2.1 The Development of North and Livvi Karelian Varieties

According to the traditional Finnish taxonomy, Karelian has two main dialects, Karelian Proper and Livvi Karelian. The former has two subgroups, North and South Karelian. All varieties are spoken in the Republic of Karelia, Russia. In addition, South Karelian is spoken in Tver Oblast, Russia, as a consequence of migration caused by wars in the 17th century (Karjalainen et al., 2013: 47; Laakso et al., 2016: 96–97; See Fig.1). According to the 2010 census of Russia, 12,369 people defined Karelian as their mother tongue (Laakso et al., 2016: 97). In Finland, Karelian is a non-regional minority language; it is mainly an autochthonous and partly immigrant language. Speakers of all varieties live dispersed across the country. Approximately 5,000 people speak Karelian fluently, and 20,000 understand Karelian in Finland (Laakso et al., 2016: 105, 108).

The geographical distribution of the dialects of the Karelian language.

Citation: Journal of Language Contact 14, 3 (2021) ; 10.1163/19552629-14030006

• Download Figure
• Download figure as PowerPoint slide

View Full Size

The geographical distribution of the dialects of the Karelian language.

Citation: Journal of Language Contact 14, 3 (2021) ; 10.1163/19552629-14030006

• Download Figure
• Download figure as PowerPoint slide

The geographical distribution of the dialects of the Karelian language.

Citation: Journal of Language Contact 14, 3 (2021) ; 10.1163/19552629-14030006

• Download Figure
• Download figure as PowerPoint slide

The development of the Karelian language into its current forms (see Fig. 1) begins with the first contacts between Proto Finnic and East Slavic peoples around Lake Ladoga in current North-Western Russia between 300–700 ad (Kallio, 2006: 157). Throughout its existence, the Karelian language has been located in the borderland of (Novgorodian) Russia and Sweden (to which Finland belonged until 1809), and Karelians have never formed a state of their own (Laakso et al., 2016: 96). Several wars have shifted the location of this border, influencing the development of the Karelian language. However, the easternmost parts of Karelia have always been under Russian rule (Palander, Opas-Hänninen, and Tweedie, 2003: 359–360), resulting in a heavy linguistic influence of Russian on Karelian. Despite this strong influence, the Karelian language and culture have endured.

At the beginning of the 20th century, only 10% of Karelians residing in the Soviet Union knew Russian. This subsequently changed, however, as the Soviet Union’s language policies prohibited the use of Karelian in official contexts. The number of speakers decreased rapidly, and the speakers became, to a greater extent, bilingual with Russian (Karjalainen et al., 2013: 11). As a result, no language planning for Karelian existed in the 1970s and 80s, which strengthened russification. Finally, during the free atmosphere of perestroika and glasnost, the language minorities of Russia started receiving official support, which led to Karelian being recognized as one of the administrative languages (Sarhimaa, 1996: 77–79). However, currently, less than half of ethnic Karelians speak Karelian (Karjalainen et al., 2013: 11).

North and Livvi Karelian are the main varieties used in the Republic of Karelia. They both have a written standard variety that is used, for instance, in the minority media in the area (Tánzcos, 2015). These written standard varieties were developed during the 1990s revitalization of Karelian in the Republic of Karelia. North Karelian is spoken in the region from the White Sea to the present-day Finnish border (see Fig. 1). The speakers of North Karelian are bilingual with Russian. In addition, contact with eastern dialects of Finnish via intensive trading and mobility has taken place from the 16th century onwards. For North Karelians, standard Finnish was a written standard variety during the Soviet period as well as the language of literature and, to some extent, radio and television. This use of Finnish was due to Soviet language policies rather than linguistic reasons (Kunnas, 2007: 58–61; Pasanen, 2006: 122–123). After the Second World War, the status of Finnish weakened in Karelia (Sarhimaa, 1996: 77–79). Due to linguistic similarities between North Karelian and Finnish (Anttikoski, 2003: 29), the languages are mutually intelligible to some extent. The eastern dialects of Finnish, in particular, are closely related to North Karelian.

The differences between the varieties occur especially in phonetics and phonology (Anttikoski, 2003: 33). The same lexemes sound different in each variety, such as the words ruato ~ ruado ‘work’ and konša ~ konzu ‘when’ in the above excerpts. The differences are often between voiced and unvoiced consonants, for instance /d/ ~ /t/, /g/ ~ /k/, /b/ ~ /p/, and /z/ ~ /s/ and in the quality of the sibilants, for instance /s/ ~ /š/ and /z/ ~ /ž/. In addition, differences in the quality of diphthongs exist in the last syllable, for instance kiäntyä ~ kiändiä in the excerpts. The speakers of North and Livvi dialects often perceive these differences as lexical and state that they do not understand all of the words (Kunnas and Arola, 2010: 126–127). Nevertheless, the varieties also have various similarities, for instance, North and Livvi Karelian share a rich suffixal morphology, basic grammar and lexicon (Karjalainen et al., 2013: 50). The written standard varieties reflect the spoken varieties to a great extent even though internal variation in the spoken varieties naturally exist (Section 3.1 briefly discusses the role of Russian in spoken Karelian, especially Livvi Karelian; see also Sarhimaa, 1999; Pyöli, 1996).

