cha"], ctx,
+              in_dir, out_dir,
+              loader, writer, C)
+
+
 #################### UTSEG ################################
 
 @batchalign.command()

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/cli/dispatch.py

@@ -47,6 +47,7 @@ Cmd2Task = {
     "morphotag": "morphosyntax",
     "benchmark": "asr,eval",
     "utseg": "utterance",
+    "coref": "coref",
 }
 
 # this is the main runner used by all functions

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/document.py

@@ -29,7 +29,8 @@ class Task(IntEnum):
     FORCED_ALIGNMENT = 9
     FEATURE_EXTRACT = 10
     MORPHOSYNTAX = 11
-    WER = 12
+    COREF = 12
+    WER = 13
 
 
     DEBUG__G = 0
@@ -51,6 +52,7 @@ TypeMap = {
     Task.FEATURE_EXTRACT: TaskType.ANALYSIS,
     Task.RETRACE_ANALYSIS: TaskType.PROCESSING,
     Task.DISFLUENCY_ANALYSIS: TaskType.PROCESSING,
+    Task.COREF: TaskType.PROCESSING,
     Task.WER: TaskType.ANALYSIS,
 
     Task.DEBUG__G: TaskType.GENERATION,
@@ -69,6 +71,7 @@ TaskFriendlyName = {
     Task.FEATURE_EXTRACT: "Feature Extraction",
     Task.RETRACE_ANALYSIS:  "Retrace Analysis",
     Task.DISFLUENCY_ANALYSIS:  "Disfluncy Analysis",
+    Task.COREF:  "Coreference Resolution",
     Task.WER:  "Word Error Rate",
     Task.DEBUG__G:  "TEST_GENERATION",
     Task.DEBUG__P:  "TEST_PROCESSING",
@@ -103,12 +106,18 @@ class Morphology(BaseModel):
     pos: str # pos like "pron"
     feats: str # string feats "Dem-Acc-S1"
 
+class Coref(BaseModel):
+    start: bool
+    end: bool
+    chain: int
+
 class Form(BaseModel):
     text: str # the text
     # MILISCEONDS
     time: Optional[Tuple[int, int]] = Field(default=None) # word bullet
     morphology: Optional[List[Morphology]] = Field(default=None) # mor
     dependency: Optional[List[Dependency]] = Field(default=None) # gra
+    coreference: Optional[List[Coref]] = Field(default=None) # gra
     type: TokenType = Field(default=TokenType.REGULAR) # whether the field is a regular word (i.e. not a filled pause, not a feature, not a retrace, etc.)
 
 class Tier(BaseModel):

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/formats/chat/generator.py

@@ -33,6 +33,8 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
     gras = []
     has_wor = False
     wor_elems = []
+    has_coref = False
+    coref_elems = []
 
     for i in utterance.content:
         mors.append(i.morphology)
@@ -43,6 +45,21 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
         else:
             wor_elems.append(i.text)
 
+        if i.coreference:
+            has_coref = True
+            coref_str_form = ""
+            for j in i.coreference:
+                coref_str = ""
+                if j.start:
+                    coref_str += "("
+                coref_str += str(j.chain)
+                if j.end:
+                    coref_str += ")"
+                coref_str_form += coref_str
+            coref_elems.append(coref_str_form)
+        else:
+            coref_elems.append("-")
+
         if bool(mors[-1]) != bool(gras[-1]):
             warnings.warn(f"Batchalign has detected a mismatch between lengths of mor and gra tiers for utterance; output will not pass CHATTER; line='{main_line}'")
 
@@ -75,6 +92,9 @@ def generate_chat_utterance(utterance: Utterance, special_mor=False, write_wor=T
     #### WOR LINE GENERATION ####
     if has_wor and write_wor:
         result.append("%wor:\t"+" ".join(wor_elems))
+    if has_coref:
+        result.append("%coref:\t"+" ".join(coref_elems))
+
 
 
     #### EXTRA LINE GENERATION ####

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/models/resolve.py

@@ -12,6 +12,7 @@ resolver = {
     },
     "whisper": {
         'eng': ("talkbank/CHATWhisper-en-large-v1", "openai/whisper-large-v2"),
+        'yue': ("alvanlii/whisper-small-cantonese", "alvanlii/whisper-small-cantonese"),
     }
 }
 

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/pipelines/__init__.py

@@ -2,7 +2,7 @@ from .pipeline import BatchalignPipeline
 from .base import BatchalignEngine
 from .asr import WhisperEngine, RevEngine, WhisperXEngine
 
-from .morphosyntax import StanzaEngine
+from .morphosyntax import StanzaEngine, CorefEngine
 from .cleanup import NgramRetraceEngine, DisfluencyReplacementEngine
 from .speaker import NemoSpeakerEngine
 

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign/pipelines/dispatch.py

@@ -6,7 +6,7 @@ Tabulate default packages and options.
 from batchalign import (WhisperEngine, WhisperFAEngine, StanzaEngine, RevEngine,
                         NgramRetraceEngine, DisfluencyReplacementEngine, WhisperUTREngine,
                         RevUTREngine, EvaluationEngine, WhisperXEngine, NemoSpeakerEngine,
-                        StanzaUtteranceEngine)
+                        StanzaUtteranceEngine, CorefEngine)
 from batchalign import BatchalignPipeline
 from batchalign.models import resolve
 
