Datasets:
File size: 6,227 Bytes
0bf2389 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 | from copy import deepcopy
import unittest
from barbet_bgd.taiwan_usage import (
audit_taiwan_usage, iter_taiwan_usage_audits, TaiwanUsageAccumulator,
)
def example(text="請使用軟體與伺服器，款項為新臺幣 1,280 元。"):
return {
"record_id": "usage-test", "primitive": "EDIT",
"task_spec": {"instruction": "依據來源修正錯誤。", "query": "保留正確內容。"},
"inputs": [{"id": "ref", "type": "reference_text", "units": [
{"id": "ref:u1", "text": text}]}],
"target": {"output_text": text, "action": {"tool": "EDIT", "arguments": {}}},
}
class TaiwanUsageTests(unittest.TestCase):
def test_native_terms_variants_names_codes_and_schema_are_allowed(self):
row = example("台北市里長林台生透過 API 查詢軟體版本 v2.3.1，並以 JSON Schema 驗證資料。")
row["inputs"][0]["units"][0]["code"] = "打印機.load(\"软件\")"
row["task_spec"]["requested_schema"] = {"properties": {"软件": {"type": "string"}}}
self.assertTrue(audit_taiwan_usage(row)["passed"])
def test_unconverted_simplified_and_converted_mainland_terms_fail(self):
for text, rule in [("请開啟檔案。", "SIMPLIFIED_FORM"), ("請登入服務器。", "TAIWAN_PREFERRED_TERM")]:
with self.subTest(text=text):
result = audit_taiwan_usage(example(text))
self.assertFalse(result["passed"])
self.assertIn(rule, {item["rule_id"] for item in result["findings"]})
def test_query_constraints_and_nested_target_values_are_scanned(self):
row = example()
row["task_spec"]["query"] = "請掃描二維碼。"
row["task_spec"]["constraints"] = {"desired_text": "開啟數據庫。"}
row["target"]["fields"] = [{"value": "默認值"}]
paths = {finding["path"] for finding in audit_taiwan_usage(row)["findings"]}
self.assertIn("task_spec.query", paths)
self.assertIn("task_spec.constraints.desired_text", paths)
self.assertIn("target.fields[0].value", paths)
def test_inline_code_mask_retains_exact_prose_offsets(self):
text = "呼叫 `软件.load()` 後使用服務器。"
result = audit_taiwan_usage(example(text))
self.assertEqual({f["matched"] for f in result["findings"]}, {"服務器"})
for finding in result["findings"]:
start, end = finding["character_span"]
self.assertEqual(text[start:end], finding["matched"])
def test_candidate_is_audited_but_not_treated_as_positive_target(self):
row = example()
row["inputs"].append({"id": "hyp", "type": "asr_hypothesis", "units": [
{"id": "hyp:u1", "text": "请使用软件。"}]})
result = audit_taiwan_usage(row)
self.assertTrue(result["passed"])
self.assertGreater(result["exempt_count"], 0)
row["target"]["output_text"] = "请使用软件。"
self.assertFalse(audit_taiwan_usage(row)["passed"])
def test_selected_candidate_never_inherits_negative_exemption(self):
for primitive in ("MATCH", "RANK"):
row = example()
row["primitive"] = primitive
row["inputs"].append({"id": "candidates", "type": "entity_candidates", "units": [
{"id": "cand:a", "text": "服務器"}]})
row["target"].update({"canonical_id": "cand:a", "ranking": ["cand:a"]})
self.assertFalse(audit_taiwan_usage(row)["passed"])
def test_faithful_copy_exemption_requires_exact_reconstruction(self):
row = example("原文件寫著：请登入服務器。")
row["primitive"] = "COPY"
row["target"]["action"] = {"tool": "COPY", "arguments": {"source_units": ["ref:u1"]}}
result = audit_taiwan_usage(row)
self.assertTrue(result["passed"])
self.assertGreater(result["exempt_count"], 0)
row["target"]["output_text"] += "附註"
self.assertFalse(audit_taiwan_usage(row)["passed"])
def test_accept_action_checks_accepted_candidate_as_positive(self):
row = example()
row["inputs"].append({"id": "hyp", "type": "asr_hypothesis", "units": [
{"id": "hyp:u1", "text": "服務器"}]})
row["target"]["action"] = {"tool": "ACCEPT", "arguments": {"candidate_id": "hyp"}}
self.assertFalse(audit_taiwan_usage(row)["passed"])
def test_known_generator_naturalness_defects_are_rejected(self):
for text in (
"計畫將於2028 年 2 月 10 日伺服器。",
"金額：2028 年 2 月 10 日。",
"版本為第二點三點一。",
"日期為2027 年 2 月 29 日。",
):
with self.subTest(text=text):
self.assertEqual(audit_taiwan_usage(example(text))["decision"], "REJECT")
self.assertTrue(audit_taiwan_usage(example("日期為2028 年 2 月 29 日。"))["passed"])
def test_synthetic_error_labels_are_not_excused_as_negative_examples(self):
row = example()
row["inputs"].append({"type": "ocr_candidates", "units": [
{"id": "cand:a", "text": "金額：不符-amount-2-93。"}]})
result = audit_taiwan_usage(row)
self.assertEqual(result["decision"], "REJECT")
def test_english_technical_prose_review_is_narrow(self):
self.assertEqual(audit_taiwan_usage(example("請比對同一位 pseudo speaker。"))["decision"], "REVIEW")
self.assertTrue(audit_taiwan_usage(example("GPU 使用率為 92.5%，保留 token ID。"))["passed"])
def test_streaming_bounded_accumulator_does_not_modify_records(self):
row = example("請登入服務器。")
original = deepcopy(row)
accumulator = TaiwanUsageAccumulator(example_limit=1)
for verdict in iter_taiwan_usage_audits(row for _ in range(10)):
accumulator.update(verdict)
result = accumulator.report()
self.assertEqual(result["records"], 10)
self.assertEqual(result["decision_counts"], {"REJECT": 10})
self.assertTrue(all(len(values) <= 1 for values in result["examples"].values()))
self.assertEqual(row, original)
if __name__ == "__main__":
unittest.main()
|