Digital resources in the Social Sciences and Humanities OpenEdition Our platforms OpenEdition Books OpenEdition Journals Hypotheses Calenda Libraries OpenEdition Freemium Follow us

Behind the Counter: Collocation Patterns in the Shopkeeper Lexicon, 1872–1949

Introduction


The category of “shopkeeper” in late Qing and Republican Shanghai was far from uniform. Terms such as 店主, 東家, 掌櫃, 業主, 行主, 坊主, and 舖主 referred to actors embedded in different economic functions, degrees of authority, legal responsibilities, and social roles. Although the previous analysis established how these shopkeeper terms rose, declined, and circulated over time, frequency alone cannot tell us how they functioned in context. To move beyond distributional patterns, we must examine the environments in which these terms appeared—the situations, conflicts, institutional settings, and thematic frames that gave them meaning. Collocation analysis provides the first systematic entry point into this contextual world. Unlike concordances, which simply extract surrounding text indiscriminately, collocations reveal which accompanying terms recur with statistical weight. They therefore allow us to reconstruct the semantic conditions under which each shopkeeper label was deployed, and to trace how those conditions changed as Shanghai’s commercial order evolved.

Collocation analysis provides a robust methodological foundation for such an inquiry. In textual corpora, words rarely appear in isolation: they acquire meaning through the company they keep. By identifying which expressions systematically occur near specific shopkeeper terms, we can reconstruct the semantic fields within which these actors were represented. Collocational evidence reveals dimensions of commercial life that are otherwise dispersed across the corpus: the vocabulary of property rights and rental arrangements; the language of credit, debt, and monetary disputes; the register of legal cases and police involvement; and the rhetorical framing of trust, reputation, and social hierarchy. As such, collocations offer a window into how the press articulated commercial practices, property relations, legal frameworks, social stereotypes, and the emergence of more formalized institutions of economic governance.

This study analyzes collocations across the same periodized corpus that we have established, covering 1872 to 1949. The method proceeds in several steps. First, collocation retrieval identifies the most frequent co-occurring terms for each shopkeeper label in successive periods. Second, network analysis visualizes the structure of the relationships between terms, showing how clusters form around specific vocabularies and how these clusters shift over time. Third, lambda-based measures of collocational distinctiveness provide a quantitative assessment of how tightly each shopkeeper term is associated with its semantic field, allowing us to track processes of specialization, divergence, and semantic consolidation.[1]

The collocational evidence discussed below suggests that the vocabulary of shopkeepers was neither stable nor homogeneous. Instead, it underwent periods of structural transformation, role differentiation, and semantic polarization. The late Qing press depicted a fragmented and loosely regulated world of shopkeeping, while the early Republic and especially the Nanjing Decade saw increasing consolidation around legal, property, and administrative domains. Wartime disruptions introduced new vocabularies of scarcity, urgency, and conflict. Ultimately, the semantic evolution of shopkeeper terms reflects the broader economic and institutional transformations of modern Shanghai, capturing how commercial actors were made legible through language.

  1. 1872–1894: A Fragmented Starting Point

The first period of the corpus, covering 1872 to 1894, presents a lexical landscape in which shopkeeper-related terms are only weakly differentiated. Although labels such as 店主, 東家, 掌櫃, 坊主, 棧主, and 行主 all appear in the press, their collocational environments are modest in size and marked by a high degree of dispersion. Rather than forming coherent semantic neighborhoods, their associated vocabularies remain diffuse, reflecting a commercial world characterized by limited institutionalization and relatively fluid role boundaries.

The collocation lists for this period show that most terms appear with only a small number of distinct lexical companions. Co-occurrences* tend to cluster around pragmatic or everyday concerns: minor disputes between merchants and customers; testimonies or statements recorded in the course of investigations; small-scale debts or unpaid balances; and routine operational matters such as pricing, supply, or delivery. These clusters are narrow and sporadic, indicating that the reporting on commercial life revolved primarily around episodic events, rather than stable institutional practices or well-defined occupational categories.

Across the different shopkeeper terms, the dominant collocates frequently include verbs associated with reporting, presenting statements, or describing transactional friction. This suggests that early press coverage framed shopkeepers more as individuals involved in localized problems or irregularities. The resulting lexical field lacks internal hierarchy: no single term stands out as occupying a more formal, institutional, or distinguished position.