@@ -27,6 +27,7 @@ DEFAULT_PACKAGES = {
     "retracing": "ngram",
     "eval": "evaluation",
     "utterance": "stanza_utt",
+    "coref": "stanza_coref",
 }
 
 LANGUAGE_OVERRIDE_PACKAGES = {
@@ -124,6 +125,9 @@ def dispatch_pipeline(pkg_str, lang, num_speakers=None, **arg_overrides):
             engines.append(NemoSpeakerEngine(num_speakers=num_speakers))
         elif engine == "stanza_utt":
             engines.append(StanzaUtteranceEngine())
+        elif engine == "stanza_coref":
+            engines.append(CorefEngine())
+
 
     L.debug(f"Done initalizing packages.")
     return BatchalignPipeline(*engines)

batchalign-0.7.6a0/batchalign/pipelines/morphosyntax/__init__.py

@@ -0,0 +1,3 @@
+from .ud import StanzaEngine
+from .coref import CorefEngine
+

batchalign-0.7.6a0/batchalign/pipelines/morphosyntax/coref.py

@@ -0,0 +1,45 @@
+import stanza
+from batchalign.utils.dp import PayloadTarget, ReferenceTarget, Match, align
+from warnings import warn
+from batchalign.document import *
+from batchalign.constants import *
+from batchalign.pipelines.base import *
+from batchalign.formats.chat.parser import chat_parse_utterance
+
+from batchalign.utils.dp import *
+
+
+
+class CorefEngine(BatchalignEngine):
+    tasks = [ Task.COREF ]
+
+    def process(self, doc, **kwargs):
+        if "eng" not in doc.langs:
+            warn("Coreference resolution is only supported for English documents.")
+            return 
+
+        detokenized = " ".join([i.strip(include_retrace=True, include_fp=True) for i in doc.content if isinstance(i, Utterance)])
+        pipeline = stanza.Pipeline(lang="en", processors="tokenize, coref")
+
+        coref_chains = pipeline(detokenized).sentences
+        coref_chains = [(j.text,
+                        [Coref(start=chain.is_start,
+                                end=chain.is_end,
+                                chain=chain.chain.index) for chain in j.coref_chains])
+                        for i in coref_chains
+                        for j in i.words]
+
+        payloads = [PayloadTarget(i[0], i[1]) for i in coref_chains]
+        references = [ReferenceTarget(j.text, (ut_id, form_id)) for ut_id, i in enumerate(doc.content)
+                        if isinstance(i, Utterance)
+                        for form_id, j in enumerate(i.content)]
+        alignment = align(payloads, references, tqdm=False)
+
+        for i in alignment:
+            if isinstance(i, Match):
+                (ut, form) = i.reference_payload
+                doc.content[ut].content[form].coreference = i.payload
+
+        return doc
+
+

batchalign-0.7.6a0/batchalign/version

@@ -0,0 +1,3 @@
+0.7.6-alpha.0
+September 27th, 2024
+initial coreference support

{batchalign-0.7.5a7 → batchalign-0.7.6a0/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a7
+Version: 0.7.6a0
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -22,7 +22,7 @@ Requires-Dist: plotly>=5.18.0
 Requires-Dist: transformers>=4.37
 Requires-Dist: tokenizers>=0.14.1
 Requires-Dist: pycountry>=22.3
-Requires-Dist: stanza>=1.7
+Requires-Dist: stanza[transformers]>=1.9.1
 Requires-Dist: scipy~=1.11
 Requires-Dist: rev_ai>=2.18.0
 Requires-Dist: rich~=13.6

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign.egg-info/SOURCES.txt

@@ -70,6 +70,7 @@ batchalign/pipelines/cleanup/support/test.test
 batchalign/pipelines/fa/__init__.py
 batchalign/pipelines/fa/whisper_fa.py
 batchalign/pipelines/morphosyntax/__init__.py
+batchalign/pipelines/morphosyntax/coref.py
 batchalign/pipelines/morphosyntax/ud.py
 batchalign/pipelines/morphosyntax/fr/case.py
 batchalign/pipelines/morphosyntax/ja/verbforms.py

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/batchalign.egg-info/requires.txt

@@ -12,7 +12,7 @@ plotly>=5.18.0
 transformers>=4.37
 tokenizers>=0.14.1
 pycountry>=22.3
-stanza>=1.7
+stanza[transformers]>=1.9.1
 scipy~=1.11
 rev_ai>=2.18.0
 rich~=13.6

{batchalign-0.7.5a7 → batchalign-0.7.6a0}/setup.py

@@ -40,7 +40,7 @@ setup(
         "transformers>=4.37",
         "tokenizers>=0.14.1",
         "pycountry>=22.3",
-        "stanza>=1.7",
+        "stanza[transformers]>=1.9.1",
         "scipy~=1.11",
         "rev_ai>=2.18.0",
         "rich~=13.6",

batchalign-0.7.5a7/batchalign/pipelines/morphosyntax/__init__.py

@@ -1 +0,0 @@
-from .ud import StanzaEngine

batchalign-0.7.5a7/batchalign/version

@@ -1,3 +0,0 @@
-0.7.5-alpha.7
-September 7th, 2024
-batch hanging utterance bug