Multilingual Clusters and Gender in Nordic Twitter

Steven Coats
English Philology, University of Oulu, Finland
steven.coats@oulu.fi

2nd Digital Humanities in the Nordic Countries Conference
March 14th, 2017

Goals of research

Explore extent of Nordic languages, English, and other languages on Twitter for the Nordic countries

Relatively little quantitative research into bi- or multilingualism and gender on social media (Eleta and Golbeck 2014; Ronen et. al 2014 don’t consider gender)

Explore similarities and differences between the multilingual networks of male and female users

Small but significant differences between genders have been found in (e.g.) relative frequencies of lexical and grammatical types in English (Argamon et al. 2007, Newman et al. 2008, Bamann et al. 2014), also in Twitter data from northern Europe (Coats 2016)

Analogous gender-based differences at a higher level of discourse organization (language choice among bi- and multilinguals)?

Data collection

Tweets collected using Tweepy (Roesslein 2015) from Twitter Streaming API 9 November 2016 – 18 February 2017
Only tweets with populated place attributes, no retweets
310.7m tweets; 1.76m from Nordic countries (Iceland, Norway, Denmark, Sweden, Finland; also Greenland, Faroes, Åland)

Data collection | Filtering

Many Twitter applications send automated messages (location updates by apps such as Foursquare, fitness apps, weather apps, etc.)
tweet_source filtered for apps more likely to be used by human agents (Twitter for iPhone, Instagram, Twitter Web Client, Twitter for Android, Twitter for iPad, Tweetbot for Mac, Twitter for Windows Phone, Twitter for Android Tablets, Twitter for Windows)
1,502,448 Tweets from Nordic countries

Data collection | Language detection

Automatic language detection algorithms can have low accuracy for tweets

Data collection | Improving language detection accuracy

Remove sequences that cause inaccurate detection (usernames, URLs, hashtags)
Remove tweets with fewer than 7 tokens
Remove tweets whose language is reported as “undetected”
Retain only tweets whose language as reported by the native Twitter algorithm agrees with the language as reported by compact language detector 2
1,024,906 tweets by 76,733 unique authors

Tweet density by language

Map data from Natural Earth, maps from Open Street Maps and Carto.

Data collection | Gender disambiguation

Name-based gender disambiguation using aggregate name information (Mislove et al. 2011)
author_name field for each tweet filtered for strings that either begin with or include as a discrete element the most common male and female given names in the Nordics (using regular expressions)
Name information from national statistical offices of Nordic countries
13,506 unique male and 15,497 unique female names used to assign gender to 39,786 Twitter authors
61.5% of Nordic tweets assigned gender in this manner (25% Iceland, 57% Norway, 61% Denmark, 63% Sweden, 70% Finland tweets)

Quantifying bilingualism

A user in the dataset was determined to be bilingual if he or she had authored at least two tweets in two different languages
Overall, 6.28% of unique users in the data qualified as bilingual
For the gendered users the percentages where slightly higher:
Males: 7.903% bilingual, 0.351% trilingual, 0.018% quadrilingual
Females: 8.643% bilingual, 0.446% trilingual, 0.017% quadrilingual

Creating networks of bi- and multilinguals

To quantify the strength of connections between language groups on Twitter in the Nordics, networks of bi- and multilingual users were created

For each language pair \((i,j)\) a contingency table of the number of bilinguals was set up:


	\(language_{i}\)	~\(language_{i}\)
\(language_{j}\)	\(O_{11}\)	\(O_{12}\)	\(= R_{1}\)
~\(language_{j}\)	\(O_{21}\)	\(O_{22}\)	\(= R_{2}\)
	\(= C_{1}\)	\(= C_{2}\)	\(= N\)

From this table a correlation coefficient \(\phi\) can be calculated:

\[ \phi_{ij}= \frac{(O_{11}O_{22}-O_{12}O_{21})}{\sqrt{R_{1}R_{2}C_{1}C_{2}}} \]

\(\phi\) ranges in value from \(-1\) (no bilingual users of \(i\) use \(j\) and all bilingual non-users of \(i\) are non-users of \(j\)) to \(1\) (all bilingual users of \(i\) are users \(j\) and all bilingual users of \(j\) are users of \(i\))

Language connection strength

For language pairs with at least 3 bilinguals, a t-statistic was calculated to assess the significance of the connections:

\[ t_{ij}= \frac{\phi_{ij}\sqrt{D-2}}{\sqrt{1-\phi_{ij}^{2}}}; D = max(R_{1},C_{1}) \]