The bipartite and term-centered networks for 1872–1894 (Fig. 1) offer a visual confirmation of this fragmented collocational environment. Rather than converging around one or two central terms, the network forms a constellation of small, disconnected clusters. Individual shopkeeper terms sit at the center of their own very limited micro-networks, each linked to a handful of collocates associated with disputes, police cases, provisioning trades, or quotidian activities. The overall structure resembles a patchwork of mini-economies of meaning rather than a unified semantic field. This absence of centrality or structural coherence reflects the loosely regulated nature of commercial life during the late Qing period. The press did not yet treat “shopkeepers” as a category embedded in clear institutional frameworks. Instead, each sub-type appears in isolation, framed by the particular contingencies of the events that brought them into the news.

Figure 1. Shopkeeper Collocation Network (left) and Key Collocation Network (right) in 1872-1894

[see all graphs in large size at bottom of blog post]

  1. 1895–1911/1912–1927: Consolidation of the Lexical Field

Although the periods 1895–1911 (late Qing) and 1912–1927 (early Republic) differ in their political and institutional structures, the collocational and network evidence shows that both participate in a shared semantic trajectory: the gradual consolidation of shopkeeper-related vocabularies around legal, administrative, and procedural frames. For this reason, the two phases are treated together as part of a continuous process of lexical structuring, even though their internal dynamics from a historical perspective— especially the emergence of labor and committee-related vocabularies after 1912—retain distinct characteristics. The purpose of this grouping is analytical: to highlight the long-term semantic movement preceding the sharper specialization of the Nanjing Decade. The turn of the century introduced clearer lexical patterns that reflected the rise and growing influence of municipal bodies and merchant organizations, both in the foreign settlements and the Chinese-administrative areas. Certain terms began to cluster around legal, administrative, and fiscal expressions, suggesting a growing bureaucratic presence in urban commercial life. At the same time, differences among the various shopkeeper labels became more pronounced, laying the foundations for the semantic specialization that will reach maturity in the Nanjing Decade.

The raw collocation profiles for 1895–1911 show an intensification of verbs and nouns associated with policing, legal procedures, and investigative actions. Terms such as 押 (collateral or custody procedures), 捕房 (police station), 拘 (detain), 解送 (escort or transfer in legal contexts), and 案 (case) appear with increasing regularity around 店主, 掌櫃, and 行主. These collocates signal that commercial actors were more frequently involved in, or at least more frequently reported within, formalized processes of policing, dispute resolution, and administrative regulation. This trend continues and becomes even more visible in 1912–1927, where the transition to Republican governance brings a vocabulary of bureaucratic actions: 議決 (deliberate and decide), 核辦 (investigate and process), 批件 (official document), and 登報 (public notification). Their proximity to 店主 suggests that shopkeepers were more tightly enmeshed in administrative routines, whether in tax matters, permits, or public announcements. The collocational evidence therefore traces the increasing bureaucratic visibility of commercial actors, reflecting the city’s expanding administrative apparatus and the broader modernization of economic regulation.

The bipartite   and term-centered networks for these two periods make the structural consolidation of the lexical field particularly clear. In 1895–1911 (Fig. 2), several shopkeeper terms begin to cluster more tightly around legal and procedural vocabulary. 店主 becomes a visible pivot in the network, with strong ties to collocates associated with police actions and formal investigations. 掌櫃 and 行主 do not yet rival 店主 in centrality but increasingly share some of the same legal vocabulary, suggesting that the boundaries between managerial and proprietorial roles are becoming more consistently represented in the press.

Figure 2. Shopkeeper Collocation Network (left) and Key Collocation Network (right) in 1895–1911

The networks for 1912–1927 (Fig. 3) reinforce this development. They exhibit enlarged clusters connected to 店主, not only in legal contexts but also in wage and labor negotiations, which begin to surface prominently in the focused networks. These networks bring into relief new types of institutional interactions—petitions, committee deliberations, and workforce disputes—that become increasingly integral to the representation of shopkeepers. Their emergence illustrates the broader reorganization of commercial life under early Republican governance, where new institutional arenas such as municipal bodies, associations, courts, and arbitration committees offered new structured channels for resolving economic and labor conflicts.


Figure 3. Shopkeeper Collocation Network (left) and Key Collocation Network (right) in 1912–1927

Lambda-based distinctiveness scores offer quantitative confirmation of the semantic consolidation suggested by the networks. Between 1895 and 1927, several core termslike 店主, 掌櫃, and to a lesser extent 行主 display rising lambda values, reflecting the tightening of their collocational profiles and the growing regularity of the language surrounding them. Their semantic fields become more clearly defined, rooted especially in legal, administrative, and procedural vocabularies. Conversely, other terms such as 坊主 and 舖主 remain lexically unstable or decline in distinctiveness, indicating their diminishing salience in the textual representation of commercial actors. This divergence marks the beginning of meaningful semantic differentiation within the shopkeeper lexicon. As some labels become increasingly formalized and institutionally situated, others fade into peripheral or specialized use. By 1912–1927, the lexical field reveals both consolidation and internal stratification, prefiguring the much sharper polarization that will characterize the Nanjing Decade.

