Concepts in Contexts: Discourse-based Semantic Networks of Ideologies in Taiwan (1945–1949)

ABSTRACT This article examines the semantic development of significant keywords in Taiwanese cultural discourse from a corpus-based inductive perspective. It contextualises these keywords ideologically and socially via extended network analysis and innovatively also via sociological analysis. The research investigated lexical patterns in what we refer to as the Taiwan Early Post-war Corpus, which consists of culturally oriented articles from 1945 to 1949. It treated shared ideologically loaded keywords as indicators of different ideologies which in turn identify the social groups who propagate them. Through this process we discovered three ideologically inclined semantic fields, which emphasise connections to the dominant cultural discourses of the central and local governments, as well as one that emphasises Taiwanese subjectivity. To understand their contemporaneous dynamics, the study examined networks associated with the authors and periodicals through shared keywords. Using the Taiwan Biographical Ontology, the study complemented ideological patterns with positional analyses of authors’ social involvements, which allowed us to treat them as proxies for various types of social, cultural, economic and political capital. We could thus characterise the habitus of each cluster and its position on an ideological map, thereby creating multi-layered networks of concepts, ideologies and social patterns.


Introduction
In 1945, Taiwan was incorporated into the Republic of China (ROC) after 50 years of Japanese colonial rule. Taiwan's reintegration represented an administrative, cultural and ideological challenge: the Nationalist government had to conceptualise Taiwan, as a linguistically and culturally heterogeneous region, within its existing epistemological framework of a culturally homogeneous modern nation represented by the ROC. We argue that in times of political limbo, such as early post-war Taiwan (1945)(1946)(1947)(1948)(1949) -i.e., before the Nationalists were forced to flee to Taiwan, where they introduced authoritarian rule -one can detect several competing ideological worldviews and link them to distinct groups of intellectuals.
These groups shared a habitus in the Bourdieusian sense and propagated different ideological standpoints. Habitus is here understood as a set of habits, skills and dispositions. It regulates dispositions to generate regulated and regular behaviour outside any reference rules (Bourdieu, 1990, 60). Habitus is neither wholly a result of free will, nor determined by structures; rather, it is created by a kind of interplay between the two over time (Bourdieu, 1984, 170). Habitus thus explains certain regularities in social structures that are shared by people with similar backgrounds, value systems or ideological standpoints. Our article focuses on the relationship between shared backgrounds and ideological standpoints to reveal regulated patterns in Taiwanese society in the early post-war period.
During this transitional period, new social structures and orders were negotiated, and ideas promoted by competing groups challenged the centre of political power, the Nationalist party (Kuomintang, hereafter KMT), which, in turn, was internally divided. We analyse a body of cultural commentary to explore how concepts developed within the discourses of different groups. The study thereby addresses the following questions: (1) What were the important discourses expressed by sociocultural keywords and their underlying semantic coherence for distinctive ideological groups? (2) What were the historical and social origins of the concepts and the ideologies expressed by these keywords? (3) How were group-specific keywords connected to the authors' socio-political backgrounds in post-war Taiwan?
Existing scholarship on the early post-war Taiwanese intellectual scene, influenced by post-1990s Taiwanese national subjectivity, mainly focuses on Sinicisation and de-Japanisation policies (e.g., Huang, 2007). Pro-Taiwanese academic discourse directly contrasts Mainlanders and their agency with Taiwanese. The KMT ideology of the early post-war period, however, has not been extensively studied. Using a representative collection of 1,168 digitised texts, our inquiry overcomes this dichotomy, showing permeability among ideological standpoints that are not delineated along ethnic lines and foregrounding strong ideological continuity from the wartime mass mobilisation policies formulated by the KMT.
Ideology and language have been studied in many ways. For Chinese-language discourse, the connection between power, ideology and the impact of ideology on linguistic features has been discussed mainly in the context of post-war China (e.g., Hodge & Louie, 1998;Link, 2013); moreover, discourses are usually viewed from the perspective of political leaders (e.g., Chang & Holt, 2009;Lams, 2018;Lams & Lu, 2018;Lu & Ahrens, 2008).
