Evaluating Inflectional Complexity Crosslinguistically: a Processing Perspective

Evaluating Inflectional Complexity Crosslinguistically: a Processing Perspective Claudia Marzi1, Marcello Ferro1, Ouafae Nahli1, Patrizia Belik2, Stavros Bompolas3, Vito Pirrelli1 1Institute for Computational Linguistics-CNR, 2Universitat Politècnica de Valencia, 3University of Patras patrizia_belik@yahoo.com, stavros.bompolas@gmail.com {claudia.marzi, marcello.ferro, ouafae.nahli, vito.pirrelli}@ilc.cnr.it Abstract The paper provides a cognitively motivated method for evaluating the inflectional complexity of a language, based on a sample of "raw" inflected word forms processed and learned by a recurrent self-organising neural network with fixed parameter setting. Training items contain no information about either morphological content or structure. This makes the proposed method independent of both meta-linguistic issues (e.g. format and expressive power of descriptive rules, manual or automated segmentation of input forms, number of inflectional classes etc.) and language-specific typological aspects (e.g. word-based, stem-based or template-based morphology). Results are illustrated by contrasting Arabic, English, German, Greek, Italian and Spanish. Keywords: paradigm-based morphology, inflectional complexity, prediction-based processing, recurrent self-organising networks. 1. Introduction There is little doubt that some languages are inflectionally more complex than others. Everybody would agree with the intuitive statement that the English conjugation system is simpler than the German system, and that the latter is, in turn, simpler than the verb system of Modern Standard Arabic. However, the naïve view is faced with two apparent paradoxes. When linguists try to pinpoint the source of this complexity, the task is far more elusive than expected, and goes well beyond a purely descriptive notion of diversity in the battery of realisational means (e.g. number of differ- ent affixes, number of cells in the corresponding paradigms, amount of stem allormophy etc.) provided by each system. Besides, there seems to be a poor correlation between our intuitive notion of morphological complexity and actual ev- idence of the pace of acquisition of more or less complex inflectional systems in child language. In some cases, ap- parently simpler inflectional markers may take more time to be acquired than formally more complex and articulated ones. What looks like a prohibitively difficult learning task in the light of the complexity and uncertainty of the infer- ence steps required for mastering it, may turn out to be rel- atively unproblematic for human speakers. In the present paper we entertain a usage-oriented, cognitively motivated approach to issues of morphological complexity, based on a neurobiologically inspired model of word processing and learning, and explore its theoretical and computational im- plications. 2. Background Assessing and understanding the comparative complex- ity of the inflectional system of a language relative to a functionally-equivalent system of another language re- mains an open question, which has animated much of the contemporary debate on the nature of word knowledge and its connection with issues of word usage (Ackerman and Malouf, 2013; Bane, 2008; Bearman et al., 2015; Juola, 1998; Moscoso del Prado Martín et al., 2004). In a crosslinguistic perspective, the way morphosyntac- tic features are contextually realised through processes of word inflection probably represents the widest dimension of grammatical crosslinguistic variation, somewhat belit- tling universal invariances along other dimensions (Evans and Levinson, 2009). Descriptive linguists have …

Content truncated to 500 words