simplemma 0.6.0

A simple multilingual lemmatizer for Python.

Purpose

Lemmatization is the process of grouping together the inflected forms of a word so they can be analysed as a single item, identified by the word’s lemma, or dictionary form. Unlike stemming, lemmatization outputs word units that are still valid linguistic forms.

In modern natural language processing (NLP), this task is often indirectly tackled by more complex systems encompassing a whole processing pipeline. However, it appears that there is no straightforward way to address lemmatization in Python although this task is useful in information retrieval and natural language processing.

Simplemma provides a simple and multilingual approach to look for base forms or lemmata. It may not be as powerful as full-fledged solutions but it is generic, easy to install and straightforward to use. In particular, it doesn’t need morphosyntactic information and can process a raw series of tokens or even a text with its built-in (simple) tokenizer. By design it should be reasonably fast and work in a large majority of cases, without being perfect.

With its comparatively small footprint it is especially useful when speed and simplicity matter, in low-resource contexts, for educational purposes, or as a baseline system for lemmatization and morphological analysis.

Currently, 38 languages are partly or fully supported (see table below).

Installation

The current library is written in pure Python with no dependencies:

pip install simplemma

• pip3 where applicable

• pip install -U simplemma for updates

Usage

Word-by-word

Simplemma is used by selecting a language of interest and then applying the data on a list of words.

>>> import simplemma
# get a word
myword = 'masks'
# decide which language data to load
>>> langdata = simplemma.load_data('en')
# apply it on a word form
>>> simplemma.lemmatize(myword, langdata)
'mask'
# grab a list of tokens
>>> mytokens = ['Hier', 'sind', 'Vaccines']
>>> langdata = simplemma.load_data('de')
>>> for token in mytokens:
>>>     simplemma.lemmatize(token, langdata)
'hier'
'sein'
'Vaccines'
# list comprehensions can be faster
>>> [simplemma.lemmatize(t, langdata) for t in mytokens]
['hier', 'sein', 'Vaccines']

Chaining several languages can improve coverage:

>>> langdata = simplemma.load_data('de', 'en')
>>> simplemma.lemmatize('Vaccines', langdata)
'vaccine'
>>> langdata = simplemma.load_data('it')
>>> simplemma.lemmatize('spaghettis', langdata)
'spaghettis'
>>> langdata = simplemma.load_data('it', 'fr')
>>> simplemma.lemmatize('spaghettis', langdata)
'spaghetti'
>>> simplemma.lemmatize('spaghetti', langdata)
'spaghetto'

There are cases in which a greedier decomposition and lemmatization algorithm is better. It is deactivated by default:

# same example as before, comes to this result in one step
>>> simplemma.lemmatize('spaghettis', mydata, greedy=True)
'spaghetto'
# a German case
>>> langdata = simplemma.load_data('de')
>>> simplemma.lemmatize('angekündigten', langdata)
'ankündigen' # infinitive verb
>>> simplemma.lemmatize('angekündigten', langdata, greedy=False)
'angekündigt' # past participle

Tokenization

A simple tokenization function is included for convenience:

>>> from simplemma import simple_tokenizer
>>> simple_tokenizer('Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.')
['Lorem', 'ipsum', 'dolor', 'sit', 'amet', ',', 'consectetur', 'adipiscing', 'elit', ',', 'sed', 'do', 'eiusmod', 'tempor', 'incididunt', 'ut', 'labore', 'et', 'dolore', 'magna', 'aliqua', '.']

The function text_lemmatizer() chains tokenization and lemmatization. It can take greedy (affecting lemmatization) and silent (affecting errors and logging) as arguments:

>>> from simplemma import text_lemmatizer
>>> langdata = simplemma.load_data('pt')
>>> text_lemmatizer('Sou o intervalo entre o que desejo ser e os outros me fizeram.', langdata)
# caveat: desejo is also a noun, should be desejar here
['ser', 'o', 'intervalo', 'entre', 'o', 'que', 'desejo', 'ser', 'e', 'o', 'outro', 'me', 'fazer', '.']

Caveats

# don't expect too much though
>>> langdata = simplemma.load_data('it')
# this diminutive form isn't in the model data
>>> simplemma.lemmatize('spaghettini', langdata)
'spaghettini' # should read 'spaghettino'
# the algorithm cannot choose between valid alternatives yet
>>> langdata = simplemma.load_data('es')
>>> simplemma.lemmatize('son', langdata)
'son' # valid common name, but what about the verb form?

As the focus lies on overall coverage, some short frequent words (typically: pronouns and conjunctions) may need post-processing, this generally concerns a few dozens of tokens per language.

