Discursive use of stability in New York Times’ coverage of China: a sentiment analysis approach

The paper investigates how the political keyword “stability” is used in The New York Times’ reporting on China from 1980 to 2020. Grounded in Critical Discourse Analysis (CDA)—particularly the Discourse-Historical Approach (DHA)—and supported by sentiment analysis, the study asks: (1) What sentiments (positive, negative, or neutral) toward China did The New York Times construct via the use of “stability” over 41 years? (2) What facets did the paper focus on in relation to China’s stability? It also asks why, by relating the discursive use of “stability” to dominant ideologies in the United States. The context is China’s longstanding emphasis on stability, rooted in traditional political philosophy and reiterated in modern governance (e.g., Deng Xiaoping’s maxim “Stability is of paramount importance”). The study’s purpose is to understand how a major US newspaper from an ideologically different context discursively constructs “stability” concerning China and how this reflects US-centric perspectives and dominant ideologies. The significance lies in extending CDA with corpus and sentiment methods to reveal long-term discursive trends and implicit evaluations in international news coverage.

The paper reviews scholarship on US media representations of China, noting sustained analytical interest and generally negative portrayals with some period-specific variation. Studies show US media frequently frame China negatively on political/ideological dimensions but more positively on economic aspects (e.g., Peng 2004; Liss 2003; Tang 2021). Diachronic analyses are fewer but reveal stable attention to political, diplomatic, and economic topics (Yan 1998). The review also discusses the methodological synergy of CDA and corpus linguistics (Hardt-Mautner 1995; Baker et al. 2008), emphasizing how corpus tools help verify CDA claims. It then surveys sentiment analysis methods (document-, sentence-, and aspect-level), contrasting machine-learning with lexicon-based approaches. Prior applications of sentiment analysis to news (e.g., NYT coverage of Russia/Islam; climate change perceptions; financial journalism during COVID-19) demonstrate its utility. The authors justify a lexicon-based approach for handling large text corpora efficiently while accounting for valence shifters, and position sentiment analysis as complementary to CDA in uncovering evaluative patterns.

Design: Descriptive, diachronic corpus-assisted CDA integrating quantitative (frequency, collocation, sentiment) and qualitative co-textual analysis. Data sources and corpus construction: News reports on China from The New York Times retrieved via LexisNexis for 1980–2020 using search terms “China,” “Sino,” and “Chinese,” ranked by “Relevance.” Excluded document types: Letter, Op-ed, News Analysis, Editorial. Using Entman’s framing theory, only reports containing at least two components (problem definition, causal interpretation, moral evaluation, treatment/recommendation) were retained. Manual cleaning removed dates, bylines, copyright. Corpus scale (1980–2020): 34,948 texts; 31,451,277 tokens; 311,273 types. Periods defined by sociopolitical context: P1 (1980–1990), P2 (1991–2000), P3 (2001–2010), P4 (2011–2020). Period-level sizes: P1: 5,201 texts; 3,859,191 tokens; 61,260 types. P2: 5,018; 4,269,135; 57,895. P3: 9,899; 8,683,077; 86,626. P4: 14,830; 14,639,824; 105,492. Annotation of target term: Using AntConc 3.5.8 (concordance and file view), instances of “stability” were categorized by co-text: (1) non-quotation related to China (analyzed); (2) quotation related to China (tagged “stabilityI,” excluded as source’s voice); (3) unrelated to China (tagged “stabilityII,” excluded). Raw and standardized frequencies of analyzed “stability” per period: P1 19 (49.23 per 10M), P2 24 (56.22), P3 90 (103.65), P4 147 (100.41); Total 280 (309.51). Sentiment analysis: Sentence-level polarity computed with R package “sentimentr” (Rinker, 2018) using the 11,709-word Jockers & Rinker lexicon, which incorporates valence shifters (negators, amplifiers, de-amplifiers, adversative conjunctions). Sentiment values aggregated per period. Collocation analysis: For non-quotation “stability” (China-related), strongest adjective and noun collocates per period were extracted using AntConc with log-likelihood statistic. Stopwords removed with tidytext (1,149 English stopwords). Window size: ±5; minimum collocate frequency: 3. Collocate lists guided selection of sentences containing “stability” + strong collocates for sentiment and thematic analysis. Selected sentence counts: P1: 19 sentences (459 tokens; 249 types); P2: 24 (761; 374); P3: 89 (2,849; 1,025); P4: 146 (4,596; 1,438); Total: 278 sentences (8,665 tokens; 3,086 types). Qualitative co-text analysis: Sentences classified by polarity (positive/neutral/negative), then thematically coded (e.g., sociopolitical, economic/financial, territorial, human rights, Hong Kong) using DHA’s nomination and predication strategies to interpret framing and evaluation patterns across periods.