2.2 Revitalization and Standardization

Language revitalization began several decades ago as minorities and autochthonous peoples that had been under severe linguistic assimilation policy experienced ethnic awakening. Today, language revitalization has spread to all continents. At the societal level, revitalization of a language means speakers are able to use their language actively in all domains, which includes domains where the use of the language has either disappeared or never existed (Huss, 2001: 278–279). Starting to use a minority language in novel domains also supports language development as it forces language users to develop new vocabulary.

Considering language use, several domains open to minority languages as a written standard variety have developed (Sulkala, 2010: 20). In Karelia, the development of written standard varieties has gone hand-in-hand with the publication of Karelian-language newspapers. For instance, the Oma Mua newspaper continues to play an important role in the revitalization process of Karelian by serving as a linguistic example for the language users; some Karelian dictionaries and glossaries are even based on the vocabulary used in Oma Mua (e.g., Õispuu 2003a, 2003b; Pyöli 2016). In addition, minority language media are important drivers in maintaining, developing, planning, and reviving languages as well as providing information. They signal competence in reporting on modern-day phenomena (Tánczos, 2015: 91–94). Standardization in the planning of a minority language thus makes the language visible (Mantila, 2010: 68) and, for this reason, standardization is often the first step in the revitalization process. The acts of standardization (i.e., development of written language) include creating new terms, renewing, establishing orthography, and choosing between alternatives (Sulkala, 2010: 17).

The greatest challenge to the standardization of Karelian seems to be the geographical variation between its dialects (Anttikoski, 2003: 34–35). The standardization of a language with substantial geographical variation has three different possible options. The first is to choose one dialect as the basis of the written standard variety. The second option is to find compromises between the different dialects. The third option is to accept the use of all the dialects in written form and use several written standard varieties (Mantila, 2010: 59–61). In Karelian, a combination of the first two of these options was briefly experimented with in the 1930s by language-internal means of creating new vocabulary and by copying from Russian using the Cyrillic alphabet. The resulting new standard Karelian was discarded soon after its creation due to political reasons. Linguistic problems also existed; the original morphology and lexicon of Karelian were reduced and modified to more closely resemble Russian. The standard was also not clearly based on any of the Karelian dialects and, thus, it was difficult to learn. (Sarhimaa, 1996: 74–76.)

Language activists’ desire for a single, common written standard variety of Karelian has long existed (Kunnas and Arola, 2010: 126); however, speakers’ experiences of substantial dialectal differences and the difficulty in finding compromises based either on the standard or on the selected forms have prevented the creation of such a standard. However, the differences between the dialects are not as substantial as in Finnish, which has a common written standard variety based on both main dialects. Rigorous language planning can overcome dialectal differences in creating a common written standard variety. Nevertheless, the third option is the current state of the standardization process, with four different standard varieties: Livvi Karelian, North Karelian, Tver Karelian, and South Karelian. The first three have been developed in Russia since the beginning of the 1990s, and the fourth has been developed in Finland in recent years. The first two varieties are the most established, whereas the latter two are in the process of standardization.

3.1 Conscious Development of Vocabularies in Karelian Standard Varieties

At the beginning of the revitalization process of the Karelian language in the 1990s, a special commission for creating neologisms was established. The commission’s task was to invent new words as Karelian lacked many layers of lexicon because the language had never been used in areas such as science, technology, urban life, or politics (Markianova, 2005: 59–60). The commission published new words in special bulletins that were distributed to schools, educational institutions, mass media, and other organizations. Overall, 8,000 socio-political terms, 1,500 terms for the natural world and species, and 700 linguistic terms were introduced in the bulletins by the year 2005. The work of the commission lasted until 2011 (Karjalainen et al., 2013: 51).

The source of neologisms can be original lexemes of different dialects that might have been forgotten or other languages’ lexemes. International words common to most languages that are currently copied from the English model tend to be introduced through Russian. Other languages, such as the close cognate language Finnish, are sources of other new words (for more on the model codes for Karelian, see Markianova, 2005: 59). However, global copying is usually avoided; instead, derivation from the Finnish model is a common strategy. Neologisms independent of foreign models are formed by language-internal means, i.e., derivation and composition (Karjalainen et al., 2013: 51).

