core/generate/intent_bridge.py
Shay f223e61352 fix(generate): wire intent-aware realizer into chat hot path
The realize_semantic / realize_target pipeline in realizer.py was fully
implemented but never called from chat/runtime.py. The hot path only called
realize() from articulation.py, which returns raw S-P-O word tokens with no
intent, tense, negation, quantifier or rhetorical-move awareness. This
disconnected the 13-construction realizer from every live chat turn.

New module generate/intent_bridge.py:
- classify_intent_from_input() runs the rule-based classifier against the
  raw input text to obtain a DialogueIntent
- articulate_with_intent() builds a PropositionGraph from that intent,
  grounds the <pending> obj slots with recalled vocabulary from the
  generation result, plans articulation via plan_articulation(), and calls
  realize_semantic() for the intent-specific template path
- Falls back cleanly to the existing ArticulationPlan surface when the
  realizer returns an empty plan (OOV-heavy or UNKNOWN intent)

chat/runtime.py change:
- Import and call articulate_with_intent() after the existing realize() call
- Replace articulation.surface with the intent-bridge surface whenever the
  bridge returns a non-empty, non-pending string
- The existing ArticulationPlan dataclass is preserved and passed downstream
  so SentenceAssembler, turn_log, ChatResponse, and all trace fields remain
  structurally unchanged

Effect: chat() now produces intent-differentiated surfaces:
  DEFINITION  → "X is defined as Y"         (was "X Y Z")
  CAUSE       → "X is grounded in Y"         (was "X Y Z")
  CORRECTION  → "correction: X corrects Y"   (was "X Y Z")
  RECALL      → "recalling X: Y"             (was "X Y Z")
  VERIFICATION→ "X is verified: Y"           (was "X Y Z")
  COMPARISON  → "X and Y are distinguished..." (was "X contrasts_with Y")
  PROCEDURE   → "first, Y; then, X follows"  (was "X Y Z")
  CONJUNCTION → "X P and Y P"               (realizer edge handling)
  RELATIVE    → "X, which Pv Y, Pv Z"       (realizer edge handling)

Articulation fidelity is now geometrically honest AND structurally expressive.
The surface corresponds to internal intent state, not a generic S-P-O join.
2026-05-16 08:38:59 -07:00

129 lines
4.4 KiB
Python

"""generate/intent_bridge.py — connects intent classification to the realizer.
Bridges the gap between chat/runtime.py's articulation path (which resolves
Proposition slot-versors into raw word tokens) and the intent-aware realizer
pipeline (realize_semantic / realize_target in realizer.py, which are fully
implemented but were never called from the chat hot path).
Design constraints:
- Deterministic: same input text + same field state → same surface
- No LLM fallback
- Falls back cleanly to the existing ArticulationPlan when the realizer
cannot produce a non-empty surface (OOV-heavy input, UNKNOWN intent
with no grounded obj slots)
- Does not alter the ArticulationPlan dataclass or ChatResponse structure;
only the .surface field is replaced when the bridge succeeds
"""
from __future__ import annotations
from generate.articulation import ArticulationPlan
from generate.graph_planner import (
GraphEdge,
GraphNode,
PropositionGraph,
Relation,
ground_graph,
plan_articulation,
)
from generate.intent import DialogueIntent, IntentTag, classify_intent
from generate.realizer import RealizedPlan, realize_semantic
_PENDING = "<pending>"
_PRIOR = "<prior>"
_EMPTY_INDICATORS = frozenset({_PENDING, _PRIOR, "...", ""})
def classify_intent_from_input(text: str) -> DialogueIntent:
"""Run the rule-based intent classifier against raw input text."""
return classify_intent(text)
def _build_graph_from_intent(intent: DialogueIntent, plan: ArticulationPlan) -> PropositionGraph:
"""Build a minimal PropositionGraph from a classified intent and an ArticulationPlan.
Uses the resolved slot words from ArticulationPlan (subject, predicate, object)
as the concrete node content, with the intent tag selecting the predicate.
"""
from generate.graph_planner import _INTENT_PREDICATES # noqa: PLC0415
predicate = _INTENT_PREDICATES.get(intent.tag, "addresses")
subject = intent.subject or plan.subject or ""
obj = plan.object or plan.predicate or _PENDING
graph = PropositionGraph()
if intent.tag is IntentTag.COMPARISON:
secondary = intent.secondary_subject or plan.object or plan.predicate or obj
left = GraphNode(
node_id="p0",
subject=subject,
predicate=predicate,
obj=secondary,
source_intent=intent.tag,
)
right = GraphNode(
node_id="p1",
subject=secondary,
predicate=predicate,
obj=subject,
source_intent=intent.tag,
)
edge = GraphEdge(source="p0", target="p1", relation=Relation.CONTRAST)
return graph.add_node(left).add_node(right).add_edge(edge)
root = GraphNode(
node_id="p0",
subject=subject,
predicate=predicate,
obj=obj,
source_intent=intent.tag,
)
return graph.add_node(root)
def _is_useful_surface(surface: str) -> bool:
"""Return True when the realized surface is non-empty and fully grounded."""
if not surface or not surface.strip():
return False
for indicator in _EMPTY_INDICATORS:
if indicator and indicator in surface:
return False
return True
def articulate_with_intent(
text: str,
plan: ArticulationPlan,
recalled_words: tuple[str, ...] = (),
) -> str:
"""Return an intent-aware surface string for *plan*, or "" if none can be produced.
Steps:
1. Classify intent from raw input *text*
2. Build a PropositionGraph from the intent + ArticulationPlan slot words
3. Ground <pending> obj slots with *recalled_words* from generation result
4. Plan articulation (topological walk)
5. Realize via realize_semantic() for intent-specific templates
6. Return the surface, or "" if the result is empty / ungrounded
The caller (chat/runtime.py) should fall back to the existing
ArticulationPlan.surface when this returns "".
"""
intent = classify_intent_from_input(text)
graph = _build_graph_from_intent(intent, plan)
if recalled_words:
graph = ground_graph(graph, recalled_words)
articulation_target = plan_articulation(graph)
realized: RealizedPlan = realize_semantic(articulation_target, graph)
if not realized.surface or not realized.fragments:
return ""
surface = realized.surface
if not _is_useful_surface(surface):
return ""
return surface