The CoNLL-SIGMORPHON 2017 shared task on supervised morphological generation required systems to be trained and tested in each of 52 typologically diverse languages. In sub-task 1, submitted systems were asked to predict a specific inflected form of a given lemma. In sub-task 2, systems were given a lemma and some of its specific inflected forms, and asked to complete the inflectional paradigm by predicting all of the remaining inflected forms. Both sub-tasks included high, medium, and low-resource conditions. Sub-task 1 received 24 system submissions, while sub-task 2 received 3 system submissions. Following the success of neural sequence-to-sequence models in the SIGMORPHON 2016 shared task, all but one of the submissions included a neural component. The results show that high performance can be achieved with small training datasets, so long as models have appropriate inductive bias or make use of additional unlabeled data or synthetic data. However, different biasing and data augmentation resulted in non-identical sets of inflected forms being predicted correctly, suggesting that there is room for future improvement.

No abstract

Abstract. Non-canonical inflection (suppletion, deponency, heteroclisis. . . ) is extensively studied in theoretical approaches to morphology. However, these studies often lack practical implementations associated with large-scale lexica. Yet these are precisely the requirements for objective comparative studies on the complexity of morphological descriptions. We show how a model of inflectional morphology which can represent many non-canonical phenomena [67], as well as a formalisation and an implementation thereof can be used to evaluate the complexity of competing morphological descriptions. After illustrating the properties of the model with data about French, Latin, Italian, Persian and Sorani Kurdish verbs and about noun classes from Croatian and Slovak we expose experiments conducted on the complexity of four competing descriptions of French verbal inflection. The complexity is evaluated using the information-theoretic concept of description length. We show that the new concepts introduced in the model by [67] enable reducing the complexity of morphological descriptions w.r.t. both traditional or more recent models.

The question of regularity within morphological paradigms has been formerly addressed within approaches falling in the scope of Canonical Typology (Corbett 2003). The aim of this paper is to provide a means for assessing the notion of morphological canonicity through original measures developed within our new morphological framework PARSLI. In particular, we introduce original measures for non-canonical phenomena such as heteroclisis, deponency, defectiveness and overabundance. We introduce PARSLI a new model for inflectional morphology using an inferential-realisational approach (Matthews 1974; Zwicky 1985; Anderson 1992). Our model precisely provides a formal representation of the lexicon/grammar interface. It relies on a formal definition of a lexical entry and a complete formal apparatus for computing all relevant form realisation rules for each lexeme, including stem formation rules. Realisation rules themselves may be expressed through any suitable realisation-based formalism (e.g. PFM or Network Morphology). We introduce several formal innovations such as inflection zones, that constitute partitions of given inflection classes. They are in particular used in modelling heteroclisis.