Language ideologies may direct the revitalization process of a language, i.e., whether the word formation is done by language-internal means or copied from a foreign model. In the case of Karelian, the ideology of linguistic purism has affected the conscious development of the standard varieties. Purism is a language ideology according to which foreign elements should be avoided at all levels of the language and its development. The ideology is based on the notion that all languages were pure at some point and that this status should be pursued in order to protect languages. Purism is a common ideology in the revitalization of languages as well as in all language development. Moreover, purism can be considered an important tool for maintaining languages, especially in the revitalization of minority languages (Spolsky, 2004: 22; Puura, 2019: 38–39). Language ideologies thus play an important role in the revitalization and standardization of a language.

Finnish as a source language was favoured over Russian at the beginning of the revitalization process. For instance, Õispuu (1997: 92) examined the creation of neologisms during the 1990s and observed that the written standard varieties used Finnish as a source language. According to him, the fact that Finnish is the closest cognate language, and that it has a stable written standard variety and status can be considered as a purist choice as a source of new words. In addition, Õispuu observed that the attitude towards Russian global copies was negative, which resulted in replacing earlier Russian copies with neologisms and Finnish copies in early issues of Oma Mua. For example, in early issues of Oma Mua the names of months were globally copied from Finnish (Õispuu, 1997: 92). However, at present, Karelian original names are in use.

Russian global copies are common and established in all varieties of Karelian, for instance, many of the Karelian particles, such as ta ~ da ‘and’ and hot’ ‘although’, are global copies from Russian and also established in written varieties (Karjalainen et al., 2013: 56; Tavi and Tavi, 2019). In addition, the current speech practices of Russian Karelians contain code alternation, i.e., ‘a switching between two basic codes yielding “mixed discourse”’ (Johanson, 1999: 55), and global copies that are used especially as occasionalisms, for example when discussing work, politics and professions. Karelian can be defined as a high-copying code, i.e., a code that contains a very high amount of copies, usually due to bilingualism (see Johanson, 2002b), as illustrated in Example 1 below in which global copies are shown in bold (Livvi Karelian, taken from Pyöli, 1994: 250).

View Table

The negative attitude towards Russian copies is likely focused on new copies due to an attempt to avoid replicating the common code alternation and global copying of the spoken discourse, in keeping with the purist ideology of standardisation. In standardization, for instance Finnish is a preferred model as it helps to maintain the original elements in the vocabulary. For example, in our data the term yhteistyöšopimuš ‘co-operation agreement’ is globally copied and phonologically adjusted to the North Karelian standard variety. The parts of the compound are yhteis ‘common’, työ ‘work’ and šopimuš ‘agreement’. The first two parts are known in some form in all Karelian varieties, but the third part is known only in North Karelian. In other lexical contexts in Karelian, a global copy from Russian, roato ‘work’, would be used to refer to work instead of the word työ.

3.2 Word Formation and Code-copying in Karelian

Forming neologisms in Karelian standards is a complex, contact-induced process that combines language-internal means of word formation, i.e., derivation and composition (for North Karelian strategies, see Zaikov, 2013: 103–112; and for Livvi Karelian strategies, see Pyöli, 2011: 177–186), and use of a foreign model. The code-copying framework offers sophisticated terminology for contact-induced changes, i.e., different types of borrowing and code-switching, within one framework. The code-copying framework is applied to describe the contact-induced features of neologisms of Karelian because when creating new words it is impossible to separate a switch from a borrowing. In fact, copying means all kinds of contact-induced changes of language, synchronic and diachronic. A copy is never identical to its original because they belong to the recipient language and are subject to its internal developments. Thus, copies are not ‘transferred’ or ‘borrowed’ from another code (Verschik, 2016: 193).

In the code-copying framework, code refers to any variety. The source language is the model code, in this study Russian, Finnish or English, and the receiving language the basic code, in this study Karelian. The term global copy corresponds to loanword and insertional code-switching. Most material, phonetic, semantic, combinational, and frequential features of an item are copied. For instance, the noun ploatja (North Karelian form) ‘dress’ (< Russian plát´e) corresponds to a loanword as is a conventionalized global copy (see Dictionary of Karelian (kks), s.v. ploatja) and pipo ‘knitted hat’ (< Finnish pipo) corresponds to insertional code-switching as it is not conventionalized in Karelian and is used in the data of this study in quotation marks. Both of these instances are defined as global copies according to the code-copying framework. Selective copy, in contrast, is used for a single copied feature of an item, such as meaning. Thus, selective copy replaces, for instance, the terms calque and loan translation (Verschik, 2016: 193–194, 2008: 54; Johanson, 2002a: 291–292). An example of a selective copy (from Pyöli, 1994: 250) in (Livvi) Karelian is the word order in numerals, as shown in Example 2a, where the numeral is followed by the header. The expression of time corresponds to the Russian expression in Example 2b.

