A Universal System for Automatic Text-to-Phonetics Conversion

Proceedings of Recent Advances in Natural Language Processing, pages 360–366, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_042 360 Abstract This paper describes an automatic text-to- phonetics conversion system. The system was constructed to primarily serve as a research tool. It is implemented in a general-purpose linguistic software, which allows it to be incorporated in a multifaceted linguistic research in essentially any language. The system currently relies on two mechanisms to generate phonetic transcriptions from texts: (i) importing ready-made phonetic word forms from external dictionaries, and (ii) automatic generation of phonetic word forms based on a set of deterministic linguistic rules. The current paper describes the proposed system and its potential application to linguistic research. 1 Introduction There are currently many commercial and academic works dealing with automatic conversion of text to phonetics (G2P). Existing solutions are based on some combination of language-specific phonetic dictionaries, corpus-based statistical models (e.g., Hidden Markov Models), deterministic models based on linguistic rules, and machine learning techniques (see review in Tomer, 2012). Importantly, most available tools are closed- source, support only a limited number of languages, and are often available only for commercial use (e.g., Baytukalov, 2019). The current paper describes a new, open-source system that generates phonetic transcriptions from texts. Its design allows it to apply, in principle, to any language. At present, the system relies on two mechanisms to generate the transcriptions: (i) 1 Website: https://chengafni.wordpress.com/cpa/ 2 The code is written in Visual Basic for Applications for MS Excel and is available under the GNU General Public License. A version of this system for LibreOffice Calc is importing ready-made phonetic word forms from external dictionaries, and (ii) automatic generation of phonetic word forms based on a set of deterministic linguistic rules. The following sections describe the transcription mechanisms and additional relevant tools. 2 Preliminaries The described system is implemented in the Child Phonology Analyzer software (CPA; Gafni, 2015)1, which was built in MS Excel due to its popularity and user-friendly interface.2 Nevertheless, the concepts behind the system are general and can be implemented in various environments. The following subsections describe the general organization of the system and guidelines for working with the data. 2.1 Tables The system uses a set of tables of definitions and rules to guide its operation. The tables are stored in separate spreadsheets in the CPA file and can be edited according to the properties of the language in question. Moreover, variants of these tables can be stored in separate files and imported by the system when needed. This feature allows users to maintain sets of definitions and rules for multiple languages.3 2.2 Organizing the Data The transcription procedures (“macros”) operate on a vector of words. Thus, the input text should be converted into a vector format prior to running the transcription macros. This can be done using the “Corpus tokenization” option of CPA’s “Data planned to appear in the future in order to free it from dependency on proprietary software. 3 The system is accompanied by an external resource containing …

Content truncated to 500 words