  1. 1928–1937: Functional Roles and Semantic Polarization

The period from 1928 to 1937 constitutes the decisive inflection point in the semantic evolution of shopkeeper terminology. All layers of analysis—raw collocations, network structures, and lambda-based distinctiveness—converge to show an unprecedented concentration and polarization of meaning. The vocabulary associated with shopkeeper terms becomes more tightly organized, more specialized, and more clearly differentiated, reflecting an urban and regulatory environment undergoing rapid institutional expansion. The press of the Nanjing Decade portrays commercial actors not merely as figures involved in daily interactions or legal disputes, but as embedded in a complex economy structured around property, rent, finance, and administrative oversight.

The raw collocation lists for this period show a marked shift toward property-related, rental, and financial terminology. Collocates such as 房屋 (buildings), 原 (former property owner), 租 (rent), 租金 (rental fees), 收回 (reclaim), and 產權 (property rights) appear with much higher relative frequency than in earlier periods. This increase signals that the semantic field of shopkeepers is now tied to a vocabulary of ownership, tenancy, and contractual obligations, domains that correspond to Shanghai’s increasingly regulated urban economy and expanding property market.

Figure 4. Shopkeeper Collocation Network (left) and Key Collocation Network (right) in 1928–1937


The networks reinforce this trend dramatically (Fig. 4). The 1928–1937 key and focused networks display sharply formed assemblies around financial and property-management vocabulary. Multiple shopkeeper terms converge on the same property-legal cluster, creating a dense core around transactions, tenant relations, and legal procedures for retaining, reclaiming, or disputing property. The effect is a clear reorganization of the semantic landscape into well-defined thematic blocs that reflect the economic and administrative priorities of the Nanjing Decade.

Despite the overall clustering around property and financial vocabulary, the semantic trajectories of individual shopkeeper terms diverge significantly, revealing a high degree of lexical specialization. 店主 transitions from the earlier emphasis on legal or debt-related contexts toward specialization in property and rent disputes. Its collocational profile includes strong ties to 房屋, 租, 房東, and 收回, showing that the term becomes central in press discussions about ownership and tenant relations. 掌櫃 retains its long-standing connection to legal and bureaucratic vocabulary. Collocates such as 案, 查, 調查, 捕房, and 核辦 indicate continuing representation in administrative processes, investigative contexts, and managerial disputes. Unlike 店主, which increasingly reflects property-based responsibilities, 掌櫃 remains embedded in institutional routines.

東家 shows the strongest shift toward organizational and committee-related vocabulary. Its proximity to 商議, 會議, 決議, and 聯席 suggests that it becomes associated with leadership roles in assemblies, guild-like deliberations, or collective decision-making bodies. This consolidates 東家 as a term carrying a connotation of higher-level authority and institutional participation. 行主 remains linked to narrower transactional functions, such as supply, delivery, or specific trade-based operations. The term does not achieve the semantic expansion seen with 店主 or 東家, indicating that it maintains a relatively restricted commercial scope within the larger system.

Figure 5. Evolution of Collocation Strength (Lambda) by Term Type

These trajectories show not only the consolidation of the overall field but also its internal polarization. Each term increasingly occupies a distinct semantic zone, reflecting clearer functional differentiation among commercial actors. The lambda-evolution results confirm that 1928–1937 is the moment of maximum lexical consolidation (Fig. 5). Figure 5 requires a brief methodological pause, as its structure may not be immediately intuitive to readers unfamiliar with collocational modeling. Each panel corresponds to a single shopkeeper term and plots the λ-values of its five strongest collocates across the five chronological periods. Rather than tracking the frequency of a word or its total number of associations, λ measures the “pull” that a term exerts on its most characteristic semantic neighborhood. A rising curve therefore signals not only that certain collocates recur more systematically, but that the term itself becomes embedded within a progressively tighter and more distinctive semantic environment. Conversely, flat or irregular trajectories reveal lexical instability or fragmented usage. Reading the graph horizontally makes it possible to compare periods within a single term, while a vertical reading highlights contrasts between terms—showing, for instance, that 東家 and 掌櫃 reach pronounced λ-peaks in the Nanjing Decade, whereas 坊主 or 舖主 display weaker or inconsistent profiles. In effect, Figure 5 visualizes the consolidation, divergence, and relative maturity of the semantic fields associated with each shopkeeper label. Most shopkeeper terms reach their highest lambda values in this decade, indicating that their collocational profiles become both more distinctive and more stable. The magnitude of the increase is especially pronounced for 店主, 東家, and 掌櫃, whose semantic fields achieve a level of specificity not present in earlier decades.

