1. Introduction

Semantic clustering is at present a robust natural language processing method (or rather a group of methods) employed in investigating semantic relationship through the shared meanings the words maintain in context (Kuhn et al., 2007; Witschard et al., 2022). It allows to cluster words and contexts through word embeddings (Word2Vec) or through transformer-based models (BERT) by calculating their similarity in the vector space and summarizing the data.

Meanwhile, we presume that apart from identifying the similarity between texts, semantic clustering integrated into cognitive framework can contribute to exploring more complicated research areas including discourse or even culture similarity. Cognitive framework allows model the relations of cognitive processes (Talmy, 2000; Langacker, 2007, Verhagen, 2007, Gallagher, Hutto, 2008); however, to proceed it requires collecting extensive data on how these processes are performed through experimental or corpus research. Since semantic clustering reveals the relationship of words, it can be attributed to knowledge domains and qualia roles – formal, constitutive, telic, and agentic (Pustejovsky, 1991, 2006) specifying discourse types and culture types. The current paper is the first attempt in exploring the cognitive process of knowledge transfer in culture integration through semantic clustering, which allows to identify the thematic attribution of knowledge domains and the qualia roles subjected to adapting the knowledge developed within a donor culture and transferred onto recipient culture. The found solution lies within distinguishing two types of knowledge transfer in culture integration, direct and indirect, exposed in immediate and mediated transfer of knowledge domains from donor into recipient culture.

Therefore, we propose a two-step research procedure which involves: 1) the corpus analysis allowing to identify the thematic attribution of knowledge domains which are directly transferred from donor into recipient culture and the degree of direct culture integration; 2) lexical-semantic clustering allowing to specify the qualia roles in indirect knowledge transfer since it explores the contexts mediated by the use of these domains. The procedure is verified featuring the lexical criterion applied to the loanwords contributing to culture integration with Russian language donor culture and Kyrgyz language recipient culture. The research objectives are consequently to reveal the thematic attribution of knowledge domains which were transferred from Russian into Kyrgyz culture directly through the integrated loanwords in web discourse, and to specify the qualia roles of the knowledge transferred indirectly through the loanword collocates. To proceed, a list of loanwords in the Kyrgyz language with Russian as etymon language was compiled, which was further applied to explore the direct and indirect knowledge transfer in culture integration.

The Theoretical Framework section provides an overview of the prerequisites for addressing cultural integration through the lexical criterion. Section Three presents the research methodology and two-step procedure of corpus analysis and lexical-semantic clustering for exploring direct and indirect transfer. Section Four demonstrates the results determining: 1) the knowledge domains representing the loanwords and the distribution of loanwords in the compiled web database as well as their thematic attribution, exposing the results of direct knowledge transfer, 2) the distribution of target words displaying high vector similarity with the loanwords in the compiled web database obtained through semantic clustering, and qualia roles representing the referents of these loanwords in the web contexts, exposing the results of indirect transfer. Section Four discusses the results in the context of semantic clustering framework and cognitive frameworks in attribution to culture integration. Section Five lists the study limitation. The Final Remarks section specifies the novel applications of lexical-semantic clustering in exploring culture integration and disintegration processes.

2. Theoretical Framework

2.1. Lexical-semantic clustering in exploring discourse and culture

Semantic clustering is generally referred to the practice of grouping elements into categories based on their meanings (Tinkham, 1993). So, the term “semantic cluster” is usually understood as a set of elements (words, concepts, etc.) that can be qualitatively or quantitatively analyzed. Guided by this general understanding of semantic clustering, various scientific theories develop specific methods for identifying and analyzing lexical-semantic clusters. For instance, cognitive network science (Siew et al., 2019) posits that lexical-semantic clusters can be represented as network structures – specifically, as sets of three nodes (concepts) interconnected by edges (semantic relations). Analyzing the overall clustering of a network reveals the efficiency of its structure for searching and retrieving information.

