Browse Prior Art Database

Analytical Method for Measuring Phonological Similarity

IP.com Disclosure Number: IPCOM000124577D
Original Publication Date: 2005-Apr-28
Included in the Prior Art Database: 2005-Apr-28
Document File: 5 page(s) / 205K

Publishing Venue

Motorola

Related People

Chen Liu: AUTHOR [+2]

Abstract

The proposed invention presents a knowledge-based method to measure phonological similarity between languages by comparing phoneme inventories along the parameters of general phoneme correspondence and inventory symmetry. Overall language relatedness is also considered.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 16% of the total text.

Analytical Method for Measuring Phonological Similarity

 

Chen Liu

, Lynette Melnar

Abstract

The proposed invention presents a knowledge-based method to measure phonological similarity between languages by comparing phoneme inventories along the parameters of general phoneme correspondence and inventory symmetry.  Overall language relatedness is also considered.

Introduction

Phonemes are language-dependent speech segments that may have multiple phonetic realizations with distinct feature specifications.  For this reason, it is impossible to reliably map phonemes across languages based only on their transcription labels or their distinctive feature matrices where subphonemic variation is not specified (as in Schultz and Waibel 2000) .  Where speech data is available, acoustic measurements may be used to determine similarity among acoustic speech models (phone HMMs).  However, speech data unavailability is a practical concern in speech technology.  The ability to make a linguistically principled, knowledge-based comparison is one way to predict subphonemic variance and overall phonological similarity between languages in the absence of speech data.  This allows for greater leverage of language resources used in speech technology applications.  For example, knowledge-based phonological similarity identification can be used to predict best candidates for phoneme model sharing or substitution in automatic speech recognition and text-to-speech synthesis systems.

Phonological Similarity

The notion of phoneme similarity is fundamental in phonological similarity research.  Current approaches, largely from studies in language reconstruction, dialectometry, and language development, generally measure phoneme similarity between aligned phonemes in pre-defined cognates (see, for example, Heggarty 2000, Kondrak 2003, Connolly 1997, Gildea and Jurafsky 1996, Kessler 1995).  Phonological similarity is then measured as the overall similarity between the cognates (genetically related words) and, by extension the cognate set.  Significantly, these approaches are all dependent on the identification of cognates and the proper alignment of their pronunciation strings, both of which require expert knowledge that is not generally available.  In addition, because cognates are genetically related words, cognate identification is restricted to related language systems.  Comparing non-related phonologies is not possible in this general method.  Comparing phoneme inventories directly is a way to predict phonological similarity that is a significant departure from this general approach.  Crucially, phoneme inventory comparison does not depend on cognate identification and alignment and is not restricted to related languages.  Instead, the method requires the availability of phoneme inventories and is universally applicable.

Measuring Phonological Similarity

The present invention is a method for measuring phonological similarity by ranking phoneme inventorie...