|
|
| import json |
| import requests |
| import random |
|
|
| ARCHETYPE_DESC = ( |
| "An abandoned clause, false start, or spoken restart. The speaker begins saying one thought or syntactic structure, " |
| "cuts themselves off, and starts their sentence over. Patterns include aborted clauses, interrupted questions, " |
| "abandoned conditionals, or sentence-initial/mid-sentence restarts. MUST be linguistically varied and natural." |
| ) |
|
|
| DIVERSE_FS_EXAMPLES = [ |
| ("Let's deploy the update tonight.", "If we could— but actually Let's deploy the update tonight.", ["If we could— but actually "]), |
| ("We need to upgrade the database version.", "Regarding our plan to— well actually We need to upgrade the database version.", ["Regarding our plan to— well actually "]), |
| ("The transaction failed due to network timeout.", "When they submitted the— so what happened was The transaction failed due to network timeout.", ["When they submitted the— so what happened was "]), |
| ("You can find the documentation on the intranet.", "Where did we put— okay You can find the documentation on the intranet.", ["Where did we put— okay "]), |
| ("The report will be ready by Thursday morning.", "As far as completing the— well The report will be ready by Thursday morning.", ["As far as completing the— well "]), |
| ("We should consider alternative vendors.", "It might be better to— or rather We should consider alternative vendors.", ["It might be better to— or rather "]), |
| ("Please send the tracking number to the customer.", "Could you check if— actually Please send the tracking number to the customer.", ["Could you check if— actually "]), |
| ("Our team finished the sprint ahead of schedule.", "In terms of progress on the— so Our team finished the sprint ahead of schedule.", ["In terms of progress on the— so "]), |
| ("There are several flights available tomorrow.", "Looking at the morning— well There are several flights available tomorrow.", ["Looking at the morning— well "]), |
| ("The conference was postponed until next month.", "Did they announce when— anyway The conference was postponed until next month.", ["Did they announce when— anyway "]), |
| ] |
|
|
| PROMPT_TEMPLATE = """<task> |
| Corrupt the input clean sentence into ONE natural spoken-disfluency example using the requested SPEECH ARCHETYPE: |
| Archetype: FALSE_START |
| Archetype Description: {archetype_desc} |
| |
| CRITICAL RULES: |
| 1. Every word, character, and casing outside the disfluent span MUST remain 100% VERBATIM IDENTICAL to the input sentence. |
| - If the input starts with a capitalized word like 'The', keep 'The' capitalized: 'When they started— The...'. Do NOT lowercase it. |
| 2. Excising the disfluent span(s) MUST recreate the input clean sentence exactly. Do NOT delete or alter any clean words or punctuation. |
| 3. Return ONLY two fields: and . |
| 4. MUST contain the exact verbatim substrings to excise, matching the exact casing in . |
| 5. FORBIDDEN PATTERNS: |
| - Do NOT use generic conversational filler preambles like 'I mean', 'I was wondering', 'I think', 'Basically', or 'So I mean'. |
| - A false start MUST contain an aborted thought, interrupted clause, or abandoned sentence fragment (e.g. 'If we could—', 'When they—', 'Regarding our—', 'Could you check—'). |
| 6. DO NOT delete, shorten, or excise valid sentence-initial subject pronouns ('I', 'We', 'It', 'You', 'They'). |
| 7. VARY your false starts widely! Adapt the aborted thought to the specific topic/context of the sentence. |
| </task> |
| |
| <examples> |
| {examples_block} |
| </examples> |
| |
| <inputs> |
| Input: {sentence} |
| </inputs> |
| |
| Respond with JSON only: |
| {{"corrupted_text": "...", "disfluent_spans": ["..."]}} |
| """ |
|
|
| test_sentences = [ |
| "The project deadline has been moved to next Friday.", |
| "You should restart the web application after modifying the config.", |
| "We noticed an unusual spike in memory consumption on node three.", |
| "Customers reported issues when checking out with international cards.", |
| "The algorithm converges much faster with a smaller learning rate." |
| ] |
|
|
| rng = random.Random(42) |
| for sent in test_sentences: |
| sampled_ex = rng.sample(DIVERSE_FS_EXAMPLES, 3) |
| ex_lines = [] |
| for clean_in, corrupted_out, spans in sampled_ex: |
| ex_lines.append(f"Input: {clean_in} |
| Response: |
| " + json.dumps({ |
| "corrupted_text": corrupted_out, |
| "disfluent_spans": spans |
| }, indent=2)) |
| examples_block = " |
| |
| ".join(ex_lines) |
| prompt = PROMPT_TEMPLATE.format( |
| archetype_desc=ARCHETYPE_DESC, |
| examples_block=examples_block, |
| sentence=sent |
| ) |
| payload = { |
| "model": "gemma4:12b", |
| "messages": [{"role": "user", "content": prompt}], |
| "stream": False, |
| "think": False, |
| "format": "json", |
| "options": {"temperature": 0.70, "top_p": 0.95, "num_predict": 256} |
| } |
| r = requests.post("http://127.0.0.1:11434/api/chat", json=payload, timeout=25) |
| res = r.json() |
| content = res.get("message", {}).get("content", "") |
| try: |
| parsed = json.loads(content) |
| raw = parsed.get("corrupted_text", "") |
| spans = parsed.get("disfluent_spans", []) |
| print(f"CLEAN: {sent}") |
| print(f"RAW: {raw}") |
| print(f"SPANS: {spans}") |
| test_clean = raw |
| for s in spans: |
| test_clean = test_clean.replace(s, "") |
| print(f"ROUNDTRIP OK: {test_clean.strip() == sent.strip()}") |
| print("-" * 60) |
| except Exception as e: |
| print(f"Error: {e}") |
|
|
