Weighted Finite-State Methods for Spell-Checking and Correction
Tommi A. Pirinen · Työväentutkimus Vuosikirja · 2014
This dissertation is a large-scale study of spell-checking and correction using finite-state technology. Finite-state spell-checking is a key method for handling morphologically complex languages in a computationally efficient manner. This dissertation discusses the technological and practical considerations that are required for finite-state spell-checkers to be at the same level as state-of-the-art non-finite-state spell-checkers. Three aspects of spell-checking are considered in the thesis: modelling of correctly written words and word-forms with finite-state language models, applying statistical information to finite-state language models with a specific focus on morphologically complex languages, and modelling misspellings and typing errors using finite-state automata-based error models. The usability of finite-state spell-checkers as a viable alternative to traditional non-finite-state solutions is demonstrated in a large-scale evaluation of spell-checking speed and the quality using languages with morphologically different natures. The selected languages display a full range of typological complexity, from isolating English to polysynthetic Greenlandic with agglutinative Finnish and the Saami languages somewhere in between.