Compare LM on Spelling Correction#

This tutorial is available as an IPython notebook at Malaya/example/compare-lm-spelling-correction.

Malaya got 3 different LM models,

KenLM
GPT2
Masked LM

So we are going to compare the spelling correction results.

[1]:

import os

os.environ['CUDA_VISIBLE_DEVICES'] = ''
os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'

[2]:

import logging

logging.basicConfig(level=logging.INFO)

[3]:

import malaya

/home/husein/dev/malaya/malaya/tokenizer.py:202: FutureWarning: Possible nested set at position 3361
  self.tok = re.compile(r'({})'.format('|'.join(pipeline)))
/home/husein/dev/malaya/malaya/tokenizer.py:202: FutureWarning: Possible nested set at position 3879
  self.tok = re.compile(r'({})'.format('|'.join(pipeline)))

[4]:

# some text examples copied from Twitter

string1 = 'krajaan patut bagi pencen awal skt kpd warga emas supaya emosi'
string2 = 'Husein ska mkn aym dkat kampng Jawa'
string3 = 'Melayu malas ni narration dia sama je macam men are trash. True to some, false to some.'
string4 = 'Tapi tak pikir ke bahaya perpetuate myths camtu. Nanti kalau ada hiring discrimination despite your good qualifications because of your race tau pulak marah. Your kids will be victims of that too.'
string5 = 'DrM cerita Melayu malas semenjak saya kat University (early 1980s) and now as i am edging towards retirement in 4-5 years time after a career of being an Engineer, Project Manager, General Manager'
string6 = 'blh bntg dlm kls nlp sy, nnti intch'
string7 = 'mulakn slh org boleh ,bila geng tuh kena slhkn jgk xboleh trima .. pelik'

Load probability model#

def load(
    language_model=None,
    sentence_piece: bool = False,
    stemmer=None,
    **kwargs,
):
    """
    Load a Probability Spell Corrector.

    Parameters
    ----------
    language_model: Callable, optional (default=None)
        If not None, must an object with `score` method.
    sentence_piece: bool, optional (default=False)
        if True, reduce possible augmentation states using sentence piece.
    stemmer: Callable, optional (default=None)
        a Callable object, must have `stem_word` method.

    Returns
    -------
    result: model
        List of model classes:

        * if passed `language_model` will return `malaya.spelling_correction.probability.ProbabilityLM`.
        * else will return `malaya.spelling_correction.probability.Probability`.
    """

[5]:

kenlm = malaya.language_model.kenlm()
kenlm

[5]:

<Model from b'model.klm'>

[6]:

malaya.language_model.available_mlm()

[6]:

	Size (MB)
malay-huggingface/bert-base-bahasa-cased	310.0
malay-huggingface/bert-tiny-bahasa-cased	66.1
malay-huggingface/albert-base-bahasa-cased	45.9
malay-huggingface/albert-tiny-bahasa-cased	22.6
mesolitica/roberta-base-bahasa-cased	443.0
mesolitica/roberta-tiny-bahasa-cased	66.1

[10]:

bert_base = malaya.language_model.mlm(model = 'malay-huggingface/bert-base-bahasa-cased')
bert_tiny = malaya.language_model.mlm(model = 'malay-huggingface/bert-tiny-bahasa-cased')
albert_base = malaya.language_model.mlm(model = 'malay-huggingface/albert-base-bahasa-cased')
albert_tiny = malaya.language_model.mlm(model = 'malay-huggingface/albert-tiny-bahasa-cased')
roberta_base = malaya.language_model.mlm(model = 'mesolitica/roberta-base-bahasa-cased')

[7]:

malaya.language_model.available_gpt2()

[7]:

	Size (MB)
mesolitica/gpt2-117m-bahasa-cased	454

[8]:

gpt2 = malaya.language_model.gpt2(model = 'mesolitica/gpt2-117m-bahasa-cased')

[20]:

model_kenlm = malaya.spelling_correction.probability.load(language_model = kenlm)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[21]:

model_bert_base = malaya.spelling_correction.probability.load(language_model = bert_base)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[22]:

model_bert_tiny = malaya.spelling_correction.probability.load(language_model = bert_tiny)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[23]:

model_albert_base = malaya.spelling_correction.probability.load(language_model = albert_base)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[24]:

model_albert_tiny = malaya.spelling_correction.probability.load(language_model = albert_tiny)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[25]:

model_roberta_base = malaya.spelling_correction.probability.load(language_model = roberta_base)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

[11]:

model_gpt2 = malaya.spelling_correction.probability.load(language_model = gpt2)

INFO:malaya_boilerplate.huggingface:downloading frozen huseinzol05/v27-preprocessing/bm_1grams.json

To correct a sentence#

def correct_text(
    self,
    text: str,
    lookback: int = 3,
    lookforward: int = 3,
):
    """
    Correct all the words within a text, returning the corrected text.

    Parameters
    ----------
    text: str
    lookback: int, optional (default=3)
        N words on the left hand side.
        if put -1, will take all words on the left hand side.
        longer left hand side will take longer to compute.
    lookforward: int, optional (default=3)
        N words on the right hand side.
        if put -1, will take all words on the right hand side.
        longer right hand side will take longer to compute.

    Returns
    -------
    result: str
    """

[12]:

strings = [string1, string2, string3, string4, string5, string6, string7]

[13]:

tokenizer = malaya.tokenizer.Tokenizer()

[26]:

for s in strings:
    tokenized = tokenizer.tokenize(s)
    print('original:', s)
    print('corrected:', model_kenlm.correct_text(' '.join(tokenized)))
    print()

original: krajaan patut bagi pencen awal skt kpd warga emas supaya emosi
corrected: kerajaan patut bagi pencen awal sikit kepada warga emas supaya emosi

original: Husein ska mkn aym dkat kampng Jawa
corrected: Husin suka makan ayam dekat kampung Jawa

original: Melayu malas ni narration dia sama je macam men are trash. True to some, false to some.
corrected: Melayu malas ini narration dia sama sahaja macam men are trash . True to some , false to some .

original: Tapi tak pikir ke bahaya perpetuate myths camtu. Nanti kalau ada hiring discrimination despite your good qualifications because of your race tau pulak marah. Your kids will be victims of that too.
corrected: Tapi tak fikir ke bahaya perpetuate myths macam itu . Nanti kalau ada hiring discrimination despite your good qualifications because of your race tahu pula marah . Your kids will be victims of that too .

original: DrM cerita Melayu malas semenjak saya kat University (early 1980s) and now as i am edging towards retirement in 4-5 years time after a career of being an Engineer, Project Manager, General Manager
corrected: DrM cerita Melayu malas semenjak saya kat University ( early 1980s ) and now as i am edging towards retirement in 4 - 5 years time after a career of being an Engineer , Project Manager , General Manager

original: blh bntg dlm kls nlp sy, nnti intch
corrected: boleh bintang dalam kelas nlp saya , nanti intch