An Ontology for Sounds and Sound Patterns

    
David Kamholz
Max Planck Institute For Evolutionary Anthropology, Leipzig (MPI EVA)

     0. Introduction

The description of sounds and sound patterns is an essential aspect of linguistic description. Although there is much theoretical disagreement on these matters, there is considerably more agreement when it comes to the practical demands of description. This paper gives an overview of the concepts required in a general ontology for sounds and sound patterns and the uses it should be adapted to. It is argued that the ontology should essentially be a phonological one, grounded in the categories of the IPA for segments, and also including additional categories for suprasegmentals.

The ontology outlined here is a preliminary one. A sample online implementation for segments has been created and should help to promote discussion; it will be accessible in time for the workshop. The hope is that those familiar with more technical aspects of ontologies will be able to suggest a more sophisticated implementation that may also be more closely integrated with GOLD.1


     1. Motivation

The immediate motivation for this paper is its relevance to a several–year project at MPI EVA, set to begin in earnest this fall, known as the Handbook of Phonological Change (Juliette Blevins, primary investigator). The goal of the project is to collect and synthesize what is known about the phonetic and perceptual causes of regular sound change in a way that will be accessible and useful to as many subfields of linguistics as possible. (The model outlined in Blevins 2004 will be employed as a general framework, but not in such a way as to exclude other views.)

A central element of the project is to create a database of regular sound change. This database will contain data both on known cases of regular sound change as well as synchronic alternations reflecting previous sound change in fossilized form. An important feature of the database will be the ability to search for sound changes based on various phonetic criteria (e.g., instances of stops becoming fricatives, velar palatalization, vowel nasalization, etc.). It thus represents a useful test case for the development and implementation of an ontology of sounds and sound patterns.


     2. Concepts the ontology should cover

Given its intended uses, what belongs in an ontology of sounds and sound patterns? This is not a theoretical question, but rather a question as to how the ontology can best serve general descriptive needs. Perhaps the most central issue in this regard is the respective roles of phonology and phonetics. I would like to make the case that it is most feasible and appropriate to incorporate the former and largely exclude the latter. The arguments for this are twofold. First, most descriptions of sounds and sound patterns employ transcriptions and rule schemas, as well as prose, all of which rely heavily on discrete, phonological concepts. Second, it is precisely such discrete concepts that an ontology is best suited to handle. Phonetic data, on the other hand, is inherently gradient. It is not very clear what a “phonetic category” would be, if not a phonological one, or how this could sensibly be incorporated into an ontology.

A reasonable compromise, I suggest, is to use the IPA as the basis for phonological features — that is, to use the categories implicit in its rows and columns and in its diacritics.2 Such a scheme provides for the description of segments and phonemes in phonological terms while still allowing relatively low–level phonetic detail to be preserved, if desired. The IPA categories are adequate for describing the majority of phonological systems, albeit not always with maximum elegance. But the goal here is generality, not elegance, and in any case, no single ontology can elegantly capture the phonological systems of all languages.

In addition to segments, the ontology should also incorporate a variety of suprasegmental concepts. These fall into three major categories: (1) domains larger than the segment, e.g. syllable, foot, phonological word, phonological phrase, intonational phrase; (2) the constituents of these domains, e.g. stress, tone, syllable onset, nucleus, coda, and rime; and (3) the edges of these domains. The status of the mora is a bit unclear, as it is arguably subsegmental (e.g. a long vowel contains two moras), but it should also be included. Further aspects of the suprasegmental ontology go well beyond my expertise; working group discussions will hopefully expand on it.


     3. Segments

We shall now take a detailed look at the proposed ontology for segments. The following is a preliminary set of features, based in the main on the categories implicit in the IPA:

Figure 1.

Following the nature of the corresponding articulatory phenomena, most features take only a single value, but two (secondary and manner) allow multiple values. Some apply only to consonants (indicated by C in the margin), others only to vowels (indicated by V). It is an open question whether and how underspecification should be allowed—e.g., whether it should be possible to make a distinction between a segment unspecified for labialization versus an explicitly unlabialized segment. Ideally, segments should be maximally specified whenever possible so as to facilitate crosslinguistic comparability, but there may be some cases when this is not desirable.

Another important question is how to support the use of “macro–features” such as sonorant, obstruent, coronal, dorsal, sibilant, liquid, rhotic, etc. The most straightforward solution is to define them in terms of their features, but the problem arises that a given phonetic segment does not always fall into the same class crosslinguistically — for example, ʔ and h sometimes pattern as sonorants, sometimes as obstruents. Still, it should be possible to come up with reasonable “default” values for the macro-features in GOLD, allowing individual COPEs to reinterpret and augment them as necessary.


     4. Transcription

Note also that it was only the use of IPA categories, not symbols, which was proposed above. A transcription system is simply a way of rendering phonetic and phonological concepts graphically; it is only the concepts themselves that belong in GOLD. Furthermore, the set of segmental categories outlined above do not always map cleanly onto the IPA itself: s and ̬ are parallel phonetically, but the former is a single character, the latter a character plus diacritic; the two characters ʧ can represent two sounds, whereas the three characters ‾ʧ are necessary to unambiguously represent a single affricate sound.

What is required to solve these issues is a mapping system that provides a full specification for how segments and suprasegmentals are to be linearized into text, and how strings of text are to be converted into ontological representations. A mapping system would additionally be an efficient way to support multiple transcription systems.3 To demonstrate what this might look like, I have written a simple program for converting strings of IPA segments in Unicode to featural representations. Used in conjunction with an additional tool for Unicode input over the web, it provides a simple way to create “ontologically aware” strings of segments. (A site demonstrating this system will be available online in time for the workshop.)


     5. The ontology in action

There are a number of distinct uses of the ontology that should be kept in mind when creating an implementation. It should be capable of handling at least the following: classes of segments based on features; phoneme inventories; conditioned allophones of phonemes; diachronic sound changes; synchronic alternations; and long-distance rules such as vowel harmony and tone spreading.

Precisely what it means to “handle” these things needs to be discussed, and I do not claim to know the best way to incorporate these capabilities into GOLD. For example, there are many complex issues involved in properly representing sound changes and synchronic alternations. Each contains a set of “from” segments, a set of “to” segments, and an environment. It will not be trivial, to say the least, to develop a set of categories to adequately represent this. In any case, the full demands of linguistic description should always be kept in mind.


1. My ideas on this subject have developed over the past several months and have changed a number of times. Thanks to Jeff Good and Juliette Blevins for guiding my ideas with helpful comments and criticism. They do not necessarily share the views presented here, which are, of course, my own.
2. It may essentially be taken for granted that the ontology must be based on the segment and its description in articulatory terms. Other systems of course exit — autosegmental phonology, Jakobsonian acoustic features, features arising from categorical perception, etc. — but none is as widespread or fully developed. In my opinion, the other systems are best elaborated in COPEs.
3. Incidentally, it also offers a solution to the question raised in Aristar's (2005) proposal.

     References

Aristar, Anthony. 2005. Phonetics ontology, available at http://emeld.org/workshop/2005/ontology-html/phonetics
Blevins, Juliette. 2004. Evolutionary Phonology: The emergence of sound patterns. Cambridge University Press.