Additionally, the current absence of morphosyntactic information is both an advantage in terms of simplicity and an impassable frontier with respect to lemmatization accuracy, e.g. to disambiguate between past participles and adjectives derived from verbs in Germanic and Romance languages. In most cases, simplemma often doesn’t change the input then.

The greedy algorithm rarely produces forms that are not valid. It is designed to work best in the low-frequency range, notably for compound words and neologisms. Aggressive decomposition is only useful as a general approach in the case of morphologically-rich languages. It can also act as a linguistically motivated stemmer.

Bug reports over the issues page are welcome.

Supported languages

The following languages are available using their ISO 639-1 code:

Available languages (2022-04-06)
Code	Language	Words (10³)	Acc.	Comments
bg	Bulgarian	213
ca	Catalan	579
cs	Czech	187	0.88	on UD CS-PDT
cy	Welsh	360
da	Danish	554	0.92	on UD DA-DDT, alternative: lemmy
de	German	682	0.95	on UD DE-GSD, see also German-NLP list
el	Greek	183	0.88	on UD EL-GDT
en	English	136	0.94	on UD EN-GUM, alternative: LemmInflect
es	Spanish	720	0.94	on UD ES-GSD
et	Estonian	133		low coverage
fa	Persian	10		low coverage, potential issues
fi	Finnish	2,106		alternatives: voikko or NLP list
fr	French	217	0.94	on UD FR-GSD
ga	Irish	383
gd	Gaelic	48
gl	Galician	384
gv	Manx	62
hu	Hungarian	458
hy	Armenian	323
id	Indonesian	17	0.91	on UD ID-CSUI
it	Italian	333	0.93	on UD IT-ISDT
ka	Georgian	65
la	Latin	850
lb	Luxembourgish	305
lt	Lithuanian	247
lv	Latvian	168
mk	Macedonian	57
nb	Norwegian (Bokmål)	617
nl	Dutch	254	0.91	on UD-NL-Alpino
pl	Polish	3,733	0.91	on UD-PL-PDB
pt	Portuguese	933	0.92	on UD-PT-GSD
ro	Romanian	311
ru	Russian	607		alternative: pymorphy2
sk	Slovak	846	0.92	on UD SK-SNK
sl	Slovenian	97		low coverage
sv	Swedish	658		alternative: lemmy
tr	Turkish	1,333	0.88	on UD-TR-Boun
uk	Ukrainian	190		alternative: pymorphy2

Low coverage mentions means one would probably be better off with a language-specific library, but simplemma will work to a limited extent. Open-source alternatives for Python are referenced if possible.

The scores are calculated on Universal Dependencies treebanks on single word tokens (including some contractions but not merged prepositions), they describe to what extent simplemma can accurately map tokens to their lemma form. They can be reproduced using the script udscore.py in the tests/ folder.

This library is particularly relevant as regards the lemmatization of less frequent words. Its performance in this case is only incidentally captured by the benchmark above.

Roadmap

• [-] Add further lemmatization lists

• [ ] Grammatical categories as option

• [ ] Function as a meta-package?

• [ ] Integrate optional, more complex models?

Credits

Software under MIT license, for the linguistic information databases see licenses folder.

The surface lookups (non-greedy mode) use lemmatization lists taken from various sources:

• Lemmatization lists by Michal Měchura (Open Database License)

• FreeLing project

• spaCy lookups data

• Wiktionary entries parsed by the Kaikki project

• Wikinflection corpus by Eleni Metheniti (CC BY 4.0 License)

• Unimorph Project

This rule-based approach based on flexion and lemmatizations dictionaries is to this day an approach used in popular libraries such as spacy.

Contributions

Feel free to contribute, notably by filing issues for feedback, bug reports, or links to further lemmatization lists, rules and tests.

You can also contribute to this lemmatization list repository.

Other solutions

See lists: German-NLP and other awesome-NLP lists.

For a more complex and universal approach in Python see universal-lemmatizer.

References

Barbaresi A. (2021). Simplemma: a simple multilingual lemmatizer for Python. Zenodo. http://doi.org/10.5281/zenodo.4673264

This work draws from lexical analysis algorithms used in:

• Barbaresi, A., & Hein, K. (2017). Data-driven identification of German phrasal compounds. In International Conference on Text, Speech, and Dialogue Springer, pp. 192-200.

• Barbaresi, A. (2016). Bootstrapped OCR error detection for a less-resourced language variant. In 13th Conference on Natural Language Processing (KONVENS 2016), pp. 21-26.