View Table

Correspondingly, placing the numeral first in Karelian (Example 3a) corresponds to the Russian meaning of exact time (Example 3b).

View Table

Moreover, in some cases both global and selective copying are employed at the same time, constituting a mixed copy (Verschik, 2016). For instance, the conventionalized Karelian word esiniekka (North Karelian form, see kks) ‘apron’ consists of two morphemes esi ‘front’ + niekka (person-related suffix). The model word is the Russian for ‘apron’, perednik, which also has two parts: pered ‘front’ + nik (person-related suffix). In Karelian, the suffix is a conventionalized global copy and the word for front is original, which makes the Karelian word esiniekka a mixed copy. Within endangered Finno-Ugric minority languages, selective and mixed copying of compounds are particularly common strategies for forming neologisms (Olthuis, 2003: 531–536).

Thus, the code-copying framework is applicable to newspaper data where copied neologisms may not at first be conventionalized in a language, although their appearance often indicates a lexical gap that requires a new word. The examples presented here included old, conventionalized copies, but also neologisms. Hence, the framework is explanatory of both old and new word formation in Karelian. In this study, the term neologism refers to any new word formed by either language-internal means or code-copying, and all analysed code-copies are considered neologisms.

4.1 The Karelian Newspapers

The newspaper Oma Mua ‘Own Land’ was established in Russia in 1990 and was published in the Livvi Karelian and North Karelian written standard varieties until 1999. At the beginning of 2000, the newspaper was divided into two separate papers according to the standard varieties. Consequently, Oma Mua continued publishing articles in Livvi Karelian, and a new newspaper, Vienan Karjala, ‘North Karelia’ published articles in North Karelian (Viinikka-Kallinen, 2010: 194). Both newspapers were published once a week; however, Vienan Karjala was published only twice a month from the second half of 2012 until the end of 2013. The printing company Periodika published both papers, and the newspapers had a joint editorial office in Petrozavodsk, the capital of the Republic of Karelia, Russia. The period of two separate newspapers continued until the end of 2013. From the beginning of 2014, Oma Mua and Vienan Karjala merged, and, once again, Oma Mua began to publish in North Karelian in addition to Livvi Karelian (Viinikka-Kallinen, 2010: 194; Tánczos, 2015: 96).

The first issues of both newspapers had only four pages. During the initial years the papers grew to eight-pages and, after merging in 2014, the issues have since consisted of twelve pages. Both newspapers and the merged volumes focus on Karelians and the Karelian language. They contain different genres, including news reports, prose, poetry, and opinion. In some of the issues children have a dedicated page with facts, stories, and assignments in Karelian. All of these genres include translations from Russian and Finnish. The newspapers have received material from readers and freelance writers, especially during their early years of publication (Tánczos, 2015: 96). Occasionally, especially in the paper’s current merged form, South Karelian and Karelian´s close cognate languages, Ludic and Veps, have been used in the texts. The use of these varieties is sometimes mentioned in the texts, unlike the use of North and Livvi Karelian varieties. The two main varieties are used concurrently without naming them on each occasion. Different articles are written in different varieties with no mixing of the two varieties. Parallel translations of the articles are rare, and are provided only, for example, for official bulletins. In addition, advertisements, for instance of elections, are occasionally published in Russian.

The distribution of both newspapers prior to merging reached 700 copies in total (Tánczos, 2015: 96) in addition to orders for pdf copies. The actual number of readers of the pdf versions is not known.¹ The pdf copies are free online. Backdated issues of Oma Mua are available from 2010 to 2013² and of Vienan Karjala from 2011 to 2013.³ The merged Oma Mua has also been available online since the beginning of 2014, although the latest two years’ editions have been available only to paying subscribers. Availability online was one of the criteria in designing the corpus for the present study.

4.2 Corpus Aggregation and Metrics

The authors collected several Karelian newspaper volumes in order to compile the Karelian newspaper text corpus (knp). The knp consists of several subcorpora, which were used in different stages of the study. One of the main ones is omvk, which encompasses the volumes of Oma Mua (om) and Vienan Karjala (vk) from 2011 to 2013. However, since vk contained fewer words and was released less frequently than om, every other issue from each volume of om was excluded in order to maintain a rough balance between the varieties in the omvk subcorpus. In addition to omvk, the knp includes one volume of Oma Mua from 2017 (om17), which is a merged version of the two former newspapers. Finally, based on the automatic classification (see Introduction), om17 was divided into two subcorpora, om17_livvi and om17_north. The knp and its subcorpora are summarized in Table 1.