Troyer et al. (1997) defined a “semantic cluster” as a group of items from the same lexical-semantic subcategory (e.g., zebra, tiger and lion for “wild animals” subcategory) in a verbal fluency task. They measured two key indices: cluster size (the number of consecutively listed items from a single subcategory) and cluster switching frequency (the rate of alternation between various semantic subcategories). Kuhn et al. (2007) defined lexical-semantic clusters as groups of source artifacts (e.g., packages, classes or methods) that use similar vocabulary; and tried to identify them using Latent Semantic Indexing (Deerwester, 1990) and clustering (Jain et al., 1999). Lexical-semantic clustering is a versatile methodology with diverse research applications. For example, semantic clustering can be used to investigate the processes of foreign language acquisition (Pérez-Serrano et al., 2022), cognitive memory strategies in aging (Shaffer et al., 2024), improving the efficiency and robustness of large language models (LLMs) (Lee et al., 2025).

While these studies employ lexical-semantic clustering to explore the structure of knowledge domains, we found no ready lexical-semantic clustering decisions to explore the extent of culture integration and the domains mediated by knowledge transfer. This study is the first attempt to develop this decision and to exemplify its efficacy in the context of determining the input of one language (donor language) into the second language (recipient language).

2.2. Culture integration exposed in direct and indirect knowledge transfer through loanwords

Following Myers-Scotton (2002, 2006) and Haspelmath (2009), in this study culture integration which is a complex phenomenon of forming shared cultural and historical knowledge, is addressed through lexical criterion, i.e., through the loanwords which fall into two types: “a) cultural borrowings which designate a new concept coming from outside, and b) core borrowings which duplicate meanings for which a native word already exists” (Haspelmath, 2009: 46). Whereas there is commonly no strong necessity to employ an alien loanword to construe a concept, it frequently penetrates the native language and culture due to a strong cultural convention in the community to use another language as a marker of cultural or ethnic identity. In most cases this is the social identity which stimulates the appearance of loanwords. Adapting this idea to exploring the distribution of loanwords in Russian and Kyrgyz, the official languages of two CIS countries, Russia and Kyrgyzstan, we expect to determine the knowledge domains of culture integration through their lexical systems integration. Importantly, in the study we do not classify these adaptation process into adoption or imposition (Winford, 2005) or reveal the reasons of borrowing, i.e., cultural pressure, loss of vitality, genealogical position, etc. (Thomason, 2001). To explore culture integration through lexical criterion of loanwords, we categorize the effect of a loanword from the donor language on the lexical stock of the recipient language as insertion without further classifying it into replacement and coexistence (Haspelmath, 2009), with the view of identifying the knowledge domains representing the loanwords in the recipient language and their distribution. To proceed, we adhere to the concept of cultural knowledge transfer following (Espagne, 1999; Feschenko et al., 2016) who view it as “culturally specific transfer of domains and conceptual modifications which appear in their importing and exporting from one culture into another” (Feschenko, Bochaver, 2016: 5). Knowledge transfer is commonly addressed through its techniques, technologies and routes (Postovalova, 2016; Demyankov, 2016; Iriskhanova, Kiose, 2016). The symptoms of culture integration are consequently the knowledge domains and their conceptual structure which are introduced or modulated through knowledge transfer. Consequently, culture integration is viewed through the integrated or modulated knowledge domains in the recipient culture development.

2.3. Qualia roles in identifying culture integration

To address knowledge domains and their structure, the cognitive framework developed a wide range of methods in exploring the cognitive processes of selection, perspective, abstraction, specificity, prominence, perspective, dynamicity, metaphoricity, integration, intersubjectivity (Talmy, 2000; Langacker, 2007). Meanwhile, these methods are seldom if ever used in computational linguistics since they are efficient in determining the principal regulations of construal processes, not their distribution (Verhagen, 2007, Gallagher, Hutto, 2008). To comply with the computational task, we address the method of qualia roles (Pustejovsky, 1991) developed within the generative lexicon framework.