This article understands ideology as 'something like a shared framework of social beliefs that organise and coordinate the social interpretations and practices of groups and their members, and in particular also power and other relations between groups' (van Dijk, 1998, 8). Ideologies are expressed through language and reproduced in society through discourse. They are part of social structures and, as such, exhibit and control the relationships of power and dominance between groups. A thorough understanding of the ideological landscapes instantiated in a particular community or at a particular time may require a comprehensive comparison of their distinctive linguistic features. Above all, words that occur with unusual frequency in a particular context may draw special attention. Ideologies have also been studied from the perspective of their cognitive, social, political, cultural and historical conditions and functions. Our study innovatively combines two approaches: linguistic and sociological. By examining language, it identifies different contemporaneous ideologies in Taiwan and the social groups that support them. The content of these ideologies is interpreted historically. The social aspect of ideologies is studied through positional analysis applied to their proponents, which sheds light on different fields of activities they were collectively involved with: their habitus. In this respect, the study reveals the larger social structures that shaped and were shaped by contemporaneous ideology. With appropriate corpus-assisted methods, important words can be identified based on rigorous quantitative criteria and 'provide a representation of socially important concepts' (Scott, 1997, 233). Significant words that are identified by comparing their frequency in a corpus with their frequency in a reference corpus are known as keywords. This distribution-based conceptualisation of keyness is closely connected to the idea of keywords outlined by Raymond Williams (2015[1976), a well-known Marxist theorist: terms presumably carrying sociocultural meanings characteristic of Western capitalist ideologies. In contrast to Williams' intuition-based approach, the corpus-based method provides a promising alternative way of establishing key terms by using distributional information. This data-driven approach to keywords is sympathetic to the notion of statistical keywords popularised by Stubbs (1996;. Although defining keyness quantitatively is itself an important methodological issue (cf. Gabrielatos, 2018), for research on ideological inclinations expressed in writings produced by distinct groups or communities there is no agreement on how we can group statistical keywords in culturally significant ways to ensure their ideological informativeness. This article contributes to this method of keyword analysis by presenting a critical case study of the ideological currents in early post-war Taiwan. We address the most important step in keyword analysis -contextualising significant keywords within their relevant social backgrounds -by (a) extending the word-level analysis to an analysis of semantic networks, (b) re-establishing the connection of keywords and authors in the post-war era with post-hoc positional analysis, and (c) exploring the connection between keywords and the periodicals they appeared in.
In this study, we adopt a bottom-up corpus-based method to discover significant words that reflect the competing discourses. While stylometrists (e.g., Burrows, 1987;Jockers, 2013) commonly investigate linguistic particularities specific to individual authors, this study focuses on within-group similarities and between-group differences. We hypothesise that more ideologically aligned writers will manifest more shared lexical patterns than ideologically divergent authors. We are interested in a corpus-based methodology of text analysis and demonstrate how to arrive at a set of ideologically interesting keywords from a list of statistical keywords. To achieve this goal, we built the representative, small-scale, full-text Taiwan Early Post-war Corpus (TEPC) from culturally oriented writings published in contemporaneous periodicals. The corpus provides a linguistic foundation for the detection of ideologically loaded vocabulary, to be interpreted against the intellectual and ideological frameworks of the late Republican period . News media are important sources of ideology because media discourse reproduces ideas previously circulated or normalised in society (Vessey, 2017).
The extension from keyword lists to networks marks an important step towards our understanding of ideological landscapes in the early post-war era. Also, the sociological approach to the history of ideas has already explored how networks among intellectual actors created an autonomous arena of discourse. Life trajectories of thinkers have been studied in the context of the accumulation of intellectual cultural capital and the sceptical mentality typical among intellectuals -i.e., their habitus (Collins, 1998). Our data-driven sociological approach followed a similar path, using data stored in the Taiwan Biographical Ontology (TBIO). 1 The TBIO was designed as a biographical database of Taiwanese elites, of both Taiwanese and Mainland origin, for the period 1900-1949, and facilitates analysis of careerpath trajectories. 2 We examined each cluster of writers as a group with a unique habitus and carried out a positional study of writers' engagements with various organisations (introduced in Hoffman-Lange, 2018), mainly in the form of their employment and participation. We thereby determined the writers' social dispositions for each ideological position.
Our network analysis, in conjunction with corpus-based keyword analysis, has three implications for the study of early post-war Taiwan. First, to repeat, scholarly discussions about intellectual developments in Taiwan have focused on Sinicisation policies specific to this newly acquired former Japanese province. They were reflected in our analysis, but a greater diversity of worldviews emerged as well. Our previous analysis of publication patterns (Dluhošová & Chen, 2016) identified more diverse clusters of authors than are usually represented in the scholarship. This article, however, analyses published content and thus contextualises the dominant Sinicising ideology among competing worldviews, each supported by different social groups. This helps to connect Taiwan's post-war intellectual development with the late Republican period in China. Second, Taiwanese secondary scholarship has discussed the specificities of some periodicals at length and identified their political stances (e.g., He, 1996;. Our network analysis complements these studies and highlights links between periodicals, or nodes in the networks, by exploring their relative closeness through keyword distributions. This enables us to draw a more complex map of both periodicals and their political stances. Finally, the study of cluster-specific involvement with organisations innovatively combines a sociological analysis of habitus with intellectual history. Variations in group habitus illuminate author positions vis-à-vis the state and its power structures (i.e., public administration, KMT party structures, and the armed forces), showing that some groups of intellectuals in Taiwan resembled 'official intellectuals' (Guiberau, 2000) or 'establishment intellectuals' (Cheek, 2015).

