Modelability of WAR metaphors across time in cross-national COVID-19 news translation: An insight into ideology manipulation

Abstract

Previous studies have compared Covid metaphors across languages and national contexts, but seldom focus on the translation issue where news narratives of the same event may be different when translated for different readers. Another unexplored question is whether, and how, successive discursive observations across time in such narratives are related. To fill these gaps, this study employs the Box-Jenkins time series analysis (TSA) method to investigate whether and how WAR metaphor usage in Chinese-English COVID-19 news reports (source articles and their translations) can be fitted with ARIMA (Autoregressive Integrated Moving Average) models. These reports come from three different sources across the year 2020: the Chinese Global Times (GT), the American New York Times (NYT) and the British The Economist (TE). Results show that WAR metaphors in the source news of GT and TE are modelable with an autoregressive and moving average model. However, no models were found to fit their translation counterparts. By contrast, WAR metaphors in both NYT’s source and translated news were not modelable. These differences are further qualitatively analyzed with examples in context. The study may contribute to the existing debates on WAR frames in COVID-19 discourse by adding a translation and TSA angle.

1. Introduction

Metaphor analysis is an important investigative tool in news discourse analysis. This is partly because news is “a very metaphorical register” (Krennmayr, 2011, p. 131). In fact, scholars argue that metaphors frame social realities through a hiding-and-highlighting process (Semino et al., 2018). They therefore link metaphorical framing with ideology manipulation in news reports of different world events, especially political events that invite conflicted opinions. Examples include the Hungarian refugee crisis (Benczes and Ságvári, 2022) and Ukrainian political crisis (Baysha and Hallahan, 2004). Following Van Dijk (2006a), we view ideology as “systems of ideas” consisting of “social representations that define the social identity of a group, that is, its shared beliefs about its fundamental conditions and ways of existence and reproduction” (p. 116). Manipulation, on the other hand, is influencing “the beliefs of the recipients, and indirectly a control of the actions of recipients based on such manipulated beliefs” (Van Dijk, 2006b, p. 362). In metaphor analyses, “ideology manipulation” therefore refers to the use of metaphors to influence peoples’ beliefs, understandings and experiences of events, thus potentially shaping their subsequent relevant actions. In the Hungarian refugee crisis example above, refugees were framed with various negatively connotated metaphors on news media, resulting in less welcoming attitudes among recipients.

The COVID-19 pandemic, as a global health crisis as well as a political event (Altiparmakis et al., 2021), has also invited a plethora of metaphor analyses in the news context1 (see Liu and Li, 2022; Charteris-Black, 2021, de Saint et al., 2021, Semino, 2021, Taylor and Kidgell, 2021). These studies discuss major metaphors used in news, their framing effects, and perceived appropriateness in a pandemic situation. They generally find that WAR metaphors are dominant in COVID-19 news reports. However, the appropriateness of WAR metaphors is debatable. Semino (2021) argues that WAR metaphors may generate counterproductive framing effects, especially when framing COVID patients as fighters who might feel guilty if they could not recover. On the other hand, Charteris-Black (2021) holds that “there is nothing inherently wrong or immoral about ‘war’ metaphors” as they can “encourage unity and co-ordination” (p. 35), but the political intentions of using WAR metaphors should be considered. For instance, Benzi and Novarese (2022) conclude from a general survey of the literature that governments may take advantage of WAR metaphors to foster obedience and thereby jeopardize democracy.

These studies have fully acknowledged that metaphor is “an eminent aspect of news discourse in which hidden opinions or ideologies may surface” (Van Dijk, 1988, p. 177). Based on this assumption, scholars have conducted general cross-national or cross-linguistic comparisons of WAR metaphors (e.g., Brugman et al., 2022). Nonetheless, two aspects are still perceptibly underexplored. The first is to inspect ostensible cross-linguistic and cross-national differences from the perspective of news translation (except Liu and Li, 2022). The second is to explore how time influences metaphorical framing, or how successive WAR metaphor usages may be correlated with each other in the source and translated news.2 For the former, comparing how newspapers from different countries metaphorically frame COVID-19 in different languages can unveil the complicated nature of news discourse in mediating ideologies for target readers who have different social, cultural and political backgrounds. For the latter, analyzing how past discursive practice (i.e., metaphor usage) impacts the present can provide a more dynamic picture by capturing “not just the way things are, but the way things move”3 (Tay, 2022b, p. 52). To fill these gaps, the current study utilizes the Box-Jenkins time series analysis (TSA) with ARIMA (Autoregressive Integrated Moving Average) models to investigate the modelability of WAR metaphors across time in the Chinese and English COVID-19 reports of three different newspapers: The Global Times from China, The New York Times from the US and The Economist from the UK. The primary objective is to investigate whether and how WAR metaphors are modelable across time in each newspaper. The second objective is to explore whether such modelability shows cross-linguistic (source news vs translated news) and cross-national (China vs the US vs the UK) differences, and to explain these differences from ideological and policy perspectives. By “modeable”, we mean that WAR metaphor usage at any time is predictable from its usage in the past periods, which may suggest some form of strategic discourse construction across time that can be further analyzed in context. On the other hand, a “non-modelable” WAR metaphor series means that present and past WAR metaphor usages are statistically unrelated, and usage across time appears to be random. Modelability of WAR metaphors can shed light on ideology manipulation by revealing how different newspapers use the same metaphor in two broadly different ways (modelable versus non-modelable) to influence readers’ beliefs, understandings and experiences of the COVID-19 pandemic and shape their actions. These actions may include their willingness to practice social distancing, take vaccines and wear face masks.