This lambda peak has several implications:

  • Lexical consolidation: The terms acquire sharply bounded semantic fields aligned with property rights, legal procedures, organizational deliberations, or specific transactional functions.
  • Role specialization: Each label becomes consistently associated with a particular economic or institutional role, reducing ambiguity and strengthening its representational coherence.
  • Sharper semantic boundaries: Overlaps diminish as terms drift apart into specialized domains, signaling a restructuring of how the press conceptualized different commercial actors.

Together, these patterns reflect the heightened economic regulation, administrative oversight, and legal formalization characteristic of the Nanjing Decade. The linguistic evidence shows how shopkeepers were increasingly represented through institutional and property-based frameworks, capturing the broader project of rationalizing the urban economy by the Nationalist government.

 

  1. 1938–1949: Reordering of the Lexical Field in Wartime

The wartime years introduced a new phase in the semantic history of shopkeeper terminology, one marked by instability, fragmentation, and uneven linguistic trajectories. The highly consolidated and functionally differentiated lexical field of the Nanjing Decade was disrupted by conditions of goods scarcity, inflation, displacement, and administrative improvisation. Successive overlapping regimes and organs managed the various sectors of city, most with little efficiency.  During this unstable period, while certain terms retain or even increase their distinctiveness, others weaken significantly, reflecting a crisis-driven restructuring of commercial life. The press depicts shopkeepers more as actors navigating emergency conditions, requisitioning measures, and volatile property relations.

The raw collocation profiles for 1938–1949 show new associations that emerge around terms such as 急 (urgent), 缺 (shortage), 荒 (scarcity), and 徵 (requisition), indicating that shopkeepers often appeared in reporting related to wartime disruptions of supply, forced allocations, and crises in essential goods. At the same time, property vocabulary intensifies further for terms like 業主 and 店主, with collocates such as 租金 (rent), 房屋 (building), 收回 (reclaim), and 產權 (property rights) appearing even more prominently than in the preceding decade. The wartime contraction of commercial stability magnifies disputes over rent, tenancy, and repossession, making these topics central to shopkeeper-related reporting. The combination of this dual trend reflects the pressures of wartime urban governance, in which commercial actors were caught between market instability and heightened state intervention.

The network structures for this period differ sharply from the dense, coherent formations of the Nanjing Decade. Several shopkeeper terms revert to forming small or isolated clusters, an echo of the fragmented networks of the late Qing period, though now shaped by emergency-related vocabulary rather than everyday disputes. This fragmentation reflects the unstable market conditions of the wartime economy, characterized by abrupt shortages, administrative requisitions, and fluctuating property relations.

Figure 6. Shopkeeper Collocation Network (left) and Key Collocation Network (right) in 1938–1949


The focused networks (Fig. 6) reveal especially strong poles around: Urgency (急, 急需, 急告), Requisition and seizure (徵, 扣押, 查封), Rent and tenancy conflicts (租金, 欠租, 催租). These clusters show how shopkeeper-related reporting became dominated by crises of supply, enforced redistribution, and intense competition over rental arrangements. The press thus represented shopkeepers not simply as commercial intermediaries but as participants in a precarious economy shaped by emergency measures and regulatory instability. Lambda evolution adds crucial depth by showing how the semantic stability of different shopkeeper terms diverged during the wartime period. Rather than rising together, as in the Nanjing Decade, the terms now split into two contrasting trajectories.

On the one hand, we see highly distinctive, semantically stable terms: 業主, 掌櫃, and 東家 maintain or even increase their lambda values. These terms remain anchored in well-defined semantic fields (property rights, legal procedures, organizational functions), suggesting that higher-status or more institutionally embedded commercial roles retained clear representational profiles despite wartime disorder. On the other hand, we observe weakening or marginal terms: 坊主, 舖主, and 行主 exhibit declining lambda values. Their collocational environments become more diffuse or irregular, indicating that these lower-status or narrower commercial categories lost linguistic coherence as the press focused more heavily on actors involved in property disputes or administrative interventions. This divergence reinforces the broader pattern of crisis-driven stratification. The wartime economy did not flatten commercial distinctions; instead, it sharpened them. Certain roles, particularly those tied to property and oversight, became more central, while others faded into marginal or ad hoc usage.