The Taiwan Early Post-war Corpus (TEPC)
The Taiwan Early Post-war Corpus includes 1,168 digitised texts in Chinese published by 694 authors in 25 culturally oriented early post-war periodicals. 3 With 1,648,644 characters, the TEPC is a domain-specific small-size corpus. It includes mainly prose-like personal essays, short stories and opinion articles. 4 Due to funding limitations, the corpus does not include all prose writings from the 25 periodicals. Our selection strives to include as many journals as possible to ensure (1) diversity of publications, and (2) representativeness and diversity of authors.
The distribution of texts across different periodicals is biased towards long-running literary supplements with several hundred issues. We did not normalise this bias, since our question was which ideological positions and stylistic preferences dominated the early post-war period. Our collections indicated which texts readers were exposed to, and how well each position was represented.

Methods
We first processed the text of the TEPC to break each article into smaller units (i.e., paragraphs) and identify the word tokens from the character sequences of each paragraph. Based on these word tokens, we extracted frequently used unigrams and bigrams for later analysis. 5 Our methodology combined different forms of quantitative analysis. All data processing and analysis used R scripts developed by the first author ( Figure 1).
First, we used hierarchical cluster analysis to group texts into different sectors according to their similarities in lexical distribution (i.e., unigrams and bigrams). Distributional cut-offs were used to identify relevant and representative lexical items as the classifying features. The output was a tree-like structure of the texts -i.e., the dendrogram that showed their subgroupings. The scree-plot method (Kaufman & Rousseeuw, 2005) suggested a five-cluster solution for our data. As shown in Figure 2, Clusters 1 and 2 form one superstructure (B), and Clusters 3, 4 and 5 form another (A). Furthermore, A's internal structure shows that Clusters 3 and 5 are closely related, whereas Cluster 4 is attached only in the next step, suggesting a looser connection to the other two. The ideological position of these clusters is discussed below.
Second, we used a quantitative corpus-linguistic method, known as multiple distinctive collexeme analysis (MDCA; see Gilquin, 2006;Gries & Stefanowitsch, 2004), as a robust statistical variant of keyword analysis (Baron et al., 2009;Pojanapunya & Todd, 2018;Rayson, 2008) to determine each cluster's keywords -i.e., its strongly associated lexical features. The description of each cluster's semantic structures and ideological position derives from these keywords. As shown in Figure 3, Supercluster A (Clusters 3, 4 and 5) features vocabulary related to several political discussions and cultural policies, indicating its relation to ideologically loaded vocabulary. Supercluster B attracted vocabulary closer to colloquial language (e.g., personal and possessive pronouns or measure words). These texts were closer to the genres of short story or essay. The distinctive keywords also suggested that articles in Supercluster B comprised literary discussions on writers' creations. That does not mean, however, that these texts did not convey certain worldviews or ideologies. The distinction based on word frequencies may be analysed as being on the level of genres with their own respective styles. Because of its literary orientation, we do not discuss Supercluster B further.
Third, we used network science (Barabási, 2016;Newman, 2010) to visualise the semantic structures of each cluster in terms of keyword associations. Additionally, as each text was connected to a writer or periodical, we constructed author and periodical networks for works in each text cluster based on their shared use of distinctive keywords to examine the intricacies of their interrelationships. A network consists of nodes, with edges connecting nodes associated by some parametric factor. The edges of each cluster's keyword network represented the co-occurrence of these distinctive keywords within the same texts. The edges of the author networks were based on the use of the same keywords by authors. Similarly, the edges of the periodical networks were based on the use of the same keywords by the periodicals. In the next section we present network analyses of keywords, authors and periodicals, and discuss how our findings clarify the dynamics of the literary and intellectual landscape. Finally, after identifying author networks based on their relationships and proximity in terms of lexical choices, we contextualised the positions of these clusters by using the TBIO to study the authors' backgrounds. The TBIO is organised around people and organisations. 6 Organisations are classified according to societal sectors regularly used for positional analysis. 7 Combinations of societal sectors for each individual point to dispositions of individuals or groups and help characterise social groups that propagate certain ideas. This allows us to combine the study of discourse with sociology. In our positional analyses, we first identified authors in each network whose biographical records had been included in the TBIO and then computed the overall distribution of societal sectors involving the cluster's authors, normalised by the total number of authors.

