This is a rule-based morphological analyzer for Meadow Mari (mhr; Uralic > Mari). It is based on a formalized description of literary Meadow Mari morphology, which also includes a number of dialectal elements, and uses uniparser-morph for parsing. It performs full morphological analysis of Meadow Mari words (lemmatization, POS tagging, grammatical tagging, glossing).
The analyzer is available as a Python package. If you want to analyze Meadow Mari texts in Python, install the module:
pip3 install uniparser-meadow-mari
Import the module and create an instance of MeadowMariAnalyzer class. Set mode='strict' if you are going to process text in standard orthography, or mode='nodiacritics' if you expect some words to lack the diacritics (which often happens in social media). After that, you can either parse tokens or lists of tokens with analyze_words(), or parse a frequency list with analyze_wordlist(). Here is a simple example:
from uniparser_meadow_mari import MeadowMariAnalyzer
a = MeadowMariAnalyzer(mode='strict')
analyses = a.analyze_words('Морфологийыште')
# The parser is initialized before first use, so expect
# some delay here (usually several seconds)
# You will get a list of Wordform objects
# The analysis attributes are stored in its properties
# as string values, e.g.:
for ana in analyses:
print(ana.wf, ana.lemma, ana.gramm, ana.gloss)
# You can also pass lists (even nested lists) and specify
# output format ('xml' or 'json')
# If you pass a list, you will get a list of analyses
# with the same structure
analyses = a.analyze_words([['А'], ['Мый', 'тыйым', 'йӧратем', '.']],
format='xml')
analyses = a.analyze_words(['Морфологийыште', [['А'], ['Мый', 'тыйым', 'йӧратем', '.']]],
format='json')Refer to the uniparser-morph documentation for the full list of options.
Apart from the analyzer, this repository contains a very small set of Constraint Grammar rules that can be used for partial disambiguation of analyzed Meadow Mari texts, as well assigning nonposs tag to all nominal forms without possessive affixes. If you want to use them, set disambiguation=True when calling analyze_words:
analyses = a.analyze_words(['Мый', 'тыйым', 'йӧратем'], disambiguate=True)In order for this to work, you have to install the cg3 executable separately. On Ubuntu/Debian, you can use apt-get:
sudo apt-get install cg3
On Windows, download the binary and add the path to the PATH environment variable. See the documentation for other options.
Note that each time you call analyze_words() with disambiguate=True, the CG grammar is loaded and compiled from scratch, which makes the analysis even slower. If you are analyzing a large text, it would make sense to pass the entire text contents in a single function call rather than do it sentence-by-sentence, for optimal performance.
Alternatively, you can use a preprocessed word list. The wordlists directory contains a list of words from a 2.6-million-word Meadow Mari corpus (wordlist_main.csv), list of analyzed tokens (wordlist_analyzed.txt; each line contains all possible analyses for one word in an XML format), and list of tokens the parser could not analyze (wordlist_unanalyzed.txt). The recall of the analyzer on the standard corpus texts is about 91%.
The description is carried out in the uniparser-morph format and involves a description of the inflection (paradigms.txt), a grammatical dictionary (mhr_lexemes_XXX.txt files), a list of rules that annotate combinations of lexemes and grammatical values with additional Russian translations (lex_rules.txt), and a short list of analyses that should be avoided (bad_analyses.txt). The dictionary contains descriptions of individual lexemes, each of which is accompanied by information about its stem, its part-of-speech tag and some other grammatical/borrowing information, its inflectional type (paradigm), and Russian translation. See more about the format in the uniparser-morph documentation.