  1. 1872–1949: Shopkeepers and their Semantic Neighbors

Our collocation evidence and analysis reveal a clear trajectory in the semantic history of shopkeeper-related terminology, that is, the recurrent lexical neighbors that form their semantic environments, across nearly eight decades of Shanghai’s urban development (Fig. 7). Figure 7 aggregates the λ-values for each shopkeeper term and presents their average strength of association across the five periods. Each line therefore represents the overall distinctiveness of a term’s semantic environment—how tightly its recurrent collocates cluster around it—rather than the specific collocates themselves. The vertical movement of the lines reflects changes in the coherence of the semantic field associated with each term: rising trajectories indicate the consolidation of a stable and characteristic lexical neighborhood, while declining or flat trajectories suggest diffuse or unstable usage. The error bars in each period capture the internal variability of each term’s collocational profile, signalling whether its semantic environment is tightly bound or internally heterogeneous. The result is a comparative view of how each shopkeeper label developed its own degree of lexical specialization, revealing, for example, the pronounced consolidation of terms such as 掌櫃, 東家, and 店主 by the Nanjing Decade, in contrast to the weaker or inconsistent profiles of 坊主 or 舖主. What began as a loosely organized and weakly differentiated lexical field progressively transformed into a stratified semantic system, characterized by functional specialization, hierarchical differentiation, and institutional embedding. The evolution of these terms paralleled the broader modernization of the commercial and regulatory landscape in Shanghai, showing how language both reflected and shaped the visibility and roles of commercial actors.

In the late Qing period, shopkeeper-related vocabulary occupied a fluid and undifferentiated semantic space. Terms such as 店主, 掌櫃, 坊主, 舖主, and 行主 appeared in relatively small collocational clusters that revolved around everyday disputes, testimonies, and petty transactions. Over time, however, these labels diverged into distinct semantic trajectories. Some became more clearly associated with property relations, legal authority, and organizational leadership, while others remained tied to local or narrow transactional contexts. This long-term differentiation reflects the increasing complexity of commercial roles in the rapidly expanding urban economy and changing morphology of the city. As the administrative frameworks, property markets, and commercial associations grew more elaborate, the press began to represent shopkeepers through more precise and functionally bounded vocabularies. The result was a shift from a dispersed semantic field to a stratified and functionally divided linguistic system.

One of the most striking long-term developments was the growing dominance of property-related and legal vocabularies. By the 1930s and 1940s, the semantic cores of several key shopkeeper terms we tied to references to: property rights (產權, 原, 業主), lease arrangements and rent disputes (租, 租金, 欠租, 收回), legal intervention and authority (案, 查封, 扣押), and economic regulation (核辦, 批示, 商議). This evolution illustrates the increasing formalization of commercial relations in Shanghai as property markets matured, administrative oversight expanded, and contractual disputes became both more frequent and more formally adjudicated. Shopkeepers were increasingly depicted as actors situated within regulated property systems and legal processes.

Figure 7. Evolution of Collocation Strength (Lambda) by Term Type


Another long-term pattern was the semantic divergence between terms associated with more elite, property-owning, or organizational roles and those tied to small-scale or marginal economic activities. Elite or institutionally embedded terms such as 東家, 掌櫃, 業主  exhibited consistently high lambda values by the 1930s–1940s, reflecting stable collocational environments and clear functional identities. 東家 gravitated toward committee deliberations and collective decision-making, 掌櫃 was associated with legal and bureaucratic vocabularies, and 業主 became strongly tied to property rights and rental conflicts.

Petty retail or local terms like 舖主, 坊主, 行主 followed a different trajectory.
Their lambda values remain low or decline over time, and their networks became fragmented. These terms drifted toward the linguistic margins, indicating reduced semantic coherence and diminished visibility in the press. This divergence reflected more than a linguistic shift: it mirrored the widening economic and institutional asymmetries that defined modern Shanghai. As property markets expanded, municipal regulation intensified, and commercial associations became more formalized, certain categories of commercial actors acquired stable institutional positions. Terms such as 東家, 掌櫃, and 業主 crystallized around the vocabularies of property ownership, legal authority, and committee deliberation. Their semantic coherence corresponds to the emergence of merchants who operated within regulated markets, handled contractual obligations, and participated in formal mechanisms of arbitration or collective decision-making.