Discussion: A Map of Early Post-war Ideology (Supercluster A)
Below we analyse the semantic structures of the text clusters with ideologically loaded vocabulary in Supercluster A. The three clusters identified by the hierarchical clustering analysis were named after (i) their core ideas (such as 'China', 'Taiwan', 'state' and 'literature'), (ii) their intended target recipients (i.e., central vs local), and (iii) their positions towards the official state ideology based on nationalism and the doctrine of the Three Principles of the People, which were aimed at building a strong modern state. We introduce each cluster, its core ideas, and its associated networks of intellectuals and periodicals.

The Nationalists' official discourse on culture
From 1925, the KMT launched a multifaceted state-building movement aimed at constructing a new political order -a prerequisite for law, economic development and social reconstruction. From the 1930s, and especially after the outbreak of the Second Sino-Japanese War in 1937, the Nationalist government launched several mobilisation programmes that can be viewed as a continuation of previous collectiveoriented efforts. Following a failed attempt at cultural transformation, including the abandonment of local customs and the production of modern citizens in Fujian Province in 1939, Friedman (2002) observes that the target of this mobilisation effort changed from lower-level elites to ordinary people. Writing about the Confucianinfluenced party-state textbooks on civic education designed to maintain social order, Culp (2002, 49) reconstructs attempts to reconstitute a 'cohesive social whole' -the collective recipient of the cultural transformation programmes -'that would structure a modern Chinese society and establish a firm relationship between national society and the party-state' to produce an organic, harmonious and hierarchical society. How were these centrally designed policies of wartime mass mobilisation transformed in Taiwan?
The keyword 'China' is the most central node with high-weighted edges in the Cluster 3 network and it dominates the first community (the green section of Figure 4a). Its immediate neighbours -'equality', 'treaty', 'international', 'state building' and their derivatives -hint that the texts in this community predominantly addressed international relations and national political issues -as one would expect after the war -and not cultural affairs.
The node with the second-highest weight degree is the term 'culture', which also dominates the second community (yellow in Figure 4a). However, it has comparatively few immediate neighbours. Among its close keywords are 'Taiwan', 'nation', 'citizen' and 'Three Principles of the People', which was a key doctrine of the Nationalist regime formulated by Sun Yat-sen . This keyword combination indicates its closeness to nationalism and a politicised approach to 'culture', as opposed to an artistic perspective.
The proximity of 'culture' to concepts connected to collective social groups (such as 'nation', 'youth' and 'state') in our corpus suggests that the regime conceptualised the cultural domain within its existing framework of mobilisation and cultural transformation. There is, however, one difference from previous Nationalist attempts at mobilisation. This time the construction of a social whole and civilising processes were aimed at Taiwanese people (as indicated by the presence of the keywords 'Taiwan' and 'Taiwanese' of the same community) who were perceived as separate from the Chinese nation. The cultural policies that aspired to create new 'citizens' thus became a core concern for a group of authors affiliated with the Taiwan Provincial Administrative Executive Office (Taiwan Sheng Xingzheng zhangguan gongshu, 台灣省行政長官公署) -i.e., the provincial government in Taiwan. These policies were often explained in editorials in the official newspaper, Taiwan xinsheng bao (e.g., Anonymous, 1945;Lu, 1945).
Terms for collectives are often linked to 'spirit'. 'National spirit' played an important role in wartime propaganda. Gu (2004) argues that it was associated with the new national culture, which served to indoctrinate and mobilise people during the war, such as during the Mobilisation of the National Spirit in 1937. The second initiative was the National Joint Pledge, which was issued in February 1939, to urge all compatriots to join forces and fight the Japanese enemy. The core vocabulary, including 'nation' and 'spirit', as well as the strong emphasis on 'education', foregrounds the later policy, known as 'National Spirit Education', which was officially launched in the 1950s and only ended in 1981 (Zhang, 2005). Our analysis suggests that the ideological mycelium was prepared well in advance of the hard indoctrination.
The unique combination of keywords in these two communities and their historical contextualisation thus innovatively situate the debate on the cultural reconstruction of Taiwan -which is usually conceptualised by Taiwanese scholars within the dichotomy of Chinese normativity and Taiwanese specificity (e.g., Huang, 2007) -in a broader framework that links it to the previous KMT state-building movement.