In what follows, we will first introduce the similarities and differences among the three newspapers and Covid policies in China, the US and the UK. Then, we will discuss time series analysis and its potential application in discourse analyses of ideology manipulation in news translation. After presenting our data and the Box-Jenkins TSA method, we will discuss the results and implications, before summarizing the major findings, limitations, and offering directions for future research.

2. Newspapers of interest and COVID policies in China, the US and the UK

The Global Times (GT), is a state-owned Chinese newspaper and has a clear partisanship with the Communist Party of China. It is regarded as “China’s Fox News” (Brady, 2015) and a “foreign mission” by the United States Department of State (Ruwitch and Kelemen, 2020). GT has an English official website (https://www.globaltimes.cn) where most news articles are directly translated from its Chinese official website (https://www.huanqiu.com). GT has been reproached by some Western scholars for spreading unfounded coronavirus theories and misinformation (Molter and DiResta, 2020).

The New York Times (NYT) is a private American daily newspaper and displays an ideological affinity with the American Democratic Party (Puglisi, 2011). NYT publishes Chinese reports on its official Chinese website, cn.nytimes.com, which are translations of NYT’s English reports on https://www.nytimes.com. NYT has been intermittently censored and blocked in China’s Mainland due to its negative reports on the country. NYT reporters were even expelled from China’s Mainland, Hong Kong, and Macau in mid-March 2020, as the Chinese government’s countermeasure to the United States Department of State’s “foreign mission” of censorship, which targeted five Chinese news agencies (The Information Department, 2020).

The Economist (TE) is a private British newspaper covering a wide range of topics including politics, economy, technology, and social issues. Like the NYT, it has been intermittently censored and blocked in China due to its negative portrayal of the Chinese politics (Horwitz, 2016). TE launched its English-Chinese bilingual mobile app in 2015, The Economist Global Business Review, which was exempted from the Chinese government’s censorship. News articles on this mobile app are the Chinese translations of the original English news articles on TE’s official website (https://www.economist.com/).

It is clear that the three newspapers have contrasting ideologies, each of which shows a general affinity to its country of origin: pro-China versus pro-West.4 In fact, the three newspapers’ disparate metaphorical frames of the COVID-19 pandemic have already been documented in Liu and Li, 2022, Liu and Li, 2023 recent research. Nevertheless, there is still a lack of inquiry into the time variable, i.e., whether previous WAR metaphor usage impacts the present, which may afford novel analytical perspectives on the ideological functions of metaphor. For example, it is reasonable to hypothesize that WAR metaphors in the source articles and in the translated articles of all three newspapers should show similar usage frequencies across time, because translations are not expected to differ significantly from their sources. If the results go against this hypothesis, they imply that the three newspapers may be manipulating the metaphorical frames and therefore the ideologies for different target readers.

Since news reportage is also impacted by each country’s Covid policies, it is necessary to introduce the Chinese, American and British governments’ policy responses to COVID-19. In early 2020, President Xi Jinping declared that this pandemic was not just a crisis but also a “people’s war” that China must fight (Kania & McCaslin, 2020). Accordingly, the Chinese government has practiced a zero-Covid policy in Chinese mainland since early 2020 through measures such as contact tracing, mass testing, border quarantine and lockdowns. For instance, Wuhan City was in lockdown from January 28 to April 8, 2020.5 Since August 2021, China has disposed of its zero-Covid policy and practiced a “dynamic clearing” policy of “external input prevention and internal rebound prevention” (Wang and Huang, 2022). More recently, in December 2022, China has loosened its nationwide Covid restrictions, such as requiring COVID-19 testing to be conducted on a voluntary basis and banning temporary lockdowns.6 As of December, 2022, China has over 9 million confirmed cases and over 30, 000 deaths.

