Computing colexification statistics for individual languages in CLICS

In the last two weeks we had a renewed interest in colexifications, especially in the third generation of the “Database of Cross-Linguistic Colexifications” (Rzymski, Tresoldi, et al., 2020). The attention was due to two different and independent requests in few days. For those unfamiliar, the concept of “colexification” (François, 2008) refers to instances in which a language uses the same lexeme to express more than one comparable concept (e.g., Russian де́рево, which can mean both “tree” and “wood”). The CLICS project, first developed by List et al. (2014), is an offspring of the transparent approaches to standardization, aggregation, and curation of linguistic data that have been promoted within the CLDF framework (Forkel et al., 2018). It uses standardized lexical databases to identify “colexification networks”.

In the first request, Ezequiel Koile, who was studying geographic linguistic patterns, asked us if it was possible to collect a list of which languages colexify specific pairs of concepts. In addition to this data not being aggregated directly in our software library, despite being indirectly accessible via the web interface, his request involved concept pairs that are not necessarily included in the database. Designed to identify communities of cross-linguistic patterns, the algorithm for CLICS excludes concept pairs with weaker signals, such as those due to pure homophony or to problems in data, as well as those that don’t have enough statistical significance to be part of a community.

A colexification network (from Rzymski, Tresoldi, et al. 2020)

A colexification network (from Rzymski, Tresoldi, et al. 2020)

Days later, another colleague, David Gil, asked if we had ever collected data on the individual colexification frequency per language. He was looking for a continuous typological variable of “tendency for a language to have colexifications”. I knew that this information was available internally, but it was not offered explicitly to the public, nor there was a straightforward way to retrieve it from the existing output or from the web interface.

As we organize the CLICS data in a relational manner and as the algorithm follows a well-structured sequence of steps, using established methods for detecting communities, it was easy to address both requests. The first one was a matter of collecting the languages for each colexification before the community detection. The result is a single table, with concepts listed alphabetically (so we know to search for “TREE and WOOD” and not “WOOD and TREE”) and languages ​​equally ordered. The most common colexifications are confirmed, but the results include a “long tail” of concept pairs with only a single colexification. Information from other resources, such as CLICS itself or Glottolog, can be easily aggregated thanks to the reference catalogs from the CLDF ecosystem.

A LITTLE [2924] SMALL [1246] 1 Hahl Mongolian [halh1238]
A LITTLE [2924] WHAT [1236] 1 Dutch [dutc1256]
ABANDON [1097] THROW AWAY [3696] 2 Ere (China) [eree1240]; Gdongbrgyad-Kamnyu [gdon1234]
[…] […] […] […]
TREE [906] WOOD [1803] 300 Abui [abui1241]; Aché [ache1246]; Adang [adan1251]; Adonara [adon1237]; Adyghe [adyg1241]; … Zacatepec Chatino [zaca1242]; Zaiwa [zaiw1241]; Zuojiang Zhuang [zuoj1238]
[…] […] […] […]
YOUNGER SIBLING [427] YOUNGER SISTER (OF WOMAN) [2421] 13 Biak [biak1248]; Busoa [buso1238]; Dadua [dadu1237];  … Ringgou [ring1244]; Sika [sika1262]
YOUNGER SIBLING [427] YOUNGER SISTER [1761] 7 Central Kanuri [cent2050]; Dutch [dutc1256]; Gurindji [guri1247]; Plateau Malagasy [plat1254]; Swahili [swah1253]; Takia [taki1248]; Yaqui [yaqu1251]
YOUR (PLURAL) [2274] YOUR (SINGULAR) [732] 13 Apali [apal1256]; Asas [asas1240]; Korak [kora1296]; Kulsab [fait1240]; … South Adelbert [sout3148]; Utarmbung [utar1238]

Table 1: Selection of languages and information from the “languages per colexification” results.

The second request involved a little more effort in coding. Not only the amount of data for each language varies (and, even more, varies for the same language in different dataset), but also the mutual coverage when considering all varieties is very small. To discuss which languages have most colexifications implies knowing the potential number of them, and there is no point in counting a concept pair if we data for that pair is not available. Likewise, it is necessary to consider whether we should count over all possible colexifications (as in the case above, more concerned with areal or family tendencies) or just those that are over the thresholds to be included in CLICS. As expected, the ratios of observed colexifications are overall very low, in the order of fractions of percentage points. It was not enough to collect these values, we also needed to offer them in a usable way, properly normalized. Ratios were thus adjusted, offering a convenient number between zero and one for each language, with higher values expressing a greater tendency to colexify.

tng-abaga abag1245 Abaga 0.0656 0.0000
huber-Achagua acha1250 Achagua 0.0995 0.2672
ids-219 basq1248 Basque 0.1144 0.3392
diacl-43200 mode1248 Modern Greek 0.0802 0.2244
tls-Kipogoro pogo1243 Pogolo 0.1148 0.3347
ids-414 zuoj1238 Zuojiang Zhuang 0.1775 0.3610

Table 2: Selection of languages and adjusted ratios from the “language colexification affinity” results.

With these data, it was possible to rely on the geographic coordinates provided by Glottolog and plot a map with the adjusted ratio for each language. As always, a visual exploration suggests patterns and trends that can later be statistically tested. This is the case here, with some areal patterns clearly visible. Polynesian languages ​​seem to be among those most prone to colexification, with high ratio founds also in Southeast Asia and the Amazon (despite some outliers). The ratios for Indo-European languages ​​suggest an intermediate position, and the languages of Papua, regardless of family, appear to be among those with the lowest tendency in the world.

