Comparison of Russian Morphological Analyzers: MyStem, Pymorphy, TreeTagger, and Stanza

This study compares the characteristics of four Russian morphological analyzers: MyStem, Pymorphy, TreeTagger, and Stanza. Specifically, it examines word segmentation, punctuation handling, tagsets, processing time, and contextual morphological disambiguation. Regarding word segmentation, when spaces were removed from a sample of ten tokens, none of the analyzers—MyStem, Pymorphy, TreeTagger, or Stanza—successfully restored the original word boundaries. MyStem produced no analysis results; Pymorphy and TreeTagger provided the part-of speech and lemma only for the final word in the sequence; and Stanza generated three incorrect tokens and lemmas. These results demonstrate that each analyzer handles ambiguous input differently. In terms of data preprocessing, MyStem and Stanza automatically separate punctuation from word tokens, yielding identical analysis results regardless of whether the input was pre-cleaned; conversely, TreeTagger and Pymorphy produce incorrect output if the data is not pre-cleaned. Each of the four analyzers exhibits distinct characteristics regarding its tagset. Processing time was measured based on the total time required to analyze a text of 99,862 characters. The median processing times were 1.76 seconds for MyStem, 4.47 seconds for Pymorphy, 24.02 seconds for TreeTagger, and 120.21 seconds for Stanza. The four tools also differ in their ability to comprehend context. TreeTagger, which does not utilize an artificial neural network model, correctly disambiguated 9 out of 15 sentences containing homonyms. In contrast, Stanza—which processes the entire sentence—correctly identified 7, while MyStem and Pymorphy failed to identify any correctly. These findings indicate that the ability to utilize context is not necessarily determined by the use of an artificial neural network model. It should also be emphasized that accuracy figures can vary depending on the tool, the method of using the tool, the state of the input data, and the criteria used to interpret the output results.