During the course of the nineteenth century, ideological language mostly expressed through isms such as liberalism, socialism or conservatism, entered the lexicon in most European languages. Previous research has based on reading key texts claimed that the su x ism was introduced to new linguistic domains during the period up to WWI, many of which do not relate to ideology. This paper uses a data-driven way to study the emergence of isms in nineteenth-century Finnish newspapers and uses word embeddings to cluster them and to trace their thematic expansion in the period. As such, the study provides a quantitatively sound way of tracking how isms relate to ideological language and more generally contributes to the understanding of the development of political language in Finland.
to cluster di erent isms and words close to them in the distributional space over time. In doing so we assume that the distributions allow for clustering isms either according to semantic similarity or similarity in rhetorical tropes or pragmatics.
In this paper we do not use methods based on comparing a word vector from one time slice to a vector for the same word in another time slice, since those methods are aimed at nding radical changes in word senses, such as discussing new sense acquired by words like gay or computer in the twentieth century. Words that we are primarily interested in this study did not undergo such radical transformations|e.g. patriotism meant more or less `love for one's country' throughout the whole nineteenth century|though context and valuation of its usage changed. Instead, we apply clustering of word vectors and demonstrate that word clusters changed as the context of the ism vocabulary was expanded over time.
Clustering isms over a long period of time in a data-driven way poses a number of
methodological problems, which requires testing and exploration. The potential bene t of doing this
lies in producing a statistically robust image of how isms developed. Earlier studies have
argued that isms transformed from the religious sphere, to the political and ideological sphere in
the late eighteenth century and early nineteenth century with pivotal isms such as patriotism,
liberalism and socialism transforming the eld. The eld of isms further expanded in the late
nineteenth century with new isms in philosophy, science and arts appeared [
This paper studies isms as particularly laden keywords in societal discourse in Finland in the long nineteenth century. We address the following research questions:
How did the vocabulary of isms expand in the period? Which isms appear as similar based on their embeddings? Are there interesting continuities in the enriched clustering that takes into account nearest neighbors of the isms? Finally, we shortly discuss the di erences in Finnish-language and Swedish-language discourse in Finland when looked upon through isms. 2.2
We use a digitalized collection of nineteenth-century Finnish newspapers freely available from
the National Library of Finland [
Both corpora are lowercased and lemmatized using LAS, an open-source language-analysis
tool [
To trace semantic shifts in word meanings we split a lemmatized corpus into double decades
(1820{1839, 1840{1859, and so on until 1900{1917) and train continuous embeddings [
Since training word embeddings is a stochastic process, the particular values of vectors do
not stay close across runs, though distances between words are quite stable. To ensure that
embeddings are stable across time slices, we follow the approach proposed in [
To investigate the expansion of the vocabulary of isms we cluster words into semantically close
groups. Since our task is mostly exploratory and the number of clusters cannot be known
in advance we apply the A nity Propagation clustering technique [
A nity Propagation has been previously used for various language analysis tasks,
including collocation clustering into semantically related classes [
The main drawback of the A nity Propagation is pairwise computations. The method is quadratic in time and memory and cannot be applied to large datasets, such as a whole corpus vocabularly. Thus, data selection is an unavoidable step. In this paper we use A nity Propagation in two experiments.
In the rst experiment, we extract from the corpus all ism words. i.e. words that end with -ism in Swedish and -ismi in Finnish and cluster only this set of words. The extraction allows us to identify how close these words to each other given other isms in the corpus.
In the second experiment, we try to put isms into a richer context and trace other words associated with them in the respective double-decades. We extract from the corpus all words which have a cosine similarity to any ism that is less than 0.5. Then we perform clustering on this enriched dataset. Finally, the clusters are ltered so that only clusters that contain at least one ism word are presented for the qualitative analysis. An output of this procedure is di erent comparing to the rst experiment, i.e. words that clustered together in the ism-only clustering can break up into di erent enriched clusters, since in the latter setting they have more exemplar options.
Clustering is performed separately for each time slice. To link clusters across time we perform visualization with Sankey charts. In the Sankey diagram, clusters from time slice t are linked to clusters in time slice t + 1 based on the number of words they have in common.
The magnitude of the link is the sum of the word frequencies (from the source cluster, that is the cluster from time slice t) of the common words between the connected clusters.
We exclude from the list words that are shorter than 5 characters for Swedish and 6 characters for Finnish. This is to lter out obvious OCR bugs such as ism, tism, rism, etc. Though the words `ism' and `ismi' exist in the Swedish and Finnish languages, they are very uncommon in nineteenth-century press.
As expected, Finnish-language and Swedish-language isms cluster di erently in terms of timing and themes that are present. There are three main reasons for this:
Swedish-language press in Finland developed earlier and included more abstract content earlier in the century, whereas newspapers in Finnish|and the Finnish written language| started maturing only in the latter half of the century. Consequently, we have been able to produce meaningful clusters of isms for 1820s onward for Swedish and only from the 1860s onward for Finnish.