A Quale is a term borrowed from philosophy and used in generative linguistics to refer to a single aspect of a word’s meaning identified on the basis of the relation between the concept expressed by the word and another concept that the word evokes (Pustejovsky, Jezek, 2016). Four qualia roles are thus explored: 1) Formal encoding taxonomic information about the lexical item (the is-a relation, what kind of thing is it, what is its nature?); 2) Constitutive encoding information on the parts and constitution of an object (part-of or made-of relation, what is it made of, what are its constituents?); 3) Telic encoding information on purpose and function (the used-for or functions-as relation, what is it for, how does it function?); 4) Agentive encoding information about the origin of the object (the created-by relation, how did it come into being, what brought it about?) (Pustejovsky, 1991, 2006). As claimed, this is the process of syntagmatic co-composition (Pustejovsky, 1991; Copestake, Briscoe, 2000) which causes sense modulation; consequently, the qualia can be attributed in context only.

For instance, in the contexts such as: a. This car weighs over 2,000 lbs.; b. We buy vehicles such as cars and buses; c. John started the car, d. You should warm your car up in winter, e. Did you lock the car?,  f. The car screeched down the road, two types of qualia are attributed to car, which are Formal qualia (sample b) and Constitutive qualia (samples a, c, d, e, in the construal of its engine, door, wheels). In the contexts such as a. He owns a two-story house, b. Lock your house when you leave, c. We bought a comfortable house, d. The house is finally finished (Pustejovsky, Jezek, 2016), all four qualia types are observed, which are Formal (sample a, house as an artifact), Constitutive (sample b, door as part of the house), Telic (sample c, purpose of the house), Agentive (sample d, origin of the house). Schematically, in (Pustejovsky, Jezek, 2016) these attributions are represented within formal semantic network in the following way (Figure 1).

Figure 1a. Qualia-roles attributed to car (Pustejovsky, Jezek, 2016: 8)

Figure 1b. Qualia-roles attributed to house (Pustejovsky, Jezek, 2016: 9)

The qualia structure framework was further adopted in ontology studies (Pustejovsky et al. (2004), in computational linguistics, e.g., in Copestake and Briscoe (1995), Cimiano, Wenderoth (2005), in the design of lexical databases, e.g., in Jezek et al. (2009), in the Brandeis Semantic Ontology proposed in Pustejovsky et al. (2006), in cross-cultural semantic studies (Lee et al., 2010; Song, Qiu, 2013). These and other works confirm the efficiency of the qualia roles framework in “mapping out the range of possible conceptual transfers available and also motivating their existence” (Copestake and Briscoe, 1995: 15).

Following these tenets, we hypothesize that semantic clustering integrated into cognitive framework reveals the paths of knowledge transfer through the distribution of a) thematic attribution of knowledge domains and b) qualia roles in the web contexts with loanword collocates.

3. Methods and Research procedure

The research is conducted within the framework of cognitive linguistics, which uses computational methods to explore how knowledge domains are distributed. This framework is used in the present study to identify the knowledge domains affected by the culture integration through knowledge transfer displayed in the loanwords and their contexts. Following the studies exploiting the notion of knowledge transfer in application to cultural contacts (Espagne, 1999; Feschenko et al., 2016), we expect to reveal its effects onto the knowledge domains distribution due to Russian-Kyrgyz contacts. To determine the integrated and modulated knowledge domains in the recipient culture, we consider two types of knowledge transfer in culture integration: 1) direct knowledge transfer which is a cognitive process exemplified in immediate incorporation of loanwords from the donor language into the recipient language appearing in the knowledge domains distribution; 2) indirect knowledge transfer which is a cognitive process exemplified in the knowledge domains distribution of target words as collocates of the loanwords. Indirect knowledge transfer displays the effectuated culture integration in the recipient language, since it reveals the knowledge domains which undergo the influence of loanwords.

