A Primer on Automatic Inference of Sound Correspondence Patterns (1): Introduction

After about three years of work on the matter, I have managed (with help of many colleagues who helped in testing) to develop a first approach for the automatic inference of sound correspondence patterns, which will soon be published with Computational Linguistics (List 2019). The key task which this algorithm solves is to take aligned data as input and to compute explicit sound correspondence patterns from the alignments.

Correspondence patterns are hereby understood as recurring alignment sites (i.e., columns per alignment) in a given dataset. In contrast to regular sound correspondences (or systematic sound correspondences, as Trask 2000 calls them), a correspondence pattern is a statement of sound correspondences across multiple languages, while regular sound correspondences are usually discussed for two languages only, at least in automatic (see Kondrak 2002) and formal approaches (see e.g., Hoenigswald 1960).

The result of an automatic correspondence pattern analysis can be thought of as some kind of a table, in which languages are placed in the columns, and correspondence patterns are placed in rows, with each cell indicating for each individual correspondence pattern, which reflex sound a given language shows for this pattern.

As an example, consider the following table, which shows four varieties of the Tableaux Phonétiques des Patois Suisses Romands (Gauchat et al. 1925), which were included as test set for the alignment algorithm presented in my thesis (List 2014).

Pattern Frequency Champéry Lourtier Plagne Courtedoux
*r 21 r r r r
*-(1) 8 ə
*t 8 t t t t
*m 8 m m m m
*f 8 f f f f
*p 7 p p p p
*-(2) 4 ə
*s 4 s ʃ s s
*k 3 k k k

The table provides some ecclectic information (and I could easily provide more), namely some “identifier” for the pattern (which I created in an ad-hoc manner here), the frequency of attested alignment sites, where the pattern occurs, and the concrete reflexes in the four varieties.

While most of the patterns look boring, showing the same reflex in all varieties (which should not be surprising, given that we’re dealing with dialect data here), some show some degree of variation, such as the two patterns which I label as *-(1) and *-(2), respectively, or the pattern *s and *ku. The first two patterns illustrate different degrees of metathesis across the varieties, as they surface in words like “du poivre”, where prototypical reflexes would be Champéry [paːvro] as opposed to Plagne [paːvər] for the first, and Champéry [pɛrdy] as opposed to Courtedoux [prədʒy]. The *s pattern illustrates the pronunciation of original [s] as [ʃ] in initials in Lourtrier, and the third pattern illustrates a specific palatalization process of velars in Courtedoux.

What is interesting and important about these data is how useful they are for additional tasks in historical linguistics. Correspondence patterns can, for example, be used to predict the pronunciation of missing reflexes in a given dataset (as I show in my forthcoming paper, List 2019), they can also be used to reconstruct a given ancestor form semi-automatically, given that all alignment sites which are assigned to the same correspondence pattern directly reflect the same common ancestor sound, and they can be used to investigate conditioning context, given that patterns that look very similar but differ regarding the reflexes of a few varieties often derive from the same proto-sound whose change was then modified in specific environments.

While correspondence patterns can be investigated manually, and people have been doing this in the past, the new computer-assisted methods which we have developed so far greatly facilitate the systematic investigation of correspondence patterns in historical linguistics. The only problem is that — in order to successfully carry out an automatic search for sound correspondence patterns, the data needs to be provided in a very good state, with a very high level of consistency and annotation.

In the following couple of weeks, I want to provide examples for different datasets, illustrating how these can be analyzed with help of the tools for automatic correspondence pattern detection, which have been developed in the past. In this context, I plan to select different datasets and show how they have to be prepared in order to properly analyze them. The posts will be accompanied by supplementary data and code, so that interested users can directly apply and test the examples discussed.

References

Gauchat, Louis and Jeanjaquet, Jules and Tappolet, Ernest (1925): Tableaux phonétiques des patois suisses romands. Relevés comparatifs d’environ 500 mots dans 62 patois-types. Publiés avec introduction, notes, carte et répertoires . Neuchâtel:Attinger.

Hoenigswald, Henry M. (1960): Phonetic similarity in internal reconstruction. Language 36.2. 191-192.

Kondrak, Grzegorz (2002): Determining Recurrent Sound Correspondences by Inducing Translation Models. In: Nineteenth International Conference on Computational Linguistics (COLING 2002). 488-494.

List, Johann-Mattis (2014): Sequence comparison in historical linguistics. Düsseldorf:Düsseldorf University Press.

List, Johann-Mattis (forthcoming): Automatic inference of sound correspondence patterns across multiple languages. Computational Linguistics ?? 1-24.

Trask, Robert L. (2000): The dictionary of historical and comparative linguistics . Edinburgh:Edinburgh University Press.


Author: Johann-Mattis List

Seit Anfang 2017 bin ich leitender Wissenschaftler am Max-Planck-Institut für Menschheitsgeschichte in Jena, in der Abteilung für sprachliche und kulturelle Evolution. In meiner Forschung nehme ich generell einen datenbasierten, empirischen und quantitativen Standpunkt in Bezug auf Sprachwandel und Sprachgeschichte ein, mit einem speziellen Fokus auf südostasiatischen Sprachen. Im Gegensatz zu rein computerbasierten Ansätzen versuche ich jedoch, meine Forschung nah an der traditionellen historischen Linguistik und der linguistischen Theorie auszurichten, weshalb ich einen computer-gestützten Ansatz im Gegensatz zu einem rein computer-basierten Ansatz verfolge.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.