Spelling Normalization of Historical Documents by Using a Machine Translation Approach

Spelling Normalization of Historical Documents by Using a Machine Translation Approach Miguel Domingo PRHLT Research Center Universitat Polit`ecnica de Val`encia midobal@prhlt.upv.es Francisco Casacuberta PRHLT Research Center Universitat Polit`ecnica de Val`encia fcn@prhlt.upv.es Abstract The lack of a spelling convention in his- torical documents makes their orthogra- phy to change depending on the author and the time period in which each docu- ment was written. This represents a prob- lem for the preservation of the cultural her- itage, which strives to create a digital text version of a historical document. With the aim of solving this problem, we pro- pose three approaches—based on statistical, neural and character-based machine trans- lation—to adapt the document’s spelling to modern standards. We tested these ap- proaches in different scenarios, obtaining very encouraging results. 1 Introduction With the aim of preserving the cultural heritage, there is an increased need for the digitalization of historical documents, a procedure which strives for creating digital text which can be searched and au- tomatically processed (Piotrowski, 2012). However, the linguistic properties of historical documents cre- ate an additional difficulty. On the one hand, human language evolves with the passage of time. On the other hand, the lack of a spelling convention makes orthography to change depending on the author and the time period in which a given document was written. This makes historical documents harder to read, and makes it even more difficult to search for certain information in a collection of documents, or any other process that must be applied to them. Spelling normalization aims to resolve these problems. Its goal is to adapt the document’s c⃝ 2018 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. spelling to modern standards, increasing docu- ments’ readability and achieving an orthography consistency. Some approaches to spelling normal- ization include creating an interactive tool that in- cludes spell checking techniques to assist the user in detecting spelling variations (Baron and Rayson, 2008). Porta et al. (2013) made use of a weighted finite-state transducer, combined with a modern lexicon, a phonological transcriber and a set of rules. Scherrer and Erjavec (2013) combined a list of historical words, a list of modern words and character-based Statistical Machine Transla- tion (SMT). Bollmann and Søgaard (2016) took a multi-task learning approach using a deep bi-LSTM applied at a character level. Ljubeˇsic et al. (2016) applied a token/segment-level character-based SMT approach to normalize historical and user-created words. Domingo et al. (2017) applied a SMT ap- proach combined with the use of data selection techniques. Finally, Korchagina (2017) made use of rule-based MT, character-based SMT and character- based NMT. In this work, we propose three approaches to tackle spelling normalization: a method based on SMT; another method based on Neural Machine Translation (NMT); and another method based on Character-Based Machine Translation (CBMT). Our main contribution are the followings: • First use (to the best of our knowl- edge) of word-based and subword-based NMT—character-based NMT was already used by Korchagina (2017)—for …

Content truncated to 500 words