README.rst

Indonesian Part of Speech Tagger and Tokenizer

Based on tagged text from UI, and and using the frameworks from NLTK [NLTK].

Tokenization

• use the default NLTK tagger with some fixes
• remove clitics if it leaves you with a word

• prefixes se|ku|kau
• suffixes ku|mu|nya and lah|kah

• the wordlist is from the Wordnet Bahasa [WN_BAHASA]

Part-of-speech tagging

A python simple POS tagger for preprocessing text before sense-tagging or parsing.

• it uses NLTK's perceptron tagger trained on the UI corpus [UI_CORPUS] http://bahasa.cs.ui.ac.id/postag/corpus
• the tokenizer and an example of calling the tagger and tokenizer is given in tag.py

Lemmatization

Not done yet! Should work with Hiroki

• deduplicate X-X ->X pos2 if not in WN
• strip superlative

• te+adj -> ter- | adj (if adj in WN)
• ter+adj -> ter- |adj (if adj in WN)

• diX -> di- + X if not in wn

Training and Setup

If you want to train the tagger, use train.py.

Training Data: https://github.com/famrashel/idn-tagged-corpus

Tagset: http://bahasa.cs.ui.ac.id/postag/downloads/Tagset.pdf

http://bahasa.cs.ui.ac.id/postag/downloads/Designing%20an%20Indonesian%20Part%20of%20speech%20Tagset.pdf

We also add a few sentences to give some examples of clitics.

If you want to setup the vocabulary for the tokenizer, run setup.py.

It uses Wordnet Bahasa through the OMW.

ToDo

• Add tagset for NLTK

** at least pull out the short description in EN and ID ** list of tags and most frequent 3 in `tagset.tsv` * Make into a proper module

• Push into NLTK
• Lemmatize with Nomoto-san's tool https://github.com/matbahasa/MALINDO_Morph
• Add extra training sentences for weird punctuation and clitics

``` punct=['•', '>', '/', '?', '!', '-', '–', '—', ';', ':',

'"', '”', '“', ')', '(', ',', '.', '``', "''", "'"]

```

• Add mapping to UPOS

UPOS  Definition      Mapped POS
ADJ   adjective       jj, jj2
ADP   adposition      in
ADV   adverb  prl, rb
AUX   auxiliary verb  md
CONJ  coordinating conjunction        cc
NOUN  noun    nn, nn2, nnc, nnc2, nng, nnu
NUM   numeral cdc, cdi, cdo, cdp, prn
PRON  pronoun prp, wp, wp2
PROPN proper noun     nnp
PRT   particle        neg, rp
PUNCT punctuation     pu!, pu", pu&, pu(, pu), pu,, pu-, pu., pu/, pu:, pu;, pu>, pu?, pu©, pu–, pu“, pu”, pu•
SCONJ subordinating conjunction       sc
VERB  verb    vbi, vbt
X     other   ., dt, dt2, fw, nns2, wrb

Maybe ToDo

These things are useful for tagging, but are done by INDRA. Maybe add add a wsd mode?

After POS tagging, ...

• Split ter/ber/di
• Un-reduplicate

### Note, now use lex['adj'], lex['ber'], ...

notber=['berdiri', 'belaja', 'bersama']
dup = re.compile(r'^(.*)-\1$')

## check for superlative te(r)-
          if lemma.startswith('ter'): # and lemma not in lexall:
              if lemma[2:] in lexadj:
                  lemma = lemma[2:]
                  pos = 'jjs'
              elif lemma[3:].lower() in lexadj:
                  lemma = lemma[3:]
                  pos = 'jjs'
          ## check for 'ber'
          elif lemma.startswith('ber') and lemma not in notber:
              if lemma[2:] in lexvrb:
                  lemma = lemma[2:]
                  pos = 'vbb'
              elif lemma[2:] in lexnon:
                  ### fixme add  'ber'
                  lemma = lemma[2:].lower()
                  pos = 'vnb'
              elif  lemma[3:] in lexvrb:
                  lemma = lemma[3:]
                  pos = 'vbb'
              elif lemma[3:] in lexnon:
                  ### fixme add noun
                  lemma = lemma[3:]
                  pos = 'vnb'
            ## check for passive di-
          elif lemma.startswith('di') and lemma.lower() not in lexall:
              lemma = lemma[2:]
              pos = 'vbd' # di

          ### check for reduplication
          if lemma.lower() not in lexdup:
              d = dup.match(lemma.lower())
              if d:
                  pos = pos + "2"
                  lemma = d.group(1)

Citations

[UI_CORPUS]

Arawinda Dinakaramani, Fam Rashel, Andry Luthfi, and Ruli Manurung. Designing an Indonesian Part of speech Tagset and Manually Tagged Indonesian Corpus. International Conference on Asian Language Processing (IALP 2014).

[NLTK]

Steven Bird, Ewan Klein, and Edward Loper (2018) Natural Language Processing with Python – Analyzing Text with the Natural Language Toolkit (online version)

[WN_BAHASA]

Francis Bond, Lian Tze Lim, Enya Kong Tang and Hammam Riza (2014) The combined Wordnet Bahasa NUSA: Linguistic studies of languages in and around Indonesia 57: pp 83–100 (URI: http://repository.tufs.ac.jp/handle/10108/79286)