Unlike China, the American and the British governments imposed much fewer Covid restrictions, though temporary lockdowns were also ordered in both countries.7 A herd immunity policy was discussed in the American and British communities, especially on the news media (Colfer, 2020, Gumel et al., 2021), but relevant authorities clarified that no official herd immunity policy was ever applied.8 As of December, 2022, there are 98 million confirmed cases and over 1 million deaths in the US,9 and over 24 million confirmed cases and over 197,723 deaths in the UK.10 All three countries encourage their citizens to get vaccinated and vaccine-related regulations and laws were in place at least in 2020, such as issuance of “vaccine passes”. As of December, 2022, around 90% of the Chinese people, and 80% of the American and the British people were vaccinated.11

3. Linking time series analysis with ideology manipulation in discourse

A time series is a set of consecutive measurements of a random variable usually made at equally spaced time intervals, and typical examples in the physical and social world include stock prices, heartbeat and rainfall (Tay, 2019; Brockwell and Davis, 2009). Time series analysis (TSA), most commonly used in Finance and Engineering, mathematically represents a given series using only aspects of its past values, in order to predict or forecast future values. In TSA, time is the (implicit) independent variable with the time intervals as its different levels, and the phenomenon observed at each time interval of the series is the dependent variable. In fact, the major difference between TSA and standard regression analysis lies in the emphasis on autocorrelations, i.e. how successive values are correlated with one another. The Box-Jenkins method (Box et al., 2015) that forms the mathematical equation using ARIMA (Autoregressive Integrated Moving Average) models is commonly used among many TSA methods (see Section 4.2 for a detailed elaboration on this method).

Discourse data can also be a time series. Hence, discourse can be examined using various time series analysis (TSA) methods with an aim to explore how previous discursive practice like metaphor impacts the present. Recently, TSA has been introduced to Critical Discourse Analysis (CDA) where power relations and ideology are the main research foci (see Tay, 2019 for the rationale for this application). A number of example studies have been provided to showcase the application of TSA in varying contexts for different purposes. For instance, TSA has been applied to analyze how lexical choices across time in news serve to construe the 2019 Hong Kong protest movement (Tay, 2021a), how a Youtuber’s linguistic choices across time construct her identity as an “amateur expert” (Tay, 2021a), and how metaphors were used in psychotherapy sessions between counselors and patients (Tay, 2017). Thematic keywords like “democracy” in press (Tay, 2019) and linguistic choices in COVID-19 press conferences (Tay, 2022a) have also been analyzed using TSA, which provides replicable and systematic ways of analyzing opinions and ideologies in discourse and thereby complements the more usual content or thematic analysis. Like any other methods, TSA can serve different research purposes. While the above-mentioned studies simply use TSA to investigate the discourse construction of certain social phenomena on social media platforms or press conferences, our research firstly attempts to apply TSA to investigate a translation issue. Moreover, since our focus is specifically on WAR metaphors, we may make novel contributions to the current debates on the use of WAR metaphors in crisis situations, from a translation perspective as well as a modelability perspective.

Specifically, the current study aims to investigate the modelability of WAR metaphors12 in COVID-19 news across time in three ideologically contrasting newspapers, and to see whether the resulting models differ in their source and translated news. Hence, the metaphors are analyzed both cross-nationally and cross-linguistically. The notion of modelability is also defined here at two levels: i) whether patterns exist in a series of WAR metaphor frequencies such that a time series model can be fitted to represent present use in terms of past use; ii) if yes, the specific details of the model and their implications.

Presumably, as the evolution of the COVID-19 pandemic is highly unpredictable, the use of WAR metaphors in news reports is hypothesized to fluctuate randomly across time, perhaps with sudden rises when mortality rates are high and drops when effective vaccines were used. Since the three newspapers are responding to the same unpredictable global event,13 their WAR metaphor usages should likewise be unpredictable and unable to be fitted with an ARIMA model (i.e., the hypothesized universal non-modelability of WAR metaphor usage). Meanwhile, as stated in Section 2, the source articles and the translated articles are expected to share this non-modelability as news translation is supposed to be trustworthy and transparent. Any modeling results that go against the above assumptions might then stem from deliberate ideology manipulation, which needs to be further analyzed in context. In a nutshell, examining the modelability of WAR metaphors in the translation of COVID-19 news in different national contexts can reveal whether and how the three ideologically contrasting newspapers construct this COVID-19 “war” differently for different target readers, thereby shaping their beliefs, understandings and experiences of the COVID-19 pandemic and potentially impacting their responses to relevant government policies.