Two types of knowledge transfer – direct and indirect – in the present study are explored in the web discourse database of the Kyrgyz language compiled and processed by the authors. Russian served as the lingua franca for Russian-Kyrgyz interactions, resulting in a large number of loanwords appearing in the Kyrgyz language. Meanwhile, their circulation within Kyrgyz culture reflects the norms and strategies of cross-cultural communication (Derbisheva, 2017, 2019; Kambaralieva, Sternin, 2019). At present due to the high rate of migration and at the same time the change in the vitality status of Russian in Kyrgyzstan (Shipilov, 2018), an efficient instrument for culture integration diagnostics is required. Since the web discourse reflects these social and cultural processes, we will use it to develop and validate the culture integration diagnostics instrument through revealing the knowledge domains affected by the knowledge transfer. Consequently, we apply corpus and lexical-semantic clustering analysis (Troyer et al., 1997; Kuhn et al., 2007, among others) to explore the direct and indirect knowledge transfer in this discourse type. Corpus analysis helps identify the distribution of loanwords and the thematic attribution of knowledge domains in the recipient language, while semantic clustering allows to reveal the distribution of target words displaying high vector similarity with the loanwords in the compiled web database obtained through semantic clustering and qualia roles (Pustejovsky, 1991; Pustejovsky, Jezek, 2016) representing the referents of these loanwords in the web contexts, exposing the results of indirect knowledge transfer.

Two criteria for exploring knowledge domains are thus exploited. While considering the distribution of knowledge domains of loanwords, both cultural and core borrowings (Haspelmath, 2009), we apply the thematic criterion. Meanwhile, this criterion is not efficient in identifying the knowledge domains of the target words (the loanword collocates) since they appear in their contextual meanings, display varied POS attribution and are far more numerous. Consequently, to proceed we apply the qualia roles criterion developed in the computational research within the generative linguistics (generative lexicon) framework.

To comply with the research objectives, we developed a two-step procedure of exploring knowledge transfer in culture integration through lexical criterion of loanwords integrating two steps: 1) identifying the domains of direct knowledge transfer through the distribution of loanwords in the compiled web database; 2) identifying the domains of indirect knowledge transfer through the distribution of collocates of the loanwords in the database.

Step 1. At this stage, we identified the lexical domains of direct knowledge transfer and its indicators: a) the ratio of loanwords from Russian as donor language into the Kyrgyz to all loanwords and to all words in the language, b) the ratio of loanword samples in web databases to the overall number of samples.

Step 1.1. To compile a list of loanwords from Russian into the Kyrgyz language, we addressed the loanwords dictionaries which identified the direction of borrowing and the donor language. Two dictionaries were applied for the purpose: 1) “The Dictionary of Loanwords” compiled by Kh. Karasaev (1986) which contains 5,100 loanwords, where the loanwords with Russian as donor language are marked with “oр” (оr), 2) “The Dictionary of the Kyrgyz Language” compiled by K. Yudakhin (1985) which contains 100,000 words, where the loanwords with Russian as donor language are marked with “р” (r). Two lists of loanwords were further integrated, the words appearing in both lists were counted as one. All word forms of each loanword were next included into the final list of loanwords. This procedure allowed to identify the key knowledge domains and their thematic attribution in performing the transfer and the lexical indicator of direct transfer considering the ratio of loanwords from Russian to the total amount of loanwords and the total number of words in the Kyrgyz language.

Step 1.2. Next, two databases of web discourse were compiled to perform the corpus search of loanwords frequency. Compiling two databases, with the first (Kyrgyz Web text corpus (Kyrgyzstan)) featuring the language use up to 2019, and the second (Kyrgyz News Corpus) featuring the language use up to 2025, was necessitated by the absence of one resource covering a representational period of loanword use. The Kyrgyz Web Text Corpus (Kyrgyzstan) is part of the open-access Leipzig Corpora, which comprises 3,036,362 sentences and 36,581,130 tokens. The Kyrgyz News Corpus was created by AI Kyrgyz specialists (as part of The Cramer Project) launched in 2025, which contains 256,364 news items (as provided in the description) collected through web scraping from a range of online news portals. The dataset encompasses a wide thematic spectrum, including politics, economics, culture, and sports. Each entry consists of the full text of a news item along with metadata on its source (The Kyrgyz News Corpus dataset, 2025). While the Kyrgyz Web Text Corpus allowed for the search process, the Kyrgyz News Corpus did not; therefore, extensive pre-processing and processing were performed in Python.

