convert_dialekt.py

import os
import re
from typing import Dict

import jsonlines
from tqdm import tqdm

TARGET = ".data/dialekt_v2_ort.vert"


def process_vert_format(vert_content: str) -> Dict[str, str]:
    # Pattern to match document boundaries and extract metadata
    doc_pattern = re.compile(r'<doc[^>]*>.*?</doc>', re.DOTALL)
    metadata_pattern = re.compile(
        r'<doc id="([^"]*)".+?rok="([^"]*)".+?misto="([^"]*)" sidlotyp="([^"]*)".+?tema="([^"]*)" pocetml="([^"]*)"')

    # Pattern to match speaker turns
    sp_pattern = re.compile(r'<sp[^>]*id="([^"]*)"[^>]*prezdivka="([^"]*)"\s*[^>]*>(.*?)</sp>', re.DOTALL)

    # Pattern to remove whitespace before punctuation
    ws_before_punct = re.compile(r'\s+([.,!?])')

    # Find all documents
    documents = re.findall(doc_pattern, vert_content)
    processed_documents = {}

    for doc in tqdm(documents):
        # Extract metadata
        metadata_match = re.search(metadata_pattern, doc)
        if metadata_match:
            # r'<doc id="([^"]*)".+?rok="([^"]*)".+?misto="([^"]*)" sidlotyp="([^"]*)".+?tema="([^"]*)" pocetml="([^"]*)"')

            doc_id = metadata_match.group(1)
            year = metadata_match.group(2)
            place = metadata_match.group(3)
            settlement_type = metadata_match.group(4)
            topic = metadata_match.group(5)
            speakers = metadata_match.group(6)

            metadata_str = (f"Rok: {year}, "
                            f"Typ sídla: {settlement_type}, "
                            f"Místo: {place}, "
                            f"Téma: {topic}, "
                            f"Počet mluvčích: {speakers}")
        else:
            raise ValueError("Metadata not found in document")

        # Initialize an empty list to hold processed document text
        processed_document = [metadata_str]

        # Find all speaker turns within the document
        for sp_match in re.findall(sp_pattern, doc):
            speaker_id = sp_match[0]
            speaker_nickname = sp_match[1]
            sp_content = sp_match[2]

            # if speaker is Y, rename him as Jiný zvuk
            if speaker_id == "Y":
                speaker_id = "Zvuk"

            # remove symbols ---, ...:,
            sp_content = sp_content.replace("---", "")
            sp_content = sp_content.replace("...:", "")
            sp_content = sp_content.replace("...", "")
            sp_content = sp_content.replace("..", "")
            sp_content = sp_content.replace("?.", "?")

            # remove tags from each line, and join text
            tokens = [line.split("\t")[0].strip() for line in sp_content.split("\n") if line.strip() != ""]
            speaker_text = " ".join(tokens)

            # replace more than one space with one space
            speaker_text = re.sub(r'\s+', ' ', speaker_text).strip()

            # remove whitespace before ., !, ?
            speaker_text = re.sub(ws_before_punct, r'\1', speaker_text)

            # - sometimes lines in oral are empty? e.g. 08A009N // REMOVE THESE LINES
            if speaker_text.strip() == "":
                continue
            # If the turn is not finished by ., ?, !, add .
            if speaker_text[-1] not in [".", "!", "?"]:
                speaker_text += "."
            # Capitalize the first letter of the speaker turn
            speaker_text = speaker_text[0].upper() + speaker_text[1:]

            # sometimes there is @ in the text, remove it
            speaker_text = speaker_text.replace("@", "")

            # capitalize first (non whitespace) letter after ., !, ?
            speaker_text = re.sub(r'([.!?]\s*)(\w)', lambda m: m.group(1) + m.group(2).upper(), speaker_text)
            speaker_text = speaker_text.replace("overlap", "překryv mluvčích")

            # Format the speaker turn and add to the processed document list
            if speakers == "1":
                processed_document.append(speaker_text)
            else:
                processed_document.append(f"[mluvčí: {speaker_nickname}] {speaker_text}")

        # Join all speaker turns into a single string for the document
        if speakers == "1":
            final_text = processed_document[0] + "\n" + " ".join(processed_document[1:])
        else:
            final_text = '\n'.join(processed_document)
        processed_documents[doc_id] = final_text

    return processed_documents


# Read the content from the file
with open(TARGET, "r") as f:
    vert_content = f.read()

# Process the content
processed_documents = process_vert_format(vert_content)

# write all splits into same json file in  .data/hf_dataset/cnc_fictree/test.jsonl
OF = ".data/hf_dataset/cnc_dialekt/test.jsonl"
os.makedirs(os.path.dirname(OF), exist_ok=True)
with jsonlines.open(OF, "w") as writer:
    for doc_id, doc in list(processed_documents.items()):
        writer.write({"text": doc, "id": doc_id})