libs/indic_nlp_library/indicnlp/tokenize/sentence_tokenize.py

#
#  Copyright (c) 2013-present, Anoop Kunchukuttan
#  All rights reserved.
#
#  This source code is licensed under the MIT license found in the
#  LICENSE file in the root directory of this source tree.
#

# Program for sentence splitting of Indian language input
#
# @author Anoop Kunchukuttan
#
"""
Sentence splitter for Indian languages. Contains a rule-based 
sentence splitter that can understand common non-breaking phrases
in many Indian languages.
"""

import regex as re
from indicnlp.transliterate import unicode_transliterate
from indicnlp import langinfo


## for language which have danda as delimiter
## period is not part of the sentence delimiters
DELIM_PAT_DANDA = re.compile(r"[\?!\u0964\u0965]")

## for languages which don't have danda as delimiter
DELIM_PAT_NO_DANDA = re.compile(
    r"[\.\?!\u0964\u0965\uAAF1\uAAF0\uABEB\uABEC\uABED\uABEE\uABEF\u1C7E\u1C7F]"
)

## pattern to check for presence of danda in text
CONTAINS_DANDA = re.compile(r"[\u0964\u0965]")

## pattern to check for presence of valid domain characters in text
CONTAINS_VALID_DOMAIN_CHAR = re.compile(r"^[a-zA-Z0-9_-]$")

## pattern to check for presence of multiple consecutive spaces in text
CONTAINS_MULTIPLE_SPACES = re.compile(" +")


def is_latin_or_numeric(character):
    """
    Check if a character is a Latin character (uppercase or lowercase) or a number.

    Parameters:
        character (str): The character to be checked.

    Returns:
        bool: True if the character is a Latin character or a number, False otherwise.
    """
    return re.match(CONTAINS_VALID_DOMAIN_CHAR, character) is not None


def is_acronym_abbvr(text, lang):
    """Is the text a non-breaking phrase

    Args:
        text (str): text to check for non-breaking phrase
        lang (str): ISO 639-2 language code

    Returns:
        boolean: true if `text` is a non-breaking phrase
    """

    ack_chars = {
        ## acronym for latin characters
        "ए",
        "ऎ",
        "बी",
        "बि",
        "सी",
        "सि",
        "डी",
        "डि",
        "ई",
        "इ",
        "एफ",
        "ऎफ",
        "जी",
        "जि",
        "एच",
        "ऎच",
        "आई",
        "आइ",
        "ऐ",
        "जे",
        "जॆ",
        "के",
        "कॆ",
        "एल",
        "ऎल",
        "एम",
        "ऎम",
        "एन",
        "ऎन",
        "ओ",
        "ऒ",
        "पी",
        "पि",
        "क्यू",
        "क्यु",
        "आर",
        "एस",
        "ऎस",
        "टी",
        "टि",
        "यू",
        "यु",
        "वी",
        "वि",
        "व्ही",
        "व्हि",
        "डब्ल्यू",
        "डब्ल्यु",
        "एक्स",
        "ऎक्स",
        "वाय",
        "जेड",
        "ज़ेड",
        ##  add halant to the previous English character mappings.
        "एफ्",
        "ऎफ्",
        "एच्",
        "ऎच्",
        "एल्",
        "ऎल्",
        "एम्",
        "ऎम्",
        "एन्",
        "ऎन्",
        "आर्",
        "एस्",
        "ऎस्",
        "एक्स्",
        "ऎक्स्",
        "वाय्",
        "जेड्",
        "ज़ेड्",
        # Indic vowels
        "ऄ",
        "अ",
        "आ",
        "इ",
        "ई",
        "उ",
        "ऊ",
        "ऋ",
        "ऌ",
        "ऍ",
        "ऎ",
        "ए",
        "ऐ",
        "ऑ",
        "ऒ",
        "ओ",
        "औ",
        "ॠ",
        "ॡ",
        # Indic consonants
        "क",
        "ख",
        "ग",
        "घ",
        "ङ",
        "च",
        "छ",
        "ज",
        "झ",
        "ञ",
        "ट",
        "ठ",
        "ड",
        "ढ",
        "ण",
        "त",
        "थ",
        "द",
        "ध",
        "न",
        "ऩ",
        "प",
        "फ",
        "ब",
        "भ",
        "म",
        "य",
        "र",
        "ऱ",
        "ल",
        "ळ",
        "ऴ",
        "व",
        "श",
        "ष",
        "स",
        "ह",
        ## abbreviation
        "श्री",
        "डॉ",
        "कु",
        "चि",
        "सौ",
    }

    return (
        unicode_transliterate.UnicodeIndicTransliterator.transliterate(text, lang, "hi")
        in ack_chars
    )


def sentence_split(text, lang, delim_pat="auto"):  ## New signature
    """split the text into sentences

    A rule-based sentence splitter for Indian languages written in
    Brahmi-derived scripts. The text is split at sentence delimiter
    boundaries. The delimiters can be configured by passing appropriate
    parameters.