View Table

All of the newspapers were converted from pdf to plain text format. Russian-language tv programme names and numerical weather report information were excluded from the conversion. The total number of tokens of the om and vk subcorpora were similar: 414,815 for om, and 412,398 for vk. The om17 subcorpus contained slightly fewer tokens, i.e., 372,693. The size, word types (i.e., types of word forms that occur in the corpus), and the lexical diversity index in the subcorpora are presented in Table 2.

View Table

The lexical diversity index is one of the various metrics for comparing different kinds of text types. It is used in text corpus analysis to define the richness of a lexicon, and it is calculated as a type-token relationship of the words. The index value for om is 0.161, vk 0.175, and om17 0.192, showing the text in om17 to be the richest. It should be noted that these indexes were calculated from unlemmatized word types; in the present study, unlemmatized word types are used because the knp contains only raw text, and, currently, no complete automatic lemmatization tools exist for the Karelian varieties. Nevertheless, some tools for analysing Livvi Karelian are in progress.⁴

As described above, omvk was divided into om17_livvi and om17_north automatically using the Naïve Bayesian classifier.

4.3 Naïve Bayes Classifier

The naïve Bayes classifier is a conventional probabilistic model, which is commonly utilized in text classification studies, such as sentiment analysis of written texts. Even though the naïve Bayes classifier is considered a simple, or ‘naïve’, and non-state-of-the-art machine learning algorithm, previous studies have shown that it achieves good text classification results with low computational costs (e.g., Pang et al., 2002). The term ‘naïve’ indicates that the classifier assumes all features (e.g., words in a text) to be independent of each other. Although this assumption is linguistically incorrect, text classification based on independent word probabilities has been sufficiently powerful with many real data applications (Feng et al., 2015). However, the naïve Bayes classifier has been rarely used for the variety classification of an endangered and under-resourced language.

As the name indicates, the naïve Bayes classifier is based on Bayes’ rule, or Bayes’ theorem, which in this context is defined as in Equation 1.

View Expanded

Equation 1. Bayes’ theorem utilised in language variety classification

Equation 1 shows the probability of language variety (i.e., North Karelian or Livvi Karelian) given features (i.e., word occurrences). In other words, based on the given text, the classifier shows the maximum probability between varieties, which is, in turn, based on the vocabulary of omvk.

The naïve Bayes model for the automatic classification of North and Livvi Karelian was implemented in Python 3 using the nltk.classify.naivebayes module in Natural Language Toolkit’s (Bird, Loper and Klein, 2009) Classify package. The classifier was generated as follows. Firstly, the 2,000 most frequent words in omvk were collected for feature extraction. Then, the omvk subcorpus (189 issues) was divided into training (149 issues) and test (40 issues) data. Instead of comparing the whole omvk subcorpus word-for-word, the feature extractor was used to indicate whether the most common 2,000 unlemmatized word forms occur in the given newspaper and to train the naïve Bayes classifier. Since the extractor searched only for the presence of a word form, the frequency of the word form was excluded from the classification. After training the classifier and calculating the classification accuracy for test data, a list of the most informative words based on the classifier was generated using the most_informative_features function. Finally, the classifier was applied to classify om17 into two subcorpora: om17_livvi and om17_viena. The most informative words in the omvk classification were used to compare the lexical features of the two varieties. The classification results of om17 were used to study new word formation strategies of neologisms (see Section 4.4). The classification results are presented in Sections 5.1 and 5.2.

4.4 Corpus-based Selection and Categorization of Neologisms

Neologisms in om17 were investigated as follows. Based on the division of om17 into om17_livvi and om17_viena, two word lists were generated accordingly. Then, the words in the lists were compared to the following dictionaries and glossaries:

• 1) Pyöli (2016) for Livvi Karelian in Finland

• 2) Fedotova and Bojko (2009) for North Karelian in Russia

• 3) Õispuu (2003a, 2003b) for the neologisms of Oma Mua in the 1990s

• 4) Makarov (1990) for Livvi Karelian in Russia

• 5) kks for Karelian dialects at the turn of the 20th century.

The first and third dictionaries utilize Oma Mua as a source, which facilitated the identification of neologisms. In addition, the lists were compared to the word lists of omvk. Similar words with the dictionaries, glossaries, the word lists of omvk and words with five or more occurrences were then deleted from the two lists, resulting in the final word lists of Karelian neologisms. Finally, the origin of the neologisms was categorized using the following four categories:

• 1) Language-internal means of word formation via derivation and composition

• 2) Words copied from or via Finnish

• 3) Words copied from or via Russian

• 4) Words copied from English.

Each category was analysed from the perspective of the code-copying framework. Section 6 presents the neologisms in om17 and the results regarding their origin.