4. Data and methods

4.1. Research data

All Chinese and English opinion articles14 on the topic of COVID-19 in 2020 across the Global Times, The New York Times and The Economist were collected (See Table 1 below for the data description). Since the corpus is too large (consisting of around half a million words) for a manual analysis, the study adopts a corpus-assisted discourse analysis: generating concordance lines based on COVID-19-related keywords15 and then conducting a word-by-word analysis for metaphors on each line. MIPVU (Metaphor Identification Procedure Vrije Universiteit, for a step-by-step guide, see Steen et al., 2010, Chapter 2) is used to identify the metaphoricity of words and SDVP (Source Domain Verification Procedure, for a step-by-step guide, see Ahrens and Jiang, 2020, p. 47, Fig. 1 ) is utilized to verify WAR domains. A linguist who is native Chinese but with previous experience of metaphor analysis in both Chinese and English languages coded and re-coded the datasets, and cases of ambiguity were resolved through regular discussions with other metaphor experts.16 Overall, there are 1,442 instances of WAR metaphors across the three newspapers’ bilingual opinion articles, and they scatter disproportionately across the six sub-corpora: 646 instances in the Chinese GT (series 1), 405 instances in the English GT (series 2), 93 instances in Chinese NYT series (series 3), 92 instances in the English NYT series (series 4), 115 instances in the Chinese TE (series 5) and 91 instances in the English TE (series 6). They are categorized into the 53 weekly time intervals in 2020. As the current study employs the concept of “COVID weeks”, weeks that do not have COVID-19 reports are excluded.17 In the end, there are 30 COVID weeks in series 1 and 2, 27 COVID weeks in series 3 and 4, 43 COVID weeks in series 5 and 6. Normalized frequency (=$\frac{\mathit{Numberofmetaphoricalwordseachweek}}{\mathit{Totalnumberofwordseachweek}}$) is used in the following time series analyses. Since the modelability of each series is analyzed separately, the disproportionate nature of WAR metaphors in each series will not affect the final result.

Table 1.

Data description.

	Global Times (GT)		The New York Times (NYT)		The Economist (TE)
Language	Chinese	English	Chinese	English	Chinese	English
Series No.	Series 1	Series 2	Series 3	Series 4	Series 5	Series 6
No. of articles	85	85	69	69	109	109
No. of words	55,593	56,494	75,195	80,738	112,596	118,902
Total words	499,518
WAR metaphor freq.	646	405	93	92	115	91
No. of COVID weeks	30	30	27	27	43	43

Fig. 1.

Normalized frequency of WAR metaphors across COVID weeks in newspapers.

4.2. The Box-Jenkins TSA method

As stated in Section 3, a common way to conduct TSA is using the Box-Jenkins method that identifies candidate ARIMA (Autoregressive Integrated Moving Average) models (Box et al., 2015). An ARIMA model consists of three elements: Autoregression (AR), which refers to a model that shows a changing variable (WAR metaphor usage in this case) regresses on its own prior values; Integrated (I), which is assigned with a value only when the raw data is non-stationary but transformed as stationary using a differencing technique (in our study, all raw data are stationary); and Moving average (MA), which is a model that uses past error terms (the distance between the actual value and the predicted value) to predict the present.

Tay (2022b) introduced the Box-Jenkins method using Python step by step, which was adopted in studies mentioned in Section 3 (Tay, 2019, Tay, 2021a, Tay, 2021b, Tay, 2022a). This method identifies candidate ARIMA models by observing the calculated (partial) autocorrelations (namely ACF and PACF) (see Tay, 2022b, p. 57, Figure 2.4 for the step-by-step guide) or by auto-ARIMA function. The first method requires the analyst to pick a candidate ARIMA model based on an observation of significant spikes in (P)ACF figures. Alternatively, auto-ARIMA selects the most optimal ARIMA model from a large pool of candidate models by automatically calculating the AIC18 value of each model. Previous studies use the manual observation method, but this method is less straightforward when the (P)ACF spikes are not in consecutive order. Since (P)ACF spikes in the current research are mostly not in consecutive order (as shown in Fig. 1, Section 5.1), we decide to use the auto-ARIMA approach implemented in Python Version 3.10.4.19

