Using Rank-Frequency and Type-Token Statistics to Compare Morphological Typology in the Celtic Languages

ABSTRACT

Previous work has used Greenberg’s synthetism index to compare three of the Celtic languages – Irish, Welsh, and Breton – but not the other three languages, namely Scottish Gaelic, Manx, and Cornish. This paper extends this earlier work by comparing all six Celtic languages, including two periods of Irish (Early Modern and Present Day). The analysis is based on a random sample of 210 parallel psalm texts (30 for each language). However, Greenberg’s synthetism index is problematic because there are no operational standards for counting morphemes within words. We therefore apply a newer typological indicator (B7), which is based solely on lexical rank-frequency statistics. We also explore whether type-token counts alone can provide similar information. The B7 indicator shows that both varieties of Irish, together with Welsh and Cornish, tend more towards synthetism, whereas Manx tends more towards analytism. Breton and Scottish Gaelic do not show a clear tendency in either direction. Rankings using type-token statistics vary considerably and do not tell the same story.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Further information about the Celtic languages can be obtained from the standard overviews by Ball and Fife (1993), MacAulay (1992), and Russell (1995).

2. We do not include all of Tristram’s statistics here; only those that have a bearing on the present experiment.

3. No standard deviations are given for the Welsh statistics from Parina (2006), as she only analysed one text sample per period. Furthermore, two different figures are provided for Welsh, depending on whether word-initial mutations are included in the morpheme counts. Generally, it seems that Tristram does not consider the mutations as morphemic. It is worth noting that Parina (2006) also analysed a text that is closer in date to our own sample – John 1.1–7 from the Welsh Bible of 1588. This had a synthetism index of 1.42 (ignoring mutations). However, this result makes no difference to Tristram’s rankings of languages.

4. The figure quoted by Popescu et al. (2009) for Hawaiian is actually −12.484, but this is clearly a misprint; recalculation from their table of individual text statistics shows that the correct value is −1.2484.

5. In our case, Spearman’s rho for the relationship between TTR and N was 0.409 (p < 0.001) but, for the relationship between TTR and V, rho was just −0.0098 (p = 0.8869).