The last community (violet in Figure 4a) is dominated by two keywords: 'state' and 'citizen'. The associated keywords point to discussions on different forms of governance. In addition to terms such as 'self-governance' and 'constitution', which are both peripheral in our network, the terms 'democracy' and 'freedom' remained strongly associated with 'China'. This foreshadows the Cold War period, when the ROC on Taiwan was dubbed 'Free China' and was part of the Free World, as opposed to the People's Republic in the Communist bloc. The basis for this concept was already present in the early postwar period and widely used by both US and Chinese politicians. See, for example, comments by Admiral Cooke (1949) or by Liu Chien-chun (1950), the President of the Legislative Yuan.
The author network in Figure 4b was traced via keyword usage. Although we have labelled Cluster 3 'Official Nationalist Discourse', its author network suggests four prominent ideological undercurrents: (1) a group promoting national policies whose members did not strongly identify with the Three Principles of the People; (2) a group asserting policies at the provincial level that emphasise cultural issues; (3) conservatives strongly adhering to KMT ideology (nationalism and the Three Principles of the People); and (4) a group discussing governance and how to legitimise the Nationalist government. The following discussion illustrates how the vocabulary of each community betrays its unique ideological orientation.
The most prominent community (yellow in Figure 4b) is dominated by two authors -Jiang Zhongzheng, also known as President Chiang Kai-shek (Jiang Jieshi, 1887-1975 and the leader of the Nationalist party, and Zeng Jinke , a prominent intellectual with close ties to the KMT's military clique. 8 While Chiang represented the highest national authority, Zeng operated at the provincial level. Our analysis of the group's dominant keywords includes 'China' and its derivatives and 'state', suggesting attention to state affairs. That this group did not use 'Taiwan' bespeaks its focus on national politics. However, the relative absence of terms such as 'Three Principles of the People' or 'Father of the Country' (a reference to Sun Yatsen, the first ideologue of the Nationalist party) in this group may indicate its distance from Sun's (left-leaning) teachings.
The most widely shared keywords in the second-largest community (green in Figure 4b) are 'culture' and its derivatives and, interestingly, 'Taiwan'. Yet the authors here were not necessarily interested in a unique Taiwanese culture and its specifics. By contrast, they focused on Sinicisation policies that represented a provincial-level agenda, and many of the authors in this community served in various provincial governmental organisations. As for the previous group, scores for the official Nationalist ideology of 'Three Principles of the People' are comparatively low. This subgroup advances the cluster of 'local' official discourse observed in Cluster 4.
The rather small third community (dark blue in Figure 4b) differs from the previous two in many of its keyword choices. The most frequent word was 'spirit', often associated with the idea of mobilisation. Next came 'nation', 'soldiers' and 'state'. Judging by the size of Ke Yuanfen's  node, he probably represented this group's ideological standing. Ke, the most senior military officer in Taiwan, naturally discussed topics related to 'soldiers'. But his writing exhibited a clear inclination toward 'nationalism' and the 'Three Principles of the People', both of which lay at the core of Nationalist ideology. Furthermore, there were few occurrences of terms related to liberalism and progressive intellectuals, such as 'cultural movement'.
The last community (violet in Figure 4b) used terms rare or absent in the work of the others. The most frequently shared item was 'people', followed by 'politics', 'democracy', 'self-governance', 'government' and 'constitutional government', which evoked democratic constitutionalism. Terms connected to 'culture', 'nation' and 'nationalism', or 'Taiwan', were absent from the writings of this group, suggesting that their discourse was not Taiwan-specific. Figure 4d shows the different societal sectors to which the authors of Cluster 3 were connected. The engagement of authors (72.9% were identified in the TBIO) with various organisations confirms that they dominated the field of power, mainly in public administration (authors were involved with an average of 4.3 different organisations in public administration), but also in KMT party hierarchies, whose boundaries with state administration were blurred. Two observations stand out. First, high-ranking technocrats were usually recruited from academia, whose boundaries with administration were permeable. Second, uniquely for this cluster, relevant prestige could be drawn from military hierarchies. The distribution of societal sectors suggests that this cluster overwhelmingly enjoyed social prestige generated by high state positions, accounting for its proximity to official discourses. Therefore, even though keyword networks suggest considerable ideological variation within the cluster, we labelled it Nationalist. Many authors were affiliated with the KMT and served at various levels of the Nationalist government. It is highly likely that the Nationalist worldview was never unified, and that multiple political and ideological cliques vied for dominance.
