|
|
|
|
|
|
|
|
import random |
|
|
from typing import List, Dict, Any |
|
|
|
|
|
from .meaning_model import StructuredMeaning, Context |
|
|
from .encoder import encode_term |
|
|
|
|
|
|
|
|
def generate_context_rich_samples( |
|
|
dictionaries: Dict[str, Any], |
|
|
count: int, |
|
|
source_language: str = "en", |
|
|
) -> List[Dict[str, Any]]: |
|
|
""" |
|
|
Generates samples that combine: |
|
|
- actor |
|
|
- action |
|
|
- object |
|
|
- time context |
|
|
- place context |
|
|
- activity context |
|
|
""" |
|
|
|
|
|
actions = dictionaries.get("actions", []) |
|
|
objects = dictionaries.get("objects", []) |
|
|
actors = dictionaries.get("actors", []) |
|
|
times = dictionaries.get("context_time", []) |
|
|
places = dictionaries.get("context_place", []) |
|
|
activities = dictionaries.get("context_activity", []) |
|
|
|
|
|
results = [] |
|
|
|
|
|
if not actions or not objects or not actors: |
|
|
raise ValueError("Missing required dictionaries: actions, objects, actors.") |
|
|
|
|
|
for _ in range(count): |
|
|
action = random.choice(actions) |
|
|
obj = random.choice(objects) |
|
|
actor = random.choice(actors) |
|
|
|
|
|
time_ctx = random.choice(times) if times else None |
|
|
place_ctx = random.choice(places) if places else None |
|
|
activity_ctx = random.choice(activities) if activities else None |
|
|
|
|
|
action_id = action.get("id", "action.unknown") |
|
|
object_id = obj.get("id", "object.unknown") |
|
|
actor_id = actor.get("id", "actor.unknown") |
|
|
|
|
|
time_id = time_ctx.get("id") if time_ctx else None |
|
|
place_id = place_ctx.get("id") if place_ctx else None |
|
|
activity_id = activity_ctx.get("id") if activity_ctx else None |
|
|
|
|
|
input_text = f"{actor_id} {action_id} the {object_id}" |
|
|
if place_id: |
|
|
input_text += f" in the {place_id}" |
|
|
if activity_id: |
|
|
input_text += f" while {activity_id}" |
|
|
if time_id: |
|
|
input_text += f" ({time_id})" |
|
|
input_text += "." |
|
|
|
|
|
meaning = StructuredMeaning( |
|
|
actor=actor_id, |
|
|
action=action_id, |
|
|
object=object_id, |
|
|
modifiers=[], |
|
|
emotion=None, |
|
|
context=Context( |
|
|
time=time_id, |
|
|
place=place_id, |
|
|
activity=activity_id, |
|
|
), |
|
|
intent="intent.describe_event", |
|
|
meta={ |
|
|
"source_language": source_language, |
|
|
"confidence": 0.95, |
|
|
}, |
|
|
) |
|
|
|
|
|
actor_enc = encode_term(actor_id, dictionaries) |
|
|
action_enc = encode_term(action_id, dictionaries) |
|
|
object_enc = encode_term(object_id, dictionaries) |
|
|
time_enc = encode_term(time_id, dictionaries) if time_id else None |
|
|
place_enc = encode_term(place_id, dictionaries) if place_id else None |
|
|
activity_enc = encode_term(activity_id, dictionaries) if activity_id else None |
|
|
|
|
|
glyphic_human = ( |
|
|
f"ACTOR[{actor_enc['human']}] " |
|
|
f"ACTION[{action_enc['human']}] " |
|
|
f"OBJECT[{object_enc['human']}]" |
|
|
) |
|
|
if time_enc: |
|
|
glyphic_human += f" CONTEXT.TIME[{time_enc['human']}]" |
|
|
if place_enc: |
|
|
glyphic_human += f" CONTEXT.PLACE[{place_enc['human']}]" |
|
|
if activity_enc: |
|
|
glyphic_human += f" CONTEXT.ACTIVITY[{activity_enc['human']}]" |
|
|
glyphic_human += " INTENT.describe" |
|
|
|
|
|
glyphic_compact = ( |
|
|
f"ACT{{{actor_enc['compact']}}} " |
|
|
f"ACTN{{{action_enc['compact']}}} " |
|
|
f"OBJ{{{object_enc['compact']}}}" |
|
|
) |
|
|
if time_enc: |
|
|
glyphic_compact += f" CTX{{TIME:{time_enc['compact']}}}" |
|
|
if place_enc: |
|
|
glyphic_compact += f" CTX{{PLACE:{place_enc['compact']}}}" |
|
|
if activity_enc: |
|
|
glyphic_compact += f" CTX{{ACT:{activity_enc['compact']}}}" |
|
|
glyphic_compact += " INT{DESC}" |
|
|
|
|
|
glyphic_tokens = ( |
|
|
f"{actor_enc['tokens']} " |
|
|
f"{action_enc['tokens']} " |
|
|
f"{object_enc['tokens']}" |
|
|
) |
|
|
if time_enc: |
|
|
glyphic_tokens += f" <CTX:TIME:{time_enc['compact']}>" |
|
|
if place_enc: |
|
|
glyphic_tokens += f" <CTX:PLACE:{place_enc['compact']}>" |
|
|
if activity_enc: |
|
|
glyphic_tokens += f" <CTX:ACT:{activity_enc['compact']}>" |
|
|
glyphic_tokens += " <INT:DESC>" |
|
|
|
|
|
results.append( |
|
|
{ |
|
|
"input_text": input_text, |
|
|
"glyphic_output_human": glyphic_human, |
|
|
"glyphic_output_compact": glyphic_compact, |
|
|
"glyphic_output_tokens": glyphic_tokens, |
|
|
"structured_meaning": meaning.to_dict(), |
|
|
} |
|
|
) |
|
|
|
|
|
return results |
|
|
|
|
|
|