The primary objective was the development of a structured dictionary of borrowed lexemes implemented in Python. In this context, a dictionary is understood not as a lexicographic product, but as a data structure — “a set of key-value pairs, with the requirement that the keys are unique (within one dictionary)” (Python Documentation, 2025). The keys were defined as the parameters of each lemma, its graphical variants, and its word forms. Such an approach, standard in natural language processing, not only facilitated the systematic structuring of lexemes but also enabled the subsequent extraction of examples containing the target units.

Manually extracted data from the Leipzig Corpora Collection were merged with borrowing examples from the Kyrgyz News Corpus, resulting in a consolidated dataset used for the construction of semantic networks. Finally, the parallel databases were integrated to further explore the use frequency of each loanword within the integrated database serving as an indicator of direct knowledge transfer featuring the lexical decision.

Step 2. At this stage, we identified the qualia roles that demonstrate indirect knowledge transfer and the indicators of indirect cultural integration. These are: a) the lexical system indicator, which shows the scaled contingency rate of collocates of loanwords from Russian as the etymon language in Kyrgyz; and b) the functional indicator, which shows the scaled average contingency rate of qualia roles that demonstrate indirect cultural integration. To proceed, we elaborated the semantic clustering procedural decision for corpus analysis. It helped identify: a) the distribution of loanword collocates, and b) the semantic clusters of loanwords within the integrated database.

We applied Word2vec – a word-embedding algorithm proposed in Mikolov et al. (2013), to vectorize our database data. This algorithm allows representing words in the text database as high-dimensional vectors (embeddings). These vectors are primarily used to measure the semantic similarity between words. Therefore, the greater the similarity between word vectors, the more similar the words are semantically. This allows determining the closest neighbors of any word in the text database and analyze them as set words that form the context of a given word.

Next, the list with the neighbors for each of the loanwords was used to determine the qualia roles through semantic analysis of the collocates within the nominal groups (loanwords and collocates) and their distribution within the compiled database. We then determined the loanwords knowledge domains and their prevalence in the web database, which describes the direct knowledge transfer with Russian as a donor language and Kyrgyz as a recipient. Finally, we identified the qualia roles and their distribution of the target words as loanword collocates, which describes the indirect knowledge transfer.

4. Results and Discussion

In this Section we present the results which describe: 1) the knowledge domains representing the loanwords and the distribution of loanwords in the compiled web database, exposing the results of direct knowledge transfer: 2) the distribution of target words displaying high vector similarity with the loanwords in the compiled web database obtained through semantic clustering and qualia roles representing the referents of these loanwords in the web contexts, exposing the results of indirect knowledge transfer. The web database and the list of target words with their vector similarity indices as well as the list of target words with their qualia roles can be accessed at: https://osf.io/78u4m/ or https://doi.org/10.17605/OSF.IO/78U4M.

4.1. Direct knowledge transfer

On Step 1.1. a list of loanwords from Russian into the Kyrgyz language was compiled. The Dictionary of Loanwords edited by Kh. Karasaev (1986) contains 793 words borrowed from Russian, while the Dictionary of the Kyrgyz Language edited by K. Yudakhin (1985) contains extra 411 words, mostly graphological and phonological variants to the loanwords presented in Karasaev (1986). Considering two input dictionaries, we further obtained a list of 1,204 loanwords from Russian as a donor language into Kyrgyz as recipient language. With the overall number of loanwords equal to 5,100 in Karasaev (1986), the ratio of loanwords from Russian (addressing only the items listed in this dictionary) is 15.55%. With the overall number of words in the Dictionary of the Kyrgyz Language edited by Yudakhin (1985) equal to 100,000, the ratio of words from Russian is 1.204%.