Finally, Figure 4c shows a network of periodicals based on shared vocabulary. The algorithm identified two subgroups, diverging from the common understanding of the contemporary Taiwanese media sphere's division into 'official' (state-endorsed) and 'popular' (minjian) periodicals outside the state domain (Chen & Zhu, 1987). This is because journals not financed by the state could still be under the sway of political cliques. Cases in point are Jianguo yuekan and Zhengqi yuekan, which had been established by individuals around the Zhengqi Group, dominated by the KMT's military clique and the Central Club clique (Xu, 2013, 62-63). Hence, we posit that the divergence reflects differences in content. The coexistence of left-leaning oppositional periodicals such as the supplement Qiao and Xin xin on the one hand and pro-KMT journals such as Zhengqi yuekan or editorials in the official newspaper Taiwan xinsheng bao on the other suggests that debates between intellectuals with competing convictions took place in various types of journals. This phenomenon and the flexibility of ideological vocabulary reflect the existence of a moderate degree of freedom of expression. Keywords considered characteristic of one ideological group could appear in different kinds of periodicals or be used by authors of divergent orientations. Received ideological labels for periodicals do not accurately represent contemporaneous dynamics.

Local official discourse
From late 1945 and throughout 1946, the local government enacted plans to Sinicise Taiwan and suppress any Japanese cultural influence. Among other challenges, it had to transform the Japanese education system into a Chinese one to turn Taiwanese into Chinese citizens (Huang, 1996). Chinese-speaking teachers had to be hired and local students had to be prepared to understand them. This educational reform is reflected in our keyword network of Cluster 4, which we labelled the 'local official' cluster.
The keyword network of Cluster 4 ( Figure 5a) is dominated by three central items -'education', 'state language' and 'local' (lit. 'from this province') -which feature as core nodes of the two communities. This reflects the promotion of the national language in Taiwan, initiated by the Taiwan Provincial Executive Administrative Office (Huang, 2007).
Discussions of 'state language' touched upon the relationships between the Chinese, Japanese and Taiwanese languages. In the dominant contemporaneous narrative, 'Taiwanese' was cast as a scriptless 'dialect' that diverged from the 'standard' of the 'state language'. The 'state language', moreover, is linked to the 'national' through the morpheme guo ('state'), while Taiwanese is linked to the 'local'. But the exact definition of the relationship between Chinese and Taiwanese was not completely resolved at the time and remained a subject of heated debate. 9 In addition to the neutral vocabulary in the blue community, several terms are closely related to educational reform, including 'educational thought', 'culture of education' and 'civic education'. This suggests that the policies were part of a bigger plan to educate new citizens.
The difference between Cluster 4 (dealing with 'education' and 'state language') and Cluster 3 (centring on post-war Nationalist discourse) lies in their focal points. While Cluster 3 is generally concerned with cultural policies at the national and central levels, Cluster 4 addresses concrete Taiwanese local realities. There is a slight ideological overlap between the two: the language policies formed part of the cultural policies. The national official discourse and the local Taiwanese discourse merged after the arrival of the KMT government in Taiwan after 1949.
The periodical network in Cluster 4 ( Figure 5c) similarly divides into two main groups in accordance with the keywords 'education' (yellow in Figure 4c) and 'state language' (blue). The selection displays all periodicals close to the Sinicisation policies, although their content differed slightly. The main participants in the debate were the official newspapers Taiwan xinsheng bao and Zhonghua ribao and their respective supplements. Their editorials (shelun, nos. 2 and 7 in Figure 5c) frequently announced policy changes. Supplements in this cluster were thematically oriented. Guoyu (no. 8, literally 'state language') was expressly launched by the Taiwan Provincial Mandarin Promotion Council to discuss the conceptualisations of the languages and introduce teaching materials for new teachers. Similarly, the supplement Jiaoyu (no. 4, literally 'education')

ASIAN STUDIES REVIEW
dealt with educational reform at all levels. The supplement Wenhai (no. 1) introduced teaching materials for traditional culture in fields such as philosophy, history and literature. Journals such as Taiwan wenhua, Zhengqi yuekan and Jianguo yuekan, though not state-sponsored, were closely affiliated with state and party organs through their editorial boards. Unsurprisingly, they supported official cultural policies.
Finally, the author network of Cluster 4 ( Figure 5b) suggests two communities. The yellow one, with the central figure He Rong (1903-1990, an educator and linguist, features shared vocabulary related to 'state language'. Not surprisingly, four of the five identifiable network members served on the State Language Council. The second community (violet) is interlinked via vocabulary related to 'education'. Five of the seven authors identified in the TBIO were employed at the Directorate of Education of the Taiwan Provincial Executive Administrative Office and directly involved in educational reform (Huang, 2007). The main societal sectors uncovered by positional analysis (60% of authors were identified in the TBIO) are higher education and public administration, through which new regulations were enforced (Figure 5d). Along with social capital from involvement in administration, the academic sector generated symbolic capital by legitimising new policies. For members of this cluster, other societal sectors played a lesser role.

