No abstract

The SIGMORPHON 2022 shared task on morpheme segmentation challenged systems to decompose a word into a sequence of morphemes and covered most types of morphology: compounds, derivations, and inflections. Subtask 1, word-level morpheme segmentation, covered 5 million words in 9 languages (Czech, English, Spanish, Hungarian, French, Italian, Russian, Latin, Mongolian) and received 13 system submissions from 7 teams and the best system averaged 97.29% F1 score across all languages, ranging English (93.84%) to Latin (99.38%). Subtask 2, sentence-level morpheme segmentation, covered 18,735 sentences in 3 languages (Czech, English, Mongolian), received 10 system submissions from 3 teams, and the best systems outperformed all three state-of-the-art subword tokenization methods (BPE, ULM, Morfessor2) by 30.71% absolute. To facilitate error analysis and support any type of future studies, we released all system predictions, the evaluation script, and all gold standard datasets. 1

The aim of this paper is to open a discussion on harmonization of existing data resources related to derivational morphology. We present a newly assembled collection of eleven harmonized resources named "Universal Derivations" (clearly being inspired by the success story of the Universal Dependencies initiative in treebanking), as well as the harmonization process that brings the individual resources under a unified annotation scheme.

This article gives an overview of how sentence meaning is represented in eleven deep-syntactic frameworks, ranging from those based on linguistic theories elaborated for decades to rather lightweight NLP-motivated approaches. We outline most important characteristics of each framework and then discuss how particular language phenomena are treated across those frameworks, while trying to shed light on commonalities as well as differences.

The article presents a semi-automatic method for the construction of word-formation networks focusing particularly on derivation. The proposed approach applies a sequential pattern mining technique to construct useful morphological features in an unsupervised manner. The features take the form of regular expressions and later they are used to feed a machine-learned ranking model. The network is constructed by applying the learned model to sort the lists of possible base words and selecting the most probable ones. This approach, besides relatively small training set and a lexicon, does not require any additional language resources such as a list of vowel and consonant alternations, part-of-speech tags etc. The proposed approach is evaluated on lexeme sets of four languages, namely Polish, Spanish, Czech, and French. The conducted experiments demonstrate the ability of the proposed method to construct linguistically adequate word-formation networks from small training sets. Furthermore, the performed feasibility study shows that the method can further benefit from the interaction with a human language expert within the active learning framework.