Minimally supervised morphological analysis by multimodal alignment

Abstract

This paper presents a corpus-based algorithm capable of inducing inflectional morphological analyses of both regular and highly irregular forms (such as brought→bring) from distributional patterns in large monolingual text with no direct supervision. The algorithm combines four original alignment models based on relative corpus frequency, contextual similarity, weighted string similarity and incrementally retrained inflectional transduction probabilities. Starting with no paired <inflection,root> examples for training and no prior seeding of legal morphological transformations, accuracy of the induced analyses of 3888 past-tense test cases in English exceeds 99.2% for the set, with currently over 80% accuracy on the most highly irregular forms and 99.7% accuracy on forms exhibiting non-concatenative suffixation.

Unsupervised Learning of Morphology with Graph Sampling

“…Various similarity measures are often used to identify such pairs, including orthographic and context similarity, or a combination of both (Yarowsky and Wicentowski, 2000;Baroni et al, 2002;Kirschenbaum, 2013). The first approach explicitly mentioning Whole Word Morphology as lin- guistic foundation has been presented by (Neuvel and Fulop, 2002), followed by a graph clustering method (Janicki, 2013) and a graph-based probabilistic model (Janicki, 2015).…”

Generating Abbreviations for Chinese Named Entities Using Recurrent Neural Network with Dynamic Dictionary

“…Except these unsupervised methods, there have been other approaches requiring additional information or selective input. Yarowsky and Wicentowski (2000) proposed to use labeled corpus to train a supervised method for transforming pasttense in English. Rogati et al (2003) introduced a stemming model based on statistical machine translation for Arabic.…”

Allomorfessor: Towards Unsupervised Morpheme Analysis

“…We currently use only orthographic features. They are used in a similar manner in [10], but our model needs less supervision and allows concatenative morphology, rather than only stem-suffix pairs. Maybe the closest work to ours is presented in [4].…”

“…We currently ignore allomorphic variation in suffixes. Information sources used in literature are orthographic similarity, word frequencies [10] and similar word contexts [9,1]. We currently use only orthographic features.…”

“…[9,1]), and learning a morphological analyzer (e.g. [10,4]). In our contribution to Morpho Challenge 2008 [7], we present an analyzer that is similar to Morfessor Baseline [3], but in addition finds common base forms for the inflected forms that derive from the same root word.…”