Morph Segmentation and Root Identification


In 2027, SIGTYP is hosting a Shared Task on Morph Segmentation and Root Identification Across Typologically Diverse Languages. Morphological segmentation, the process of decomposing words into minimal meaningful units, is a long-standing problem in computational linguistics. At the abstract level, these units are known as morphemes; in surface morphological segmentation, the focus is on their overt realizations, or morphs, in individual word forms. Morphological segmentation underpins applications ranging from lexicography and language documentation to subword tokenisation for language models. Yet most existing resources and benchmarks cover only a small set of well-resourced languages, follow incompatible annotation conventions, and treat words in isolation from the sentences in which they occur.


Overview of the Shared Task

This shared task addresses surface morphological segmentation across typologically diverse languages. Given a word form within a sentence context, participating systems must:
a. Segment the word form into surface morphs, i.e., minimal meaningful units that concatenate to reconstruct the original word form;
b. Identify the root among the resulting morphs.

DataSets and Task Setup

This shared task builds partially on the SIGMORPHON 2022 Shared Task on Morpheme Segmentation and extends it in the following ways:
1. Broader language diversity. The task covers 40 typologically, genealogically, and geographically diverse languages.
2. Harmonised evaluation data. Newly annotated development and test data follow a common sampling method and annotation guidelines, enabling meaningful comparisons across languages.
3. Heterogeneous training data. Training data come from already existing resources and vary considerably in size, quality, lexical material selection, and annotation schema, reflecting the present-day reality of lexicographic resources.
4. Integration with Universal Dependencies (UD). Word forms to be segmented are provided with sentence contexts drawn from UD treebanks as well as with morphological categories following the Universal Features schema.
5. Written and spoken data combined. For most languages, the shared task uses written word forms. For a few languages, however, phonetic transcriptions are used instead, either because no standardized orthography is available or because the existing writing system is not well suited to morphological segmentation.

Important Links

  ↣  Shared Task GIT Page (provides all the details and regular updates)!  

  ↣  Register for the Task!  

Important Dates

  Training data Release: ↣  15 January 2027
  Test data Release: ↣  15 February 2027
  Submissions Due: ↣  1 March 2027
  Results Notification : ↣  15 March 2027
  System descriptions are Due: ↣  22 March 2027
  Camera-ready Due: ↣  29 March 2027

Task Organizers

Magda Ševčíková, Charles University
Zdeněk Žabokrtský, Charles University
Priya Rani, Charles University
Andrew Dryer, Charles University
Hana Hledíková, Charles University
Vojtěch John, Charles University
Anja Nedoluzhko, Charles University
Aleš Papáček, Charles University
Enora Rice, University of Colorado

Contact

    Please contact Priya Rani if you have any questions