P-values were calculated from the t-distribution (two-tailed)

Cluster for males

(Random sample of male users = number of female users)

Gender differences in the networks

Female network is slightly denser (21 language nodes, 13 significant connections)
Male network is slightly sparser (19 nodes, 17 significant connections)

Significance of gender differences

To test whether the male-female differences in correlations (\(\phi\)-values) were significant, a z-score was calculated via a Fisher transform of \(\phi\) values according to the formula:

\[ z_{(i,j)} = \frac{z_{\phi_{(i,j)_{m}}}-z_{\phi_{(i,j)_{f}}}}{\sqrt{\frac{1}{n_{(i,j)_{m}}-3}+{\frac{1}{n_{(i,j)_{f}}-3}}}} \]

(Sheskin 2000: 792)

For languages \((i,j)\):

\(z_{\phi_{(i,j)_{m}}}\) is the Fisher transformed value of \(\phi\) for males

\(z_{\phi_{(i,j)_{f}}}\) is the Fisher transformed value of \(\phi\) for females

\(n_{(i,j)_{m}}\) and \(n_{(i,j)_{f}}\) are the number bilingual speakers in the male and female networks.

Significance of gender differences II

Summary and conclusions

English is the central language in all the networks, followed by Swedish
Females are slightly more likely to be bilingual; but few bilinguals use two Nordic languages (negative phi values)
Male phi values are higher for English-Norwegian, English-Danish, English-Arabic and English-Urdu (among others); female phi values are higher for English-Finnish, English-Swedish, and Swedish-Finnish, as well as English-Tagalog

Denmark and Norway have higher rates of use of English on Twitter overall; in the context of language choice, higher female use of features whose frequency is changing (“change from above”) for Sweden and Finland (Labov 1990)?

Acknowledgements

Thanks to

Dorthe Larsen (Danmarks Statistik) for Danish name lists
Joar Skott (Statistika centralbyrån, Sweden) for Swedish name lists
Finland’s Center for Scientific Computing for data storage and computing resources

References

Almende, B.V., and B. Thieurmel. 2016. “visNetwork: Network Visualization using vis.js Library”. R package version 1.0.2. https://CRAN.R-project.org/package=visNetwork

Argamon, S., M. Koppel, J. W. Pennebaker and J. Schler. 2007. Mining the blogosphere: Age, gender, and the varieties of self-expression. First Monday 12/9. http://pear.accc.uic.edu/ojs/index.php/fm/article/view/2003/1878

Bamann, D., J. Eisenstein and T. Schnoebelen. 2014. Gender identity and lexical variation in social media. Journal of Sociolinguistics 18(2), 135–160. http://onlinelibrary.wiley.com/doi/10.1111/josl.12080/full

Cheng, J., B. Karambelkar and Y. Xie. 2017. “leaflet: Create Interactive Web Maps with the JavaScript ‘Leaflet’ Library”. R package version 1.1.0. https://CRAN.R-project.org/package=leaflet

Coats, S. 2016. Grammatical feature frequencies of English on Twitter in Finland. In L. Squires (ed.), English in computer-mediated communication: Variation, representation, and change. Berlin: De Gruyter. 179–210.

Csardi G., and T. Nepusz. 2006. “The igraph software package for complex network research”. InterJournal: Complex Systems. http://igraph.org

Eleta, I. and J. Golbeck. 2014. “Multilingual use of Twitter: Social networks at the language frontier”. Computers in Human Behavior 41, 424–432.

Labov, W. 1990. “The intersection of sex and social class in the course of linguistic change.” Language Variation and Change 2, 205–254.

Newman, M. L., C. J. Groom, L. D. Handelman and J. W. Pennebaker. 2008. “Gender differences in language use: An analysis of 14,000 text samples”. Discourse Processes 45(3), 211–236. http://dx.doi.org/10.1080/01638530802073712

References II

Roesslein, J. 2015. Tweepy. Python package [Computer software]. http://www.tweepy.org

Ronen, S., B. Gonçalves, K. Z. Hu, A. Vespignani, S. Pinker, and C. A. Hidalgo. 2014. “Links that speak: The global language network and its association with global fame”. PNAS 111(52), E5616–E5622. http://dx.doi.org/10.1073/pnas.1410931111

Sheskin, D. 2000. Handbook of parametric and non-parametric statistical procedures, 2nd ed. Boca Raton: Chapman and Hall.

Sites, D. 2013. Compact language detector 2. https://github.com/CLD2Owners/cld2