Full list of the 115 segmentation labels: number, name, category, and example

This document is a reference catalog that lists all 115 segmentation labels that Bareun's segmentation (Segmenter) model distinguishes in a single table. The principles, such as what a segmentation label is and why it is learned into the model weights, are explained in Korean morphology created by the 115 segmentation rules, so this page focuses on the full list.

Categories at a glance

Full label table

Frequently asked questions

Q. How many segmentation labels does Bareun have in total?

Based on the internal segmentation model, the 114 labels 000 to 113 plus the whitespace handling label SP make 115 in total. Among these, combining the labels with _오류 in their names and the special ㅂ-irregular conjugation forms, the presentation materials introduce 14 error detection labels.

Q. Where can I see the list of the 115 labels?

The full table in this document shows the number, name, category, and example together. The labels are not a separate rule file. They are encoded in the weights of the trained segmentation model (seg-model.onnx), and the numbers in the table may change in part with the model version.

Q. What are the labels marked as error detection?

They are labels that identify morphological errors at the segmentation stage (irregular conjugation errors, contraction and vowel assimilation errors, and so on). While restoring morphemes, Bareun also emits the signal that "this eojeol has the wrong form," and this signal leads to the spell checker.

Q. What does it mean when the example is empty?

The example cell is filled only for labels where a reliable representative case has been confirmed. Even when it is empty, the label name (phenomenon_detail_variant) tells you which Korean morphological change it covers.