By contrast, terms associated with petty retail (舖主, 坊主, 行) gravitated toward increasingly unstable semantic environments. Their fading distinctiveness was not merely a linguistic artifact. It reflected the vulnerability of small-scale retailers who lacked access to the institutional protections afforded by property ownership or guild-like representation. As the local economy polarized, these actors were more exposed to fluctuations in supply, rent increases, and the pressures of wartime requisitioning. Their diminishing semantic visibility in the press paralleled their declining foothold within an urban economy that rewarded more capitalized, legally legible, and organizationally integrated commercial roles.

Collocation analysis does more than describe patterns of lexical behavior. It reveals how the press encoded the shifting social patterns of commercial life. The network structures make visible the thematic neighborhoods surrounding each term, clusters that reflect evolving zones of economic activity, from property management to bureaucratic adjudication or emergency provisioning. Lambda values, in turn, measure the stability and strength of these associations, showing which roles were consistently enmeshed in coherent institutional vocabularies and which drifted toward irregular or crisis-driven contexts.

Conclusion

The semantic evolution of shopkeeper-related terminology in the Shenbao from 1872 to 1949 reveals not only shifting linguistic patterns but a deeper reorganization of Shanghai’s commercial world. Collocations provide a window into the everyday contexts in which merchants and small traders appeared in the press, contexts shaped by disputes, policing, property relations, committee deliberations, and, eventually, wartime emergency. What emerges is a gradual tightening of the semantic field from the late Qing’s dispersed vocabulary of petty conflicts toward the early Republic’s more bureaucratically inflected language, culminating in the Nanjing Decade’s fully institutionalized framing of commercial actors. By the 1930s, the press no longer portrayed shopkeepers simply as participants in local exchange but as actors embedded in a regulated property regime and a maturing legal-administrative order. The disruptions of the Sino–Japanese War fractured this coherence once more, yet even then the press continued to cast certain commercial roles, especially those connected to property ownership and administrative oversight, as central nodes in a strained but still recognizable economic system.

At a broader level, these shifts point to transformations that cannot be reduced to terminology alone. The changing semantic environments of 店主, 掌櫃, 東家, and 業主 reflect the uneven remaking of Shanghai’s economy: the consolidation of property markets, the professionalization of commercial governance, the rise of municipal regulation, and the vulnerabilities of small-scale retail under conditions of inflation, requisitioning, and political instability. In this sense, the language of shopkeeping offers an indirect but revealing proxy for the restructuring of the local economy in a city where legal norms, administrative instruments, and market practices became increasingly intertwined with the consolidation of state power under the Nationalists.

At the same time, the interpretive reach of collocation analysis has clear limits. Collocations illuminate the contexts in which terms appear, but they do not reveal the identities of the actors involved, the institutional networks that connected them, or the broader thematic repertoires that shaped public discourse. They highlight recurrent associations but cannot explain the deployment of terms in news writing, the editorial decisions behind framings, or the competing discourses circulating across different media. Moreover, our analysis is limited to a single newspaper, the Shenbao, while there existed alternative repertoires in other Chinese newspapers or in in the English-language press. To move from semantic environments to social worlds requires additional analytical layers. The next essays in this series, drawing on named entity and institutional network analysis, and on topic modeling, will push further in that direction. These approaches will complement the lexically grounded perspective developed here and offer a more comprehensive account of how commercial actors were represented, regulated, and imagined in modern Shanghai. In this sense, collocation analysis served as a crucial starting point: it mapped the semantic terrain which we plan to examine with a new set of lenses to understand its institutions, actors, and discursive structures. The approach implemented in this essay therefore marks the foundation for a deeper inquiry into the everyday life of shopkeepers in a rapidly transforming city.

  • In this essay, I use indifferently collocation and co-occurrence, even if I mean collocation. Technically in NLP, co-occurrence refers broadly to any two words appearing near each other within a chosen context window—such as a sentence, paragraph, or sliding window—indicating simple proximity without implying a meaningful relationship. Collocation, by contrast, describes pairs or groups of words that occur together more frequently than chance would predict, forming statistically significant and often semantically meaningful combinations (measured through metrics like Mutual Information or t-score). In short, co-occurrence captures contextual neighborhood, while collocation identifies habitual, reinforced, and linguistically salient word pairings.