Overall sentiment trends: Sentence-level sentiment values for non-quotation “stability” contexts were: Period 1 (1980–1990): +0.09 (only positive period); Period 2 (1991–2000): −0.19 (lowest); Period 3 (2001–2010): −0.06; Period 4 (2011–2020): −0.02. This indicates a generally negative stance from 1990 onward, aligning with broader US–China relations and US foreign policy shifts. Polarity distributions (N=278 sentences): Overall: Positive 140 (50.4%), Neutral 9 (3.2%), Negative 129 (46.4%). By period: P1: Positive 13 (68.4%), Negative 6 (31.6%); P2: Positive 6 (25%), Negative 18 (75%); P3: Positive 36 (40.4%), Neutral 4 (4.5%), Negative 49 (50.1%); P4: Positive 85 (58.2%), Neutral 5 (3.4%), Negative 56 (38.4%). Frequency of target term: Non-quotation, China-related “stability” occurrences increased over time (raw counts per period: 19, 24, 90, 147; standardized per 10M words: 49.23, 56.22, 103.65, 100.41), with a sharp rise from P2 to P3. Collocate patterns (top items by log-likelihood): - P1: political, prosperity, economic, peace, Kong, Hong (political/economic stability linked to Hong Kong; regional peace/prosperity). - P2: social, political, peace, unemployment, period, party, financial (emphasis on social issues, unemployment, CPC, human rights). - P3: social, political, country, party, regional, economic, financial (focus on CPC-led political stability; rising attention to economic/financial stability during global financial crisis). - P4: social, maintenance, financial, prosperity, political, economic, government (salience of “stability maintenance” mechanisms; government as key actor; continued economic/financial dimensions). Thematic emphases by period: - P1: Positive focus on China’s economy, regional development, and world peace; negatives tied to political instability concerns, Hong Kong’s uncertain future, and repercussions of 1989 Tiananmen (e.g., credit rating downgrade). - P2: Predominantly negative (75%): unemployment/social stability threats; human rights/religion (Falun Gong, dissidents, restrictions on expression/publishing); leadership transition after Deng Xiaoping’s death; some positives recognizing China’s role in Asian/international stability. - P3: Mixed but net negative sentiment value: positives on financial/economic stability and social order; negatives on press freedom, territorial disputes, corruption, widening wealth gap, protests (e.g., constraints framed as enforcing strict social stability). - P4: More positive sentences but sentiment value still slightly negative: economic/financial management and currency strength linked to political stability; negatives on cybersecurity controls, COVID-19 responses, Hong Kong unrest; portrayal of government “stability maintenance” and clampdowns. Interpretive highlights: The NYT recurrently frames “stability” politically (persistent “political stability” collocate), with negative sentiment clustering around sociopolitical and territorial issues, and positive sentiment around economic/financial stability and contributions to peace/prosperity. The discourse is characterized as US-centric, aligning with dominant ideologies and foreign policy narratives.

The findings address the research questions by showing that NYT’s discourse on China’s “stability” is predominantly negative outside the 1980s, with sentiment patterns tracking major shifts in US–China relations and US foreign policy. The newspaper consistently foregrounds political dimensions of stability, often linking it to CPC rule and social control, which contributes to a critical framing of sociopolitical and territorial issues (human rights, censorship, protests, Hong Kong, maritime disputes). Conversely, economic and financial stability are portrayed more positively, acknowledging China’s performance and contributions, particularly around the 2007–2008 financial crisis. This asymmetry suggests a politically self-interested, US-centric perspective and reflects enduring dominant ideologies within American media discourse. The persistence of negative political/ideological framing alongside positive economic framing mirrors earlier cross-newspaper findings (e.g., Peng 2004), indicating limited change over decades. Thus, the discourse of “stability” functions as a lens for othering and for legitimizing US positions in regional security narratives, while recognizing China’s economic weight. Methodologically, integrating sentiment analysis with DHA helps uncover evaluative patterns at scale and contextualize them historically, reinforcing the interpretive claims of CDA with quantitative evidence.

The study demonstrates that The New York Times’ use of the political keyword “stability” in China-related reporting (1980–2020) is largely negative in sociopolitical/territorial contexts and positive in economic/financial contexts. This pattern aligns with long-standing US media ideologies and foreign policy trajectories, revealing a predominantly political, US-centric framing of China’s stability. Contributions include: (1) extending corpus-assisted CDA with sentiment analysis to diachronically map evaluative stances; (2) offering empirical evidence on how political keywords shape national images in international news discourse across ideologically diverse societies; and (3) showing sentiment analysis to be time-efficient, verifiable, and sufficiently accurate for supporting CDA in large corpora. Future research directions include comparative analyses across outlets and periods and developing domain-specific sentiment lexicons tailored to news discourse to better capture nuanced evaluative language.

Potential researcher subjectivity in categorizing “stability” into three types and in sentence-level thematic coding may affect interpretations. The use of cross-domain sentiment lexicons may miss domain- and genre-specific semantic features salient to news discourse, given contextual variation in language use. Developing specialized, news-specific sentiment lexicons could improve detection of evaluative nuances in future work.