On Step 1.2. we performed the corpus search of loanwords (including their word forms) frequency in the two compiled databases, the Kyrgyz Web Text Corpus (Kyrgyzstan) featuring the language use up to 2019, and the Kyrgyz News Corpus. The analysis of loanwords frequency revealed considerable differences in their web representation; therefore, only 64 most frequent loanwords with frequency exceeding 800 in the base form in the Kyrgyz Web Text Corpus (Kyrgyzstan) were subjected to further analysis. To determine the borderline frequency in selecting these words, we adopted the following procedure. The overall frequency of 1,204 loanwords from Russian as a donor language into Kyrgyz as a recipient language in the Kyrgyz Web Text Corpus (Kyrgyzstan) in the base forms amounted to 197,827. The average word frequency was 208.24. Meanwhile, almost one third of the loanwords either did not appear in the corpus, e.g., жаркеп (‘жаркое’, ‘roast’), кечур (‘кучер’, ‘coachman’), майыр (‘русский офицер’, ‘major’), матике (‘дама’, ‘дамка’, ‘dame’), нагирет (‘награда’, ‘award’), обосол (‘новосел’, ‘new house dweller’), памбаркут (‘панбархат’, ‘panne’), селкебай (‘целковый’, ‘one rouble’), тастаберне (‘удостоверение’, ‘certifying document’), турдаден (‘трудодень’, ‘workday’), тыпылыс (‘бязь тифлисская’, ‘calico’), чомун (‘в темную’, ‘with eyes closed’) чыстай (‘чистый’, ‘аккуратный’, ‘clean’), ыштан (‘штаны’, ‘панталоны’, ‘pantaloons’), etc., as seen mostly due to being the phonological and graphological variants. Additionally, a vast number of loanwords were infrequent (did not exceed 10), e.g., арыстан (‘арестант’, ‘prisoner’), агитатор (‘агитатор’, ‘agitator’), аккорд (‘аккорд’, ‘chord’), анархист (‘анархист’, ‘anarchist’), балалайка (‘балалайка’, ‘balalaika’), галифе (‘галифе’, ‘riding breeches’), драп (‘драп’, ‘drape’), колонизатор (‘колонизатор’, ‘colonizer’), лексикография (‘лексикография’, ‘lexicography’), материалист (‘материалист’, ‘materialist’), пленум (‘пленум’, ‘plenum’), шинель (‘шинель’, ‘overcoat’), etc. Three reasons for their infrequent use were identified: they either relate to the objects and notions which are not used in contemporary culture, not used in web discourse, or use other phonological and graphological word forms. Only 132 words out of 1,204 in the Kyrgyz Web Text Corpus (Kyrgyzstan) in the base form demonstrated the frequency which exceeded 400; with half of these being still infrequent, only the other half (disregarding the ones which were listed in the dictionary, however used only in their variants, e.g., борум (‘форма’, ‘form’)) was used for further analysis.

Next, we established the knowledge domains of 64 loanwords following the thematic criterion. Since we considered the loanwords lacking their context use, we adhered to their major dictionary meaning. These related to political, social, financial, cultural areas, as well as to calendar months, jobs, metrics, technology, evaluation spheres, establishments. Consequently, we can claim that both cultural and core borrowings (Haspelmath, 2009) appeared in this list. Next, we identified the distribution of these loanwords in the compiled web database, exposing the results of direct knowledge transfer. Their frequency in the Kyrgyz Web Text Corpus (Kyrgyzstan) in the base forms amounted to 119,476 (ranging from 800 to 13,500); in all forms – to 360,016. Their frequency in the Kyrgyz News Corpus in the base forms amounted to 173,399 (ranging from 411 to 18,034); in all forms – to 503,989. Considering the overall number of words in the Kyrgyz Web Text Corpus (Kyrgyzstan) equal to 36,581,130, we can claim that 13.108% of all words (tokens) functioning within web discourse appear from Russian as a donor language. Contrasted with the ratio of words from Russian listed in Yudakhin (1985), which is 1.204%, this number exceeding it more than 10 times shows that the borrowed words are highly active in contemporary discourse which proves that direct knowledge transfer has stimulated extensive conceptualization of the areas listed above.

Importantly, we ranged the frequency of these words to determine the key areas of direct knowledge transfer. As described above, the Kyrgyz Web Text Corpus (Kyrgyzstan) allowed for the search procedure, while pre-processing and processing was performed for the Kyrgyz News Corpus, given that the dataset itself was assembled through web scraping. Duplicates (880 items) were removed, and the data were cleaned of noise such as references to publishing outlets, tabulations, formatting artifacts, and copyright symbols. From the preprocessed dataset, web items containing at least one borrowed lexeme were extracted on the basis of the constructed dictionary. These texts were then segmented into sentences, and only sentences containing borrowings were retained in a new dataset. Each sentence was assigned a unique identifier, allowing the traceability of the extracted sentence back to its original news item in the main dataset. The resulting entries were stored in the form of id, lemma, word form, and the corresponding source sentence. All of the aforementioned steps, including the pre-processing of the raw dataset and the extraction of loanword-containing fragments, were implemented using Python and regular expressions. 