5.1 Classification Accuracy and the Most Informative Words

The naïve Bayes classifier obtained an overall classification accuracy of 100% on the test set of 40 newspapers from omvk. Rather than demonstrating the high performance of a simple naïve Bayes algorithm, the perfect classification accuracy in this case reveals major differences between the vocabularies of the North Karelian and the Livvi Karelian newspapers. Table 3 shows 10 words that were the most informative for the naïve Bayes classifier. For instance, the word päivy ‘day’ is approximately 60 times more likely to occur in the Livvi Karelian newspaper. The counterpart in North Karelian would be päivä, which is an example of the many lexical differences that are actually phonological in nature (see Section 2.1 for a description). Interestingly, the most informative features in automatic classification were defined as the presence (i.e., the word form received a true value) of Livvi Karelian instead of the presence of North Karelian words. Table 3 also shows ‘false’ values, which indicate that a word is more likely to be absent in comparison to another variety, for instance šitä is approximately 60 times more likely to not occur in Livvi than in North Karelian, i.e., the word is North Karelian. However, the majority of the informative features have ‘true’ values.

View Table

The 2,000 most informative words contain several Russian global copies. The Russian global copies in Table 3 are all conventionalized words, i.e., they were copied at least some decades ago and have become established (e.g., Backus, 2010) in spoken Karelian (e.g., Tavi, 2018) and in several dictionaries (e.g., kks; Pyöli, 2016). Most of them are more likely to appear in Livvi Karelian (see word forms 7 and 9 in Table 3: libo ‘or’, ruado-h ‘work-sg.ill.’), and the absence of some words is characterized as a North Karelian feature in the classifier (see word forms 10–13 in Table 4: ruod-uo ‘work-sg.part.’, libo ‘or’, ruado-h ‘work-sg.ill.’, hos ‘although’). Livvi Karelian is often considered as having more Russian influence than other dialects of Karelian largely for geographical reasons (see Section 2.1) and Table 3 and 4 demonstrate that old, conventionalized Russian global copies are more likely to appear in Livvi Karelian texts than North Karelian texts, which along with phonology create the lexical differences between the varieties (for an example of same text written in North and Livvi Karelian, see Section 2.1). Russian global copies of Livvi Karelian constitute about 4% of the most informative words. Fifteen of the most frequent Russian global copies of the 2,000 most informative words are listed in Table 4.

View Table

Next, the naïve Bayes classifier was used to test whether the vocabulary of the issues of the new merged Karelian newspaper is closer to the North or Livvi Karelian variety. Since the classification accuracy achieved 100% on the omvk test set, the classification of om17 into om17_livvi and om17_north was expected to be reliable.

5.2 Classification of om17

The automatic classification results for om17 show that the majority of the issues contained more linguistic content written in Livvi Karelian than in North Karelian. Of the total number of issues, 80% were classified as Livvi Karelian, whereas only 20% were classified as North Karelian. It should be noted that even though all issues of volume 2017 include linguistic content from both varieties, the classification result indicates the dominance of the North or Livvi variety in a classified issue, i.e., in 80% of the issues, most articles were written in Livvi variety and in 20% of the issues, most articles were written in North Karelian variety. This was confirmed by checking several classification results manually.

Based on the automatic classification, om17_livvi and om17_north were formed. To reveal whether the new word formation strategies follow the older, conventionalized vocabulary or whether new strategies have emerged, neologisms were collected manually from the om17_livvi and the om17_north corpora.

6.1 Language-Internal Means or Selective Copying?

Language-internal means in word formation encompasses derivation and composition (see Section 3.2). Therefore, derivation and composition are summarized as language-internal means in Table 5. Karjalainen et al. (2013: 51) give derivation and lexical calques as examples of original word formation. Lexical calques that are considered selective copies are under the category of original word formation strategies. Many of the neologisms in this category have a model from another language; about 6% of the compounds and derivatives in om17_north and about 5% in om17_livvi are selective copies formed on the basis of the Finnish model. These neologisms were possible to define as selective copies despite the similarities in the word formation of Finnish and Karelian; however, the real percentage of the selective copies may be higher. In some of the cases, the line between language-internal means of word formation and the Finnish model is ambiguous as the two languages are very closely related and share a great deal of lexicon, derivative items, and word formation strategies.

Example 4a shows an instance of a selective copy that has a Finnish derivative as the model. The Finnish model word is the derivative samanlaisuus. Examples of compound words with the meanings of their components selectively copied from Finnish are the North Karelian determinative compound in Example 4b and the Livvi Karelian determinative compound in 4c. Example 4b is formed with Karelian lexical items noja ‘a rest’ and stuula ‘a chair’. Stuula is an old Russian global copy whereas in Finnish the word is tuoli. Similarly, in 4c, laukku ‘a store’ is an old Russian global copy whereas in Finnish the word would be kauppa ‘a store’. Keskus ‘a centre’ is a Finnish global copy in Karelian. The examples of the om17_livvi and om17_north corpora indicate that the compounds have a Finnish model in meaning, which indicates that they are selective copies.

