The Fijian Language GIS Project. NZGRC 2022.

INTRODUCTION: Both cultural geographers and linguists recognize the value of mapping linguistic information. While cultural geographers tend to focus on the spatial differentiation of language as an artefact of the cultural landscape (Jordan, 2014), linguists tend to be more interested in the evolution of language and the complexities of language variation (Haynie, 2014). Linguistic geography, or geolinguistics, is an interdisciplinary field that recognizes the importance of spatial patterns and geographic relationships as drivers of change leading to the structural diversity of language (Haynie, 2014). With roots in traditional core ideas of cultural geography, such as diffusion theory (e.g. wave model (Hagerstrand, 1966) and gravity model (Olsson, 1965)), the field of geolinguistics has benefited in recent decades from advances in computational analysis (e.g. data mining) and geographic information systems (GIS).
The Fijian Language GIS Project is an interdisciplinary research effort funded by The Resona Foundation for Asia and Oceania and led by Associate Professor Ritsuko Kikusawa of the National Museum of Ethnology, Osaka, Japan. The goals of the project are to: 1) develop a GIS database of Fijian communalects collected and recorded by Dr Paul Geraghty of The University of the South Pacific, 2) use the database to conduct research on linguistic variations from a spatial perspective focusing on horizontal and vertical evolution of language, and 3) produce information suitable for dissemination to the public through museums and other venues.
METHODS (GIS Database of Fijian Communalects):
A “communalect” is the smallest subdivision of a dialect and refers to community of native-born speakers who share a common variety of speech (Pawley and Sayaba, 1971). Differences among communalects are subtle and most recognisable by native speakers as varieties of Fijian speech that indicate a person’s home locality (Geraghty, 2018, Pawley and Sayaba, 1971). The geographic area of a communalect is small, often comprising a single village, a group of villages, or a small island (Pawley and Sayaba, 1971).
Over the past 40 years Geraghty has recorded 100-word lists for roughly 260 communalects of Fiji. The Geraghty 100-word list is similar in concept to the Swadesh list used by linguists for historical and comparative language study (Swadish, 1971). The list contains 100 lexical items chosen by Geraghty as a representative sample useful for comparing variation among communalects. In addition to a 100-word list for each of 300+ Fijian communalects, Geraghty has compiled a 100-word list for the Bauan dialect which has emerged since the 1840s, as the standard literary language and lingua franca of Fijians (Geraghty, 1983, Pawley and Sayaba, 1971).
To build the GIS database of communalects it was necessary to first reference communalect names to geographic locations. This was done by assigning communalect names to a GIS dataset of villages and also by assigning communalect names to a GIS dataset of land areas known as mataqali (landowning units associated with Fijian clans (Crocombe, 1987)) . To complete the GIS database, a table of 100-word lists for each of the communalects was linked to each communalect geographic feature. In total there are approximately 1100 villages in the database, each with an assigned communalect name. While there are approximately 270 communalects in total, roughly 150 communalects have 100-word lists.
(Kadavu Pilot Study): As a pilot study we used data from the island group of Kadavu to evaluate the potential for using the GIS database for a dialectometric analysis of communalect variation. Kadavu is a relatively isolated and sparsely populated island 80 kilometres south of the main island of Viti Levu. Travel to and from Kadavu passes through two points of entry, one at an airport at Vunisea and the other at a ferry terminal at Kavala. Automobile travel in Kadavu is limited and many people travel between villages via fiberglass boats with outboard motor. All villages are situated on the coast.
We explored the hypothesis that lexical items from the 100-word list for communities near the points of entry would be more similar to standard Fijian (Bauan dialect) than communalects further away from these points of entry. We reasoned that communities near the points of entry would be more likely to adopt artefacts of speech as a result of more frequent encounters with people moving to and from the main island. There are 71 villages, comprising 17 communalects in the Kadavu island group.
RESULTS: A metric of linguistic distance between the lexical items for each of the 13 communalects’ 100-word lists and the 100-word list for standard Fijian (Bauan dialect) was derived a sequence comparison algorithm that examines the number and alignment of phonetic segments of lexical items to produce a metric of linguistic distance for each lexical item (List et al., 2008). As a result, a table was created of linguistic distances for each lexical item for each communalect, which was linked to villages in the GIS database. To visualise linguistic distance Inverse Distance Weighting (IDW) was used to create a continuous surface of linguistic distance for each lexical item.
Visual examination of four lexical times (nikua, lasu, caka, and levu) allows us to evaluate the hypothesis that lexical items for communities near the ports of entry have greater similarity to standard Fijian than communalects more distant from the entry points. What we see is at best inconclusive, and more likely suggests the hypothesis is not supported by the data. We see, for example, that data for the lexical item nikua supports the hypothesis in Kavala, but not Vunisea. Likewise, the lexical item levu supports the hypothesis in Vunisea but not Kavala. If the data for individual lexical items are inconclusive, what if the data were visualised in aggregate? A surface based on the average linguistic distance for a selection of 10 lexical items can be visually interpreted to suggest that on average the communalects near Kavala are less similar to standard Fijian (i.e., greater linguistic distance) while the communalects near Vunisea tend to be neither similar nor dissimilar to standard Fijian. A visualization of the standard deviation of the mean for the same 10 lexical items can be interpreted to suggest that communalects at the furthest ends of the island vary little in relation to standard Fijian and in the middle of the island the variation is greater.
DISCUSSION AND CONCLUSION: The pilot study described above is a first attempt to analyse the GIS data collected as part of the Fijian Language GIS Project. An important aim of many geolinguistic studies is to model and visualise the hypothesised relationship between geographic proximity and linguistic distance, a branch of geolinguistics referred to as dialectrometry (Haynie, 2014). A study by Szmrecsanyi (2011) in the United Kingdom analysed the frequency of grammatical characteristics in 34 dialects to map their linguistic distance in relation to their geographic distance. The matrix of linguistic distances was reduced to three dimensions using Multidimensional Scaling which were mapped in a RGB (red-green-blue) colour scheme to visually compare linguistic distance to geographic distance. The possibility of using similar methods for visualising the relationship of linguistic distance and geographic distance in Kadavu and other parts of Fiji is one avenue of analysis that we look forward to exploring.