The -ismi was not a productive su x in the Finnish language but used through cognate loans and through analogous derivation of foreign words. Consequently, isms are in general less common and ism words less productive in Finnish than in Swedish but nonetheless used especially as Finnish political language in the nineteenth century developed through an interplay between the two main languages in the country.
The political outlook of the two languages was slightly di erent. From the 1880s onward
the Finnish and Swedish newspapers were printed in nearly equal amounts. At this time
the language spheres also started specializing. Swedish speakers lived mostly in larger
towns and around the coast, whereas Finnish speakers occupied the whole country [
Aligning the clusters in the Sankey plots provides a possibility of visually exploring how the vocabulary of isms developed over the course of the century. As can be seen in Figure 1, for Swedish there is quite a steady expansion of isms from the 1820s onward. As the models for producing the clusters rely on enough datapoints for training, particular clusters appear with a delay compared to rst uses of particular words. For instance, patriotism appears the rst time in the corpus in 1791 and liberalism 1820, but the clusters in which they are part of (but not necessarily cluster representatives or most frequent ones) appear in 1820{1839 and 1840{1859, as can be seen in Swedish clusters. The word socialism appears the rst time in 1840 and also appears in the cluster for 1840{1859 respectively, since it immediately became popular and the amount of newspapers in Swedish had already grown.
Figure 1 suggests that there is a clear continuity in the politically laden isms which start from a cluster with patriotism, fanatism (Eng. fanaticism) and despotism in one cluster in 1820{1839 and continue with an expansion over the consecutive double decades. Most frequent isms in the political clusters are patriotism, socialism and despotism up to 1859, and then boulangism, fanatism, anarkism, nationalism and kapitalism (Eng. capitalism) up to 1917. There is some uctuation between the political clusters, like liberalism and patriotism being quite tightly associated until the last time slice of the investigated period, and some unsurprising continuities, like konservatism (Eng. conservatism) and liberalism being in the same clusters through out. Still, it seems that there is less uctuation between the distinctly political clusters and the other clusters. Also the the religious isms (starting from pietism), and medical isms (rheumatism) come across as reasonably stable. The philosophical, artistic and scienti c isms are also distinguishable, albeit they are less clear cut.
For Finnish, the data is too scarce to produce meaningful clusters for more than three time slices Even though the Finnish corpus for the 1880{1899 double decade is comparable in size with the Swedish corpus, the number of distinct isms in Finnish is smaller than in Swedish: 44 for Finnish and 125 for Swedish.
With scarcer data the distinctness of the clusters is even clearer. Clusters with socialism as
the most frequent ism are rather dominant both for Swedish and Finnish, but the role of
socialism as a pivotal ism is even more pronounced for the latter as is also indicated by [
Both Swedish-language and Finnish-language clusters include separate clusters for rheumatism (with spelling variations), which are almost self-containing. Rheumatism, albeit an ism based strictly on spelling, does not cluster with other isms, but has a distinct use in medical discourse of the time. This shows that our clustering method is e ective, but it is also indicative of the fact that historical language use made a distinction of di erent types of isms. Some simply ended with the su x, while others were seen as belonging to groups of other isms. Rheumatism also stands out as a speci c type of term in the newspaper medium as it was very often used as a stand alone word in advertisements or lists of illnesses. 4
There are alternative ways to build diachronic embeddings. The recent line of research is aimed
at smooth time representation [
For examples see Hufvudstadsbladet, 23.11.1907, nro 320, p. 8; Wiborgs Nyheter, 23.01.1903, nro 18, p. 3; Uusi Suometar, 04.06.1905, nro 128, p. 8 representatives are marked with italic, isms are highlighted with bold.
We are grateful to Simon Hengchen and Mark Granroth-Wilding for the help with data preparation. This work has been supported by the European Unions Horizon 2020 research and innovation programme under grants 770299 (NewsEye) and 825153 (EMBEDDIA). [1] Domagoj Alagic, Jan Snajder, and Sebastian Pado. Leveraging lexical substitutes for unsupervised word sense induction. In Thirty-Second AAAI Conference on Arti cial Intelligence, 2018. [2] Cesare Cuttica. To use or not to use ... the intellectual historian and the isms: A survey and a proposal. Etudes Episteme, 23, 2013. [3] Haim Dubossarsky, Simon Hengchen, Nina Tahmasebi, and Dominik Schlechtweg. Time-out: Temporal referencing for robust modeling of lexical semantic change. In The 57th Annual Meeting of the Association for Computational Linguistics (ACL), 2019. [4] Max Engman.
Sprakfragan:
Finlandssvenskhetens uppkomst 1812-1922.
Svenska litteratursallskapet i Finland, 2016.