View Table

Selective copying from Russian is rarer, which may be due to slightly different word formation strategies. Russian compounds are open, unlike in Finnish and Karelian. For instance, ‘Minister for Foreign Affairs’ is in Russian minístr inostránnyh del, whereas both Finnish and Karelian use the determinative compound ulkoministeri (‘outer+minister’). In addition, Russian compounds may be appositional with two parts; the second head is semantically dominant and the first is a specifier, or the first head defines the type, and the second noun defines the specifier. The parts are joined with a hyphen, for example, shkóla-internát ‘boarding school’, where the first head is semantically dominant and the second is the specifier (an example of this kind of selective copying is presented in 5a). This type of use of compounds is fixed in the Russian language (Timberlake, 2004: 151).

Russian derivational suffixes have been copied to Karelian via globally copied lexemes without gender marking in the masculine form. For instance, the Russian nik occurs in many conventionalized global copies of Karelian, such as pruasniekka (< prázdnik ‘feast’), and is used to derivate new words, such as pruasniekkaniekka ‘a guest at a feast’. However, neologisms derived using old suffixes copied from Russian are rare in the data as only one example occurs in om17_livvi and om17_north (see Example 5b).

View Table

The compounds can be formed by global copying, or one part of the compound can be an original neologism or a global copy. Thus, the resulting neologism can also be a mixed copy. In om17, one component of the compound and the entire meaning of the compound can be copied in a neologism. The Livvi Karelian neologism sotsialurahasto in Example 6 is a mixed copy (see also Example 4c for a mixed copy). The first semantic component, the modifier sotsiualu, is a copy from the Russian sotsialnyj and has already become a conventionalized word in Karelian. The word has an ending -alu formed based on the Finnish model -aali used in compounds. The second semantic component, the head rahasto, is a global copy from Finnish. The meaning of the endocentric compound has been copied from Finnish.

View Table

No model can be traced in most of the instances of the neologisms created by derivation and composition. Of these two strategies, composition is more common than derivation. Instances of neologisms created by derivation are given in Examples 7a and 7b and by composition in 7c. Examples 7a and 7b are not new concepts, but they are not included in dictionaries and may not be conventionalized. Thus, they can be interpreted as neologisms.

View Table

Of the neologisms formed on the original basis, 3% are derivations and 97% are compounds in om17_north, and 2% are derivations and 98% are compounds in om17_livvi. The strategies for word formation are similar to Finnish selective copies, which may be due to a selective copying of the Finnish word formation model and to similarities with word formation typology. Because of the similarities and close relation between Karelian and Finnish, interpreting whether the neologism is formed by language-internal means or selective copying with Finnish model is elusive.

6.2 Finnish Global Copies

Finnish has served as an example and a source of new words for both North and Livvi Karelian during the revitalization process. In om17_north, Finnish global copies are the second largest group representing 29% of neologism occurrences. In om17_livvi, Finnish global copies also form the second largest group of neologisms with 24% of the total occurrences of neologisms. This result is rather surprising as Karelian neologisms are often reported to be Russian global or selective copies; however, the assumption that Karelian extensively copies from Russian may have its roots in spoken Karelian and its code alternation, as mentioned in Section 3.2. Using the Finnish model is, however, supported by the ideologies of the conscious development of Karelian written standard varieties.

The Finnish global copies resemble words formed by language-internal means; Finnish global copies are Finnish derivatives and compounds. Example 8 gives instances of Finnish derivations that are global copies. The word in Example 8a is derived in Finnish with the suffix -ja, which refers to an actor. In Example 8b, the suffix describing collectiveness is -isO. This word has already been lexicalized in the Finnish language.

View Table

Example 9 contains four instances (9a-9d) of globally copied Finnish compounds. All of these are determinative compounds; the first component is the modifier, and the second is the semantic head.

View Table

Copying international words from Finnish is also a strategy for creating neologisms in om17_north and om17_livvi. According to Section 3.1, the source for international words is expected to be Russian, although favouring Finnish over Russian has been a strategy since the 1990s. Examples 10a-c demonstrate international words copied via Finnish. For instance, in Example 10c, the word in Russian is bum, which would most likely receive the Karelian nominative ending -a, and the resulting word would be bu(u)ma. The nominative ending -i in the word in Example 10c is Finnish and is typical for adapting global copies. The ending reflects that the model language is Finnish. The model code is more ambiguous in Example 10a, where only the case-ending similar to Finnish is a clue to the Finnish model.