Literary discussions and Taiwanese subjectivity
The keyword network in Figure 6a suggests that Cluster 5 is heterogeneous, with two main and five minor communities. Unlike Cluster 3 (official KMT discourse), where core vocabulary was tied to 'China', Cluster 5 is dominated by vocabulary related to 'Taiwan' and its derivatives, as shown in the largest community (blue in Figure 6a). The relationships between the keywords suggest that the cluster is seemingly unconnected to official policies, with a predilection for 'literature', a topic alien to the previous two clusters. This large community also included discussions on two important figures in Taiwanese literature and culture, Yang Kui (1906-1985) and Yang Yunping (1906-2000, whose prestige traversed the 1945 divide (Dluhošová, 2020b) and whose work was discussed in the contemporaneous media. The term 'movement', indicating change, hints that these keywords were likely used by progressive intellectuals -which again differed from previous clusters.
The second big community of keywords (violet in Figure 6a) is linked to Taiwan and Taiwanese, dominated by the patronising phrase 'ours', which, together with terms such as 'Taiwan' and 'inner regions', suggests use by Taiwanese. The term 'inner regions' consists of a morpheme 'in' (nei) that refers to Japan, as opposed to 'out' (wai), which implies a peripheral Taiwanese position. Conversely, 'compatriot', also meaning 'sibling from the same womb', should have signalled a close, familial relationship between the peoples of China and Taiwan. This was a signal from the politically dominant Mainland Chinese minority to the Taiwanese population. Terms such as 'to expropriate', 'to govern' [by illegitimate ruler] and 'imperialist' indicated a discussion often led by the Mainlanders about methods of government. In comparison to the similar discussion on ways of ruling in Cluster 3, this one was strongly localised and related to the previous Japanese rule. The origin of this cluster's authors can thus be identified by characteristic lexical choices. It may be surprising that a cluster with literary-oriented vocabulary such as Cluster 5 is linked with the ideologically oriented -that is, propagandist -Clusters 3 and 4 (cf. Figure 2). Pertinent debates from Cluster 5 were concerned with the definition of Taiwan, and its literature and culture, as opposed to 'China'. Although 'China' is not a prominent keyword in the network, it is subliminally present in the discussion. Cluster 5 represents reactions by intellectuals of Chinese and Taiwanese origin to Sinicising cultural policies.
These reactions were concentrated around a few heated debates on the essence of Taiwanese new literature (core keywords in the blue community in Figure 6a). The postwar 'new' was conceptualised against the Japanese colonial 'old'. But while Mainland intellectuals considered the Japanese legacy to be a 'heterogenous other', Taiwanese authors, like Lin Shuguang or Yang Kui, cherished the continuity of Taiwanese literature from the previous period. Taiwanese literature was inspired by the modern Chinese literature of the May Fourth Movement of the 1920s and had developed against the strengthening cultural policies of Japanisation. The legacy of the May Fourth Movement in Taiwan was another hotly contested topic. Some authors saw Taiwanese literature as an heir of the May Fourth Movement legacy, some wanted to re-install Chinese literature in Taiwan (e.g., Qian Gechuan and Xu Shoukang), and some -mainly Mainland authors with a radical proletarian background (e.g., Luotuo Ying) -even disputed its importance and wanted a 'new' literature disentangled from previous legacies (see, e.g., Peng, 1997).
The network of periodicals ( Figure 6c) shows a rather high degree of homogeneity, pointing us to periodicals in which literary discussions took place. The author network of Cluster 5 comprises two large communities (Figure 6b). Both encompassed authors of Taiwanese and Mainland origin. Both were preoccupied with Taiwan and its literature. Their differences are subtle: the yellow community in Figure 6b tends to discuss 'new literature' and 'Taiwan new literature', with a hint of more mobilising vocabulary such as 'literary movement' and 'to build Taiwan', and evidence of literary debates about the work of Yang Yunping and Yang Kui. The shared vocabulary of the violet group also included vocabulary pointing to the uneven relationship between China, Taiwan and Japan. The limited polarisation of authors and concepts confirms that this is ideologically a rather coherent cluster.
Our positional analysis (52% of the authors were identified in the TBIO) further shows that this cluster attracted different groups professionally involved in the production of periodicals, such as publishers, journalists or editors, all belonging to the societal sector of the mass media ( Figure 6d). As observed elsewhere (Dluhošová, 2019), involvement in periodicals or teaching were natural career choices for Republican-era intellectuals. However, the positional analysis also identified heightened involvement in public administration for authors of Cluster 5; academia was only the sixth most prominent sector. The salient involvement in administration may be attributed to two factors. Firstly, Mainland intellectuals who voiced political views mostly came to Taiwan as government employees. Secondly, some Taiwanese intellectuals worked in culture-related state organisations.
