yajwiz 0.1.1

Klingon NLP toolkit

yajwI’ is a Klingon NLP toolkit that includes basic tokenization, morphological analysis and POS tagging.

It heavily uses the boQwI’ dictionary.

Installation

yajwI’ requires Python 3.8 or newer.

It can be installed from PyPI:

pip install yajwiz

Tokenization

The library includes very simple tokenization.

>>> import yajwiz
>>> yajwiz.tokenize("Hegh neH chav qoH. qanchoHpa' qoH, Hegh qoH.")
[('WORD', 'Hegh'), ('SPACE', ' '), ('WORD', 'neH'), ('SPACE', ' '), ('WORD', 'chav'), ('SPACE', ' '), ('WORD', 'qoH'), ('PUNCT', '.'), ('SPACE', ' '), ('WORD', "qanchoHpa'"), ('SPACE', ' '), ('WORD', 'qoH'), ('PUNCT', ','), ('SPACE', ' '), ('WORD', 'Hegh'), ('SPACE', ' '), ('WORD', 'qoH'), ('PUNCT', '.')]

Morphological analysis

The yajwiz.analyze function parses a word and returns a list of possible parses and a lot of extra information.

>>> yajwiz.analyze("yInwI'")
[
    {
        'WORD': "yInwI'",
        'POS': 'N',
        'XPOS': 'N',
        'PARTS': ['yIn:n', "-wI':n:suff"],
        'LEMMA': 'yIn',
        'SUFFIX': {'N4': "-wI'"},
        'XPOS_GSUFF': 'N',
        'UNGRAMMATICAL': True
    },
    {
        'WORD': "yInwI'",
        'POS': 'V',
        'XPOS': 'VT',
        'PARTS': ['yIn:v:t_c', "-wI':v:suff"],
        'LEMMA': 'yIn',
        'SUFFIX': {'V9': "-wI'"},
        'XPOS_GSUFF': "VT.wI'"
    }
]

Currently the analyzer is very permissive and does allow using wrong plurals and possessive suffixes (eg. yInwI’ instead of yInwIj). It will try to mark this kind of errors with 'UNGRAMMATICAL': True. It detects the following errors:

• Using -pu’, -wI’, -lI’, etc. when the noun is not a person noun

• Using -Du’ when the noun is not a body part

• Using -vIS without using -taH

• Using -lu’ with an illegal verb prefix

• Using intransitive verbs with prefixes indicating object

There is also a simpler function yajwiz.split_to_morphemes, that returns a set of tuples of strings (usually there will be only one tuple in the set):

>>> yajwiz.split_to_morphemes("yInwI'")
{('yIn', "-wI'")}

List of Parts of Speech

XPOS	Explanation
VS	Stative verb
VT	Transitive verb
VI	Intransitive verb
VA	Transitive and intransitive verb
V?	Verb with unknown transitivity
NL	Person noun
NB	Body part noun
PRON	Pronoun (including ‘Iv and nuq: it is a noun that can function as a copula)
NUM	Number
N	Other noun
ADV	Adverb
EXCL	Exclamation
CONJ	Conjunction
QUES	Question word (other than ‘Iv and nuq)
UNK	Unknown

CONLL-U files and POS tagger

CONLL-U files are a popular data format for storing annotated linguistic data.

yajwI’ can generate CONLL-U files filled with morphological information (it does not support dependency parsing).

Below is an example script that first parses a text without a trained POS tagger, then trains a POS tagger with it and finally parses the text with the tagger and saves the result to a CONLL-U file.

import yajwiz

with open("prose-corpus.txt", "r") as f:
    text = f.read()

conllu = yajwiz.text_to_conllu(text)

tagger = yajwiz.Tagger()
tagger.train(yajwiz.conllu_to_tagged_list(conllu))

conllu = yajwiz.text_to_conllu(text, tagger)

with open("prose-corpus.conllu", "w") as f:
    f.write(conllu)

Without a trained POS tagger, ambiguous words will be left without a tag:

# Hegh neH chav qoH.
1   Hegh    _       _       _       _       _       _       _       _
2   neH     _       _       _       _       _       _       _       _
3   chav    _       _       _       _       _       _       _       _
4   qoH     qoH     NOUN    N       _       _       _       _       _
5   .       .       PUNCT   PUNCT   _       _       _       _       _

# qanchoHpa' qoH, Hegh qoH.
1   qanchoHpa'      qan     VERB    V?.pa'  Person=3|ObjPerson=3,0  _       _       _       SuffixV3=-choH|SuffixV9=-pa'
2   qoH     qoH     NOUN    N       _       _       _       _       _
3   ,       ,       PUNCT   PUNCT   _       _       _       _       _
4   Hegh    _       _       _       _       _       _       _       _
5   qoH     qoH     NOUN    N       _       _       _       _       _
6   .       .       PUNCT   PUNCT   _       _       _       _       _

After training the tagger, it will take the “best guess” when deciding the POS.

# Hegh neH chav qoH.
1   Hegh    Hegh    VERB    VT      Person=3|ObjPerson=3,0  _       _       _       _
2   neH     neH     ADV     ADV     _       _       _       _       _
3   chav    chav    VERB    VT      Person=3|ObjPerson=3,0  _       _       _       _
4   qoH     qoH     NOUN    N       _       _       _       _       _
5   .       .       PUNCT   PUNCT   _       _       _       _       _

# qanchoHpa' qoH, Hegh qoH.
1   qanchoHpa'      qan     VERB    V?.pa'  Person=3|ObjPerson=3,0  _       _       _       SuffixV3=-choH|SuffixV9=-pa'
2   qoH     qoH     NOUN    N       _       _       _       _       _
3   ,       ,       PUNCT   PUNCT   _       _       _       _       _
4   Hegh    Hegh    VERB    VT      Person=3|ObjPerson=3,0  _       _       _       _
5   qoH     qoH     NOUN    N       _       _       _       _       _
6   .       .       PUNCT   PUNCT   _       _       _       _       _

In this example the tagger made a mistake: it classified the first Hegh as VT, although it should be N. I don’t have a correctly tagged corpus, so evaluating the tagger is currently impossible. :(

Copyright

yajwiz (c) 2020 Iikka Hauhio

This program a contains the boQwI’ dictionary (data.json) that is licensed under the Apache License 2.0.

The Python files are also licensed under the Apache License 2.0. See the LICENSE file for more details.