Methodological Note

The analysis presented in this essay is based on a tokenized version of the corpus of articles extracted from the Shenbao, restricted to items containing shopkeeper-related terminology. The working dataset (shop_bind_tok2) comprises 67,946 rows, each corresponding to a document-level unit (typically an article or advertisement segment) with pre-tokenized text (space-separated tokens) and a document identifier (DocId) that encodes the year of publication. The Chinese text was consistently tokenized using the segmenting model developed in the ENP-China project.

From this base, I used the lexicon of shopkeeper-related terms defined previously, including 坊主, 店主, 棧主, 莊主, 號主, 行主, 鋪主, 館主, 店主婦, 店東, 東家, 東主, 店家, 掌櫃, 商户, and 業主. The script first extracts the publication year from each DocId and assigns documents to five historically meaningful periods: 1872–1894, 1895–1911, 1912–1927, 1928–1937, and 1938–1949. For each period, I subset the corpus to documents that contain at least one of the shopkeeper terms and then built a quanteda corpus* using the pre-tokenized Text field. A large, manually curated stopword list in traditional Chinese, including classical particles, function words, discourse markers, and noise terms specific to the press corpus, was applied to remove high-frequency, low-information items.

Collocations were extracted using the quanteda library, with a window size of 5 tokens and focusing on bigrams and trigrams. For each period, the script computed all 2-gram and 3-gram collocations above a minimum frequency threshold  and then retained only those collocations that contained at least one of the shopkeeper terms. The function returned, for each collocation, its raw count and several association measures, notably lambda and z-scores. In a second filtering step, I restricted the dataset to statistically meaningful collocations by requiring either lambda ≥ 3 or z ≥ 2 and removed duplicates that resulted from overlapping token sequences. For each shopkeeper term and period, I also computed a normalized frequency (counts per 1,000 documents with that term) in order to take into account the differing number of documents across periods.

The outputs for all periods were then aggregated into a single table (all_collocations_combined), which served as the central dataset for subsequent analysis. From this combined table, I produced several derived datasets: (1) a general collocation_summary by period and term (number of collocations, mean lambda, mean count, maximum lambda); (2) a set of “period-specific” collocations, defined as those that appear only in one of the five periods; and (3) a set of “persistent” collocations, defined as those that recur with the same term in three or more periods. These derived tables were exported as CSV files and used both for quantitative inspection and for manual checking of key collocates.

To visualize and interpret the structure of the collocational field, I generated a series of bipartite and term-centered networks for each period. In these networks, shopkeeper terms and their collocates are treated as nodes, and edges are weighted by collocation counts (with a minimum edge weight threshold to avoid noisy links). Graphs were constructed using igraph and plotted with ggraph (Fruchterman–Reingold layout), producing a set of periodized network visualizations that highlight clustering, centrality, and fragmentation. In parallel, I plotted the evolution of lambda values over time. One set of graphs shows the average lambda by period and shopkeeper term, while another tracks the lambda trajectories of the strongest collocations across periods. A faceted version focuses on the top collocates for each term, providing a compact view of how the semantic environment of each label stabilizes or shifts across the five periods.

Beyond these core steps, the script also implements several exploratory analyses designed to probe the temporal dynamics of collocational change in more detail. First, I constructed “semantic vectors” for each shopkeeper term and period by pivoting the lambda values of their top collocations into a wide matrix (terms × collocates), and then measured cosine similarity between consecutive periods. This procedure yields a simple index of semantic stability or disruption across time for each term and helps identify major shifts in their collocational profiles. Second, the script includes a temporal trend module that calculates Spearman correlations between period indices and lambda values for collocations that occur in at least three periods, thereby highlighting collocates whose association with a given term strengthens or weakens systematically over time. Third, a rule-based categorization of collocates into semantic or “discourse function” classes (transactional, regulatory, spatial, temporal, etc.) provides a coarse-grained way to examine the changing balance of commercial, legal, social, and locative vocabularies across periods. Finally, the script contains experimental components for topic modeling (LDA on pseudo-documents built from collocations), derivation of collocation embeddings, and event-like detection of abrupt lambda shifts. These additional modules were used primarily as heuristic devices to test the robustness of the main patterns and to explore possible extensions for future work, not as direct sources of the arguments presented in this essay.

The methodological choices outlined above are intended to provide a transparent and replicable framework for analyzing semantic environments around shopkeeper terminology. They also lay the groundwork for subsequent studies that will move from collocations to named entities, institutional networks, and topic modeling in order to reconstruct more directly the social and institutional ecologies in which these commercial actors operated.
*  A quanteda corpus is a structured R object that stores a collection of texts together with their metadata (e.g. date, source, document ID), in a format optimized for text analysis (tokenization, collocations, document-term matrices, etc.).