Surprisingly, the third most important sector in the positional analysis is private business. Some Taiwanese authors were members of the local elite, involved in private business, and thus independent of the state (Dluhošová, 2020a). Some elite intellectuals produced their own periodicals, often in opposition to state periodicals. These authors were naturally more active in cultural organisations, unlike those in previous clusters, which stayed closer to the local and central official discourse.
The final surprising sector with an important role in this cluster is politics. Usually, authors with close cultural ties do not have strong relations to politics. In this case, however, authors did not necessarily have connections to the KMT, but were often connected to associations with political goals, such as the Taiwan Revolutionary Alliance (which was active during the war in China) or the Taiwan Local Autonomy Alliance (which operated in the Japanese colonial period in Taiwan), both of which lobbied for Taiwanese interests. To capture their political stances, more refined categorisation should be applied.
The positional analysis thus elucidates the habitus of intellectuals in Taiwan and reveals their position to be a continuation of that of Chinese Republican-era intellectuals -i.e., those active in publishing and education. It supports the observation of the gradually changing position of intellectuals: pre-war era intellectuals predominantly 'outside' the state entered the state domain during wartime mobilisation (Dluhošová, 2019). Intellectuals' early post-war proximity to state hierarchies is symptomatic of the rise of 'establishment intellectuals' (Cheek, 2015, 21) in Taiwan, which is a feature of all the clusters under scrutiny.

Conclusion
This article makes three key contributions. The first is methodological. The corpusbased analysis of early post-war writings helps to clarify contemporaneous ideological dynamics. Lexical patterns divide texts into two superclusters connected to style and content. So a novel contribution of this article is to illustrate how one can scrutinise the ideological inclinations in the context of intellectual history through linguistic analysis. Furthermore, this study extends prior work on the early post-war discourse landscape in several important ways: we adopted a rigorous quantitative method for computational text analytics, facilitating an extensive analysis of authors and periodicals neglected by previous scholarship; we drew insights from quantitative corpus linguistics and used statistical methods to identify keywords and phrases from particular text collections; and we used network analysis to reconstruct lexically grounded interrelationships between authors and periodicals. This powerful quantitative technique not only greatly reduced the risk of limiting the discussion to a narrow canon but also generated comprehensive visualisations that captured the ideological scope of the period.
The second contribution is interpretative. The new methods elucidated dynamics among different types of writing, publications and authors. We introduced three ideological standpoints: Nationalist official discourse on culture, local official discourse, and oppositional literary discussions on Taiwanese subjectivity. Prior studies of this era have usually understood the first two categories as the KMT's ideology. Our analysis pointed out different foci (central and provincial) and two sets of different actors. We also linked official discourses in Taiwan with the KMT's previous state-building attempts, thereby contributing to knowledge of the party's ideological development after the war and before the Cold War period. The three categories can be internally heterogeneous, suggesting ongoing ideological negotiations among participants in the debate. This observation overcomes the conventional wisdom about this period as exclusively involving negotiations between the dominant Mainland group and dominated Taiwanese. Furthermore, from a lexical perspective, discussions about Taiwanese literature were highly 'political'. In the late 1940s the divide between the fields of literature and power was thin. Some authors may have argued for the autonomy of Taiwanese literature on the discursive level, but they did so in response to pressures from the field of power. Finally, the journal networks challenge the received division of periodicals into 'official' and 'nonofficial' based on their funding bodies, which does not fully capture the relationship between these periodicals and their ideological standpoints. As our analysis showed, even 'non-official' periodicals could endorse official policies.
Lastly, by uncovering how different ideological standpoints structured the Taiwanese media sphere and reflected the positions of political and cultural elites, this article contributes to the sociology of ideas. Our study demonstrates that proponents of divergent ideologies can share a distinct habitus: those supporting the Nationalist official standpoints were high-ranking bureaucrats with prior academic careers and politicians, but also high-ranking army officers. Those supporting the local Sinicisation policies, meanwhile, were mainly educators and bureaucrats at the provincial level. Finally, those backing the notion of unique Taiwanese literature and culture were affiliated with the media and public administration, but also private business, securing them independence from the state. Intellectually, they form distinctive ideological groups, and strategic manipulation of shared discourses may allow them to occupy certain positions on an imaginary ideological map. In the future, we intend to enlarge the scope of our study to situate the early post-war Taiwanese case within the larger late-Republican ideological map, marking both the uniqueness of the Taiwanese intellectual sphere and its conceptual continuities with the Mainland, which gained in significance with the arrival of Mainland Chinese intellectuals.