    The sentence splitter can identify non-breaking phrases like
    single letter, common abbreviations/honorofics for some Indian
    languages.

    Args:
        text (str): text to split into sentence
        lang (str): ISO 639-2 language code
        delim_pat (str): regular expression to identify sentence delimiter characters. If set to 'auto', the delimiter pattern is chosen automatically based on the language and text.


    Returns:
        list: list of sentences identified from the input text
    """

    if lang == "ur":
        from indicnlp.urduhack.tokenization import sentence_tokenizer

        if len(text.split()) < 2:
            sentences = text.split()
        else:
            sentences = sentence_tokenizer(text)
        return sentences

    # print('Input: {}'.format(delim_pat))
    if delim_pat == "auto":
        if langinfo.is_danda_delim(lang):
            # in modern texts it is possible that period is used as delimeter
            # instead of DANDA. Hence, a check. Use danda delimiter pattern
            # only if text contains at least one danda
            if CONTAINS_DANDA.search(text) is None:
                delim_pat = DELIM_PAT_NO_DANDA
                # print('LANG has danda delim. TEXT_CONTAINS_DANDA: FALSE --> DELIM_PAT_NO_DANDA')
            else:
                delim_pat = DELIM_PAT_DANDA
                # print('LANG has danda delim. TEXT_CONTAINS_DANDA: TRUE --> DELIM_PAT_DANDA')
        else:
            delim_pat = DELIM_PAT_NO_DANDA
            # print('LANG has no danda delim --> DELIM_PAT_NO_DANDA')

    ## otherwise, assume the caller set the delimiter pattern

    ### Phase 1: break on sentence delimiters.
    cand_sentences = []
    begin = 0
    text = text.strip()
    for mo in delim_pat.finditer(text):
        p1 = mo.start()
        p2 = mo.end()

        ## NEW
        if p1 > 0 and text[p1 - 1].isnumeric():
            continue

        ## Prevents splitting on "." in URLs/emails in indic texts.
        if lang != "en":
            if is_latin_or_numeric(text[p1 - 1]):
                if p1 + 1 < len(text) and is_latin_or_numeric(text[p1 + 1]):
                    continue

        end = p1 + 1
        s = text[begin:end].strip()
        if len(s) > 0:
            cand_sentences.append(s)
        begin = p1 + 1

    s = text[begin:].strip()
    if len(s) > 0:
        cand_sentences.append(s)

    if not delim_pat.search("."):
        ## run phase 2 only if delimiter pattern contains period
        # print('No need to run phase2')
        return cand_sentences
    #     print(cand_sentences)
    #     print('====')

    #     return cand_sentences

    ### Phase 2: Address the fact that '.' may not always be a sentence delimiter
    ### Method: If there is a run of lines containing only a word (optionally) and '.',
    ### merge these lines as well one sentence preceding and succeeding this run of lines.
    final_sentences = []
    sen_buffer = ""
    bad_state = False

    for i, sentence in enumerate(cand_sentences):
        words = sentence.split(" ")
        # if len(words)<=2 and words[-1]=='.':
        if len(words) == 1 and sentence[-1] == ".":
            bad_state = True
            sen_buffer = sen_buffer + " " + sentence
        ## NEW condition
        elif sentence[-1] == "." and is_acronym_abbvr(words[-1][:-1], lang):
            if len(sen_buffer) > 0 and not bad_state:
                final_sentences.append(sen_buffer)
                sen_buffer = sentence
            else:
                sen_buffer = sen_buffer + " " + sentence
            bad_state = True
        elif bad_state:
            sen_buffer = sen_buffer + " " + sentence
            if len(sen_buffer) > 0:
                final_sentences.append(sen_buffer)
            sen_buffer = ""
            bad_state = False
        else:  ## good state
            if len(sen_buffer) > 0:
                final_sentences.append(sen_buffer)
            sen_buffer = sentence
            bad_state = False

    if len(sen_buffer) > 0:
        final_sentences.append(sen_buffer)

    for i in range(0, len(final_sentences)):
        final_sentences[i] = CONTAINS_MULTIPLE_SPACES.sub(
            " ", final_sentences[i].strip()
        )

    return final_sentences