Statement on AI-Assisted Development

This study benefited from AI-assisted development by Claude Sonnet 4.5 (Anthropic) in an iterative dialogue process that made substantial contributions to both the technical implementation and analytical interpretation of the collocation study. Claude Sonnet 4.5 played a crucial role in developing and refining the R scripts for collocation analysis, contributing sophisticated code solutions for several complex challenges: First, in implementing comprehensive collocation extraction methods using multiple statistical measures (frequency, MI, t-score, log-likelihood) within the quanteda framework.  Second, in creating robust data processing pipelines that handled the specific challenges of historical Chinese texts, including compound term identification and proper encoding of Chinese characters. This included developing custom functions for filtering noise while preserving meaningful historical language variations. Third, in implementing diachronic analysis capabilities that allowed for period-based comparisons, including code for normalizing frequencies across subcorpora of different sizes and visualizing temporal patterns in collocational relationships. Fourth, in developing comprehensive output formats and visualization methods that facilitated both quantitative analysis and qualitative interpretation, including sorted tables, statistical summaries, and export functions for further analysis. All code generated through this process was tested, validated, and adapted to the specific requirements of the historical corpus, exemplifying how AI can accelerate and enhance digital humanities methodology while maintaining scholarly rigor. The script is available on GitHub.

The present study also employed ChatGPT assistance (5.1) in two complementary aspects of the research process: the interpretation of pre-existing computational outputs and the refinement of written exposition. All datasets, statistical measures, and visualizations, including collocation tables, temporal distributions, lambda-based distinctiveness scores, and the various network graphs, were generated independently by the author using established computational methods. ChatGPT support intervened only after these materials had been produced, serving as an interpretive aid in organizing and synthesizing patterns already identified in the data. In this capacity, the AI contributed to the analysis of datasets and network visualizations, helping to articulate how semantic clusters emerged and how these patterns varied across the five historical periods analyzed. Similar assistance was provided in interpreting the lambda evolution graphs, which required translating statistical distinctiveness into historically meaningful trajectories of consolidation, divergence, and semantic instability. In addition to this analytical support, ChatGPT played a secondary role in shaping the prose of the essay. It helped transform analytical notes and visual observations into coherent narrative sections. ChatGPT also contributed to stylistic refinement by drafting preliminary formulations that were subsequently revised, expanded, or reworked by the author to ensure historical accuracy. These interventions were particularly useful in sections that required integrating multiple computational outputs such as collocations, networks, and lambda values into a unified interpretive argument. Throughout the process, ChatGPT functioned strictly as an auxiliary tool: it did not generate data, compute metrics, or independently produce historical claims. Its role was limited to assisting in the interpretation of outputs derived from the author’s own computational workflow and to supporting the development of polished academic prose. The empirical findings, periodization, and historiographical interpretations presented in this study remain entirely the result of the author’s independent research and critical judgment.

[1] Lambda (λ) is a statistical measure of how strongly a word is associated with its most characteristic collocates relative to the corpus as a whole. Unlike simple frequency counts, which only show how often a term appears, lambda captures the extent to which a word draws specific vocabulary into its semantic orbit. High lambda values signal a stable and distinctive contextual environment, while low values indicate diffuse or inconsistent usage. In this essay, lambda serves to quantify the degree of semantic specialization or fragmentation surrounding each shopkeeper term over time.

All figures in large size

Figure 1.

Shopkeeper Collocation Network (left) in 1872-1894


Key Collocation Network (right) in 1872-1894

Figure 2.

Shopkeeper Collocation Network (left) in 1895–1911


Key Collocation Network (right) in 1895–1911

Figure 3.

Shopkeeper Collocation Network (left) in 1912–1927


Key Collocation Network (right) in 1912–1927

Figure 4.

Shopkeeper Collocation Network (left) in 1928–1937


Key Collocation Network (right) in 1928–1937

Figure 5. Evolution of Collocation Strength (Lambda) by Term Type

Figure 6.

Shopkeeper Collocation Network (left) in 1938–1949


Key Collocation Network (right) in 1938–1949

Figure 7. Evolution of Collocation Strength (Lambda) by Term Type


OpenEdition suggests that you cite this post as follows:
Christian Henriot (November 20, 2025). Behind the Counter: Collocation Patterns in the Shopkeeper Lexicon, 1872–1949. Blog Virtual Shanghai. Retrieved May 14, 2026 from https://doi.org/10.58079/156ih


You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.