World map of language colexification affinity

World map of language colexification affinity

It’s never superfluous to remember that these data demand the same care of any linguistic research: some characteristics, especially for languages ​​for which we have a single source, may be due to biases in their collection, or even problems in data management. Nonetheless, it is satisfying to watch science at work. First, we have made the efforts of hundreds of linguists over the centuries accessible by good data practices. The results of such practices allowed to publish a resource like CLICS, which other researchers have identified as a viable answer for their questions. In collaboration, we could offer the data they needed, which might nurture the cycle again. But there is a significant difference because of the “data FAIRness” (Wilkinson et al., 2016) approach we have adopted: the sources and the manipulation are transparent and reproducible, so that solutions can be improved and errors can be corrected as a normal part of the research process.

The code for collecting these data has already been merged to pyclics, and can be used from the command-line “clics” tool. The two tables here discussed, along with the vector maps for both adjusted ratios, are available on Zenodo (DOI: 10.5281/zenodo.4148257).


Forkel, R., J.-M. List, S. Greenhill, C. Rzymski, S. Bank, M. Cysouw, H. Hammarström, M. Haspelmath, G. Kaiping, and R. Gray (2018): Cross-Linguistic Data Formats, advancing data sharing and re-use in comparative linguistics. Scientific Data 5.180205. 1-10. DOI:  10.1038/sdata.2018.205.

François, A. (2008): Semantic maps and the typology of colexification: intertwining polysemous networks across languages. In Vanhove, M. (ed.) From polysemy to semantic change. Benjamins: Amsterdam, 163–215.

List, J.-M., Mayer, T., Terhalle, A. & Urban, M. (2014): CLICS database of crosslinguistic colexifications. Forschungszentrum Deutscher Sprachatlas: Marburg, DOI: 10.5281/zenodo.1194088.

Rzymski, C., T. Tresoldi, S. Greenhill, M. Wu, N. Schweikhard, M. Koptjevskaja-Tamm, V. Gast, T. Bodt, A. Hantgan, G. Kaiping, S. Chang, Y. Lai, N. Morozova, H. Arjava, N. Hübler, E. Koile, S. Pepper, M. Proos, B. Epps, I. Blanco, C. Hundt, S. Monakhov, K. Pianykh, S. Ramesh, R. Gray, R. Forkel, and J.-M. List (2020): The Database of Cross-Linguistic Colexifications, reproducible analysis of cross- linguistic polysemies. Scientific Data 7.13. 1-12. DOI: 10.1038/s41597-019-0341-x.

Wilkinson M. D., M. Dumontier, IJ. J. Aalbersberg, G. Appleton, M. Axton, A. Baak, N. Blomberg, J.-W. Boiten, L. B. S. Santos, P. E. Bourne, J. Bouwman, A. J. Brookes, T. Clark, M. Crosas, I. Dillo, O. Dumon, S. Edmunds, C. T. Evelo, R. Finkers, A. Gonzalez-Beltran, A. J. G. Gray, P. Groth, C. Goble, J. S. Grethe, J. Heringa, P. A. C ’t Hoen, R. Hooft, T. Kuhn, R. Kok, J. Kok, S. J. Lusher, M. E. Martone, A. Mons, A. L. Packer, B. Persson, P. Rocca-Serra, M. Roos, R. van Schaik, S.-A. Sansone, E. Schultes, T. Sengstag, T. Slater, G. Strawn, M. A. Swertz, M. Thompson, J. van der Lei, E. van Mulligen, J. Velterop, A. Waagmeester, P. Wittenburg, K. Wolstencroft, J. Zhao and B, Mons -Show (2016): The FAIR Guiding Principle for scientific data management and stewardship. Scientific Data 3.160018. DOI: 10.1038/sdata.2016.18.

Cite this blog post
Tiago Tresoldi (2020, November 4). Computing colexification statistics for individual languages in CLICS. Computer-Assisted Language Comparison in Practice. Retrieved June 22, 2024, from

One thought on “Computing colexification statistics for individual languages in CLICS

  1. David Gil

    I would like to thank the CLICS team for their prompt and very useful response to my query. It is nice to feel part of a community of researchers who are able and willing help each other out to the best of our abilities.

    I would like to bring up a general conceptual issue that has always bothered me about the CLICS project. I have always had the somewhat uncomfortable feeling that “colexifications are in the eye of the beholder”, where the beholder is more often than not a WIERD, or “Standard Average European” lexicographer whose language distinguishes between concepts A and B, and therefore “expects” to find A and B distinguished in other languages, and when they turn out not to be, then “Aha, it’s a colexification”. But if the beholding lexicographers were from some other part of the world, then wouldn’t they have started with a quite different list of concepts, and discovered lots and lots of colexifications in, say, English, that we never think of as being colexifications? For example the word “dog”, which colexifies two different subspecies of dogs the distinction between which might be very important in some languages of New Guinea but nobody ever thought to include in the list of basic concepts. At issue, essentially, is the universality of the CONCEPTICON on which CLICS is based.

    The latest computations and accompanying map, in the above posting, provide a nice method of putting my scepticism to empirical test. If indeed colexifications are in the eye of the beholder, then I would predict that WEIRD or Standard Average European languages should emerge with fewer colexifications than languages from other parts of the world. When I first saw the map, and the really high rates of colexification in languages of the Pacific, I took this as support for my scepticism: surely, they seem to have so many colexifications because they simply care about different things than we do. But then I saw that languages of New Guinea tend to have low rates of colexification, and this is something that I would not have predicted; in fact, this seems to me provide potential support for the universal appropriateness of the CONCEPTICON, which I was calling into question. So maybe my scepticism was a bit misplaced… Bottom line: perhaps an analysis of the world-wide geographical distribution of colexifications, as represented in the above map, could provide a method for a post facto evaluation of the universal validity of the CONCEPTICON list, specifically, the less exceptional WEIRD or Standard Average European languages turn out to be, the more universally appropriate the CONCEPTICON list is.


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.