Table 1 presents a list of loanwords extracted from two databases, showing the movement of loanwords from Russian (the donor culture) to Kyrgyz (the recipient culture). Their frequency (considering all word forms) in the compiled database is indicated.

Table 1. Frequency of loanwords (abs and ratio, %) in the compiled web database

Таблица 1. Частотность заимствованных слов в собранном веб-корпусе (абсолютные и относительные (в процентах) величины)

Table 1 shows that several loanwords demonstrate the highest frequency exceeding 2 per cent, which are апрель (‘April’), банк (‘bank’), бюджет (‘budget’), градус (‘degree’), депутат (‘depute’), директор (‘director’), документ (‘document’), журналист (‘journalist’), комитет (‘committee’), медицина (‘medicine’), метр (‘meter’), парламент (‘parliament’), партия (‘party’), район (‘region’), сентябрь (‘September’), техника (‘technics’), экономика (‘economics’). These loanwords clearly demonstrate the main areas of knowledge that are transferred directly in web discourse: banking and finance; management and documentation; technology and economics; city management; and measurement and time control.

Рисунок 2. Слева: Семантическая сеть, состоящая из слов и семантических связей между ними. Красные узлы означают заимствованные слова из русского в кыргызский язык, серые – их ближайших соседей, определенных с помощью Word2vec. Цвет ребер определяется степенью семантической близости: более сильные связи обозначаются фиолетовым цветом, более слабые – желтым. Справа: Фрагмент этой сети.

Next, the list with the nearest neighbors for each of the 825 loanwords was used to determine the qualia roles through semantic analysis of the collocates within the groups (loanwords and collocates) and their distribution within the compiled database.

Below, we specified the procedure which helped identify the qualia roles of collocates within the groups (loanwords and their collocates). Four key principles were observed. All word forms of each of the 64 highly frequent loanwords were considered to identify the vector similarity with the left and right collocates appearing in the same sample (the distance between the loanword and the collocates was not observed). To explore the distribution of qualia roles, only the collocates with vector similarity index exceeding 0.6 with 5 most frequent collocates for each word form were addressed. Overall, 4,126 collocates serving as target words were subjected to semantic analysis. In case the target word and the loanword were one and the same word, the target word was not considered. In most cases, each loanword form displayed contingency with several target words. For instance, the loanword батир (‘flat’) with батир, батирге, батирде, батирден, батирди, батирдин, батири, батирим, батирлер, батирлери, батирлерибиз, батирлүү, батирсиз demonstrated contingency with 55 target words. Among them the word form батирлери had 5 collocates, батири (‘flat’) (0.79), мүлктөрү (‘property’) (0.75), мүлкү (‘property’) (0.74), эмереги (‘furniture’) (0.73), элиталык (‘elite’) (0.73). Four target words were consequently addressed in identifying the qualia roles, where мүлктөрү (‘property’) and мүлкү (‘property’) were attributed to the Formal role displaying taxonomic information about the lexical item (Pustejovsky, 1991; Pustejovsky, Jezek, 2016), where in the hierarchy the property domain relates to superordinate category contrasted to the basic categorization of the domain of flat. Meanwhile, эмереги (‘furniture’) was attributed to the subordinate level, since in the hierarchy of semantic representation the furniture domain displays the part-whole relations contrasted to the domain of flat; consequently, we identify its qualia role as Constitutive (ibid.). The target word элиталык (‘elite’) encodes the information on its sphere of use (who will use it); therefore, we attribute the qualia role as Telic displaying purpose and function (ibid.). Consequently, the qualia structure of батирлери is as follows (Figure 3):

Рисунок 3. Квали-роли слова батирлери