View Table

Finnish global copies have preserved their phonological structure in Karelian, which can be seen, for instance, in the nominative case-ending -i instead of the Karelian ending -a. Thus, the type of copying can be defined as global. Contact-induced change is difficult to define in the contacts between close cognate languages (Riionheimo, 2007: 33). The novelty of an item may be the only clue to the change being contact-induced.

6.3 Russian Global Copies

Due to the bilingualism of Karelians in Russia, Russian is a natural choice as a source of new words that are difficult to form by language-internal means. However, Russian is not the first choice in copying, as stated in Section 6.2. Russian global copies form only 9% of the neologisms in om17_north and 10% in om17_livvi. Since Russian has influenced Livvi Karelian more than North Karelian (Section 2.1), the small difference in the percentages of the Russian copies is unexpected. As shown in Tables 3 and 4, Livvi Karelian has more Russian global copies than North Karelian.

Several of the Russian global copies are international, as in Example 11. Consequently, Russian is a source of international words along with Finnish.

View Table

The phonology of international words reveals the model code of the copy. For instance, in Example 11c, the Finnish word is hormoni, and, hence, the /g/ ~ /h/ alternation originates from the Russian model. Additionally, Example 13b has -i- in the ending of the word, but in Finnish, the original word sivusto is more common than the English model. In Russian, sajt is used for site. Examples 11a and 11f come from the Russian nominative inflection in singular and plural forms sessija: sessii, revitalizatsija: revitalizatsii. Another possible explanation for Example 11a is frequency-based copying because the form sessii here represents the nominative singular form in Karelian. In Russian, also the genitive singular form is sessii and this is used frequently also in the accusative case. Similarly, most Russian adjectives globally copied in Karelian have the ending -oi, for instance pervoi ‘the first’ (< Russian pervyj). The ending -oj appears in Russian in the nominative case of only some adjectives of masculine gender but it appears frequently in many other cases in all genders. Thus, frequency of the ending is also a possible explanation for the form of adjectives. The voiced sibilant /z/ in Example 11f originates from the Russian model since the Finnish language has only one sibilant /s/.

In addition to international words, some administrative words have been copied from Russian (Examples 12a and 12c), which is typical considering that these types of words are culture-specific. Russian global copies can also be words that are first copied to the spoken language and then used in the standardization of the language, such as polučči-u ‘(s)he receives’ in Example 12b.

View Table

Even though most neologisms are nouns, some verbs have been copied, as in Example 12b. The verb in the example may have been copied previously to spoken language but is not listed in dictionaries. This may be because Karelian has language-internal means to convey the same meaning, for example, ottoa vastah ‘take in’ (kks s.v. vassassa, vastah). These types of global copies are, however, rare among neologisms in om17_north and om17_livvi.

The ideology of purism in the revitalization process explains the overall results of Russian global copies; they are only the third largest group, and many of the words are of English origin conveyed by way of the Russian model. The words also describe culture-specific referents in Russian. The motivation for using other means than the Russian model to create words may be found in the desire to maintain a language’s originality and distinguish it from Russian. This supports the purist ideology of language revitalization.

6.4 English Global Copies

Copying international words straight from English under the influence of social media and the internet is possible for all languages of the world. However, this strategy has not been considered in previous literature concerning the Karelian language (e.g., Õispuu, 1997; Markianova, 2005; Karjalainen et al., 2013). As stated in Sections 6.2 and 6.3, English words are often copied using the Finnish or Russian model. Nevertheless, some of the global copies with an English origin do not seem copied from a Finnish or Russian model, which can indicate that they originate from the English model. The newspapers in the knp indicate that copying from English is possible, yet rare; English global copies constitute 1% of the neologisms in both om17_north and om17_livvi.

Four instances (a-d) of English global copies are shown in Example 13, with Russian counterparts. The Finnish counterparts are written according to the English form.

View Table

Example 13a is morpho-syntactically adapted, whereas the word in Examples 13b-13d is in the basic form. The word in 13c, wifi, is common to many languages; for example, the word has exactly the same form in Finnish, indicating that Finnish or English is at least the orthographic model for the words instead of Russian. Example 14 describes the syntactic use of the words online and offline.

View Table

The words online and offline should have a different case government in the context. The meaning ‘to write somewhere’ requires the inessive case in Karelian and, therefore, the correct syntax would be kirjutti šitä onlinessa ta offlinessa. Thus, the words have not been morpho-syntactically adapted, which indicates that the words are global copies but not yet conventionalized.

Indications that English may be a new model language for Karelian word formation exist in om17. However, Finnish and Russian remain the models for copying international words. This may be due to the ideologies of revitalization, according to which models should be as close to Karelian as possible. In the future, it is possible that more words will be copied from English as contact between the languages becomes more intense.