5. Results

5.1. Time series plots and (P)ACF calculation

Fig. 1 below displays the time series plots of WAR metaphor frequency (y-axis) across COVID weeks (x-axis)20 in 2020 in the Chinese and English versions of GT, NYT and TE. A general observation of the “shape” of each series reveals that the Chinese and English versions of each newspaper share quite similar patterns in describing COVID-19 as a “war”. Nonetheless, exceptions exist either in the first few weeks (GT) or to the end of the year (NYT and TE). In other words, these newspapers sometimes narrate the COVID-19 differently for different target readers, though they overall tend to keep a similar tone in the source and translated news.21 Moreover, in both GT and TE, WAR metaphors are obviously more frequently used in the Chinese news whereas the NYT does not see much difference in this regard. In fact, the NYT also seems to differ from the GT and TE in that it does not use as many WAR metaphors as the other two newspapers early in the year, which may point to the NYT’s lesser degree of concern about the coronavirus’s destructive impact in the beginning or its opposition to (former) President Trump’s war mongering talk about Covid at that time. Meanwhile, GT and TE do not share the same patterns across time: the frequency in TE is kept low in several consecutive weeks in the middle of the year whereas more rapid changes between consecutive weeks are witnessed in the GT.

In summary, each series is fluctuating in nature, and no obvious long-term trend is observable anywhere. This implies that linear modeling methods are likely to be ineffective. Instead, more localized patterns of fluctuations should be investigated in terms of how they might contribute to a more dynamic narrative of COVID-19 in each series. Fig. 2 visualizes the autocorrelation functions (y-axis) from 0 to 12 lags (x-axis), which provides clues on how values 0, 1, 2…12 intervals apart are positively or negatively correlated with one another. Spikes beyond the blue bands mean that the corresponding (P)ACF values are statistically different from zero at the 95% confidence level, which reveals how consecutive series values correlate with one another and can be used to identify candidate models. Although all series except series 5 have significant spikes (i.e., certain autocorrelations between consecutive series values exist), the spikes are not in consecutive order and therefore identifying ARIMA models by simply observing the spikes (see Tay, 2022b, p. 61) is less straightforward. Instead, the auto-ARIMA feature in Python is used to generate the most optimal model for each series based on the criterion of minimum AIC.

Fig. 2.

Autocorrelation functions of WAR metaphors across three bilingual newspapers.

5.2. Time series modeling outcomes

Fig. 3 gives an example of how auto-ARIMA selects the “best model” for series 1 among a variety of options. Results show that series 1 (Chinese GT) is fitted with an ARIMA (5, 0, 0) model, or simply AR(5) model, and series 6 (English TE) is fitted with an ARIMA (0, 0, 1) model, or simply MA(1) model, whereas the rest are white noise series whose values are randomly distributed across time and therefore no optimal models are generated (see Table 2 below for a summary). The results are surprising for at least two reasons: first, all three newspapers were reporting the same unpredictable global health crisis (as stated in Section 3), so all WAR metaphor series should be non-modelable. The modelable WAR metaphor use in the source news of GT and TE may suggest certain deliberate discourse construction strategy; second, Chinese and English articles correspond to each other, with one being the source article and the other being its translation, and thus modelability of WAR metaphors in the two languages are not supposed to differ significantly. The results instead show that both GT and TE have their source news modelable, but translated news non-modelable.

Fig. 3.

Auto-ARIMA results for series 1.

Table 2.

Summary of time series modeling results.

	Source news	Translated news
Global Times	ARIMA (5,0,0)	No model
The New York Times	No model	No model
The Economist	ARIMA (0,0,1)	No model

To investigate whether the modeling patterns are significantly correlated with the severity of the pandemic, Figs. 4 and 5 22 are plotted below to compare WAR metaphor frequency and the number of Covid infections across time in the two modelable series: Chinese Global Times and English The Economist.23

Fig. 4.

Confirmed cases in China and WAR metaphor use in Chinese GT across time.

Fig. 5.

Confirmed cases in the UK and WAR metaphor use in English TE across time.

By visual inspection, it seems that in both Chinese Global Times and English The Economist, the ups and downs in WAR metaphor series do not correspond to the Covid infections in China and the UK.24 Pearson correlation coefficients were further computed to assess the relationship between infection rates and WAR metaphor frequency. Results show that there was no correlation between the two variables in both Chinese GT and English TE (r = -0.150, N = 30, p =.429; r = -0.237, N = 43, p =.126). In other words, the modeling patterns in the GT and TE source news are more likely to result from certain deliberate discourse strategies that warrant further investigation below.

6. Discussion

6.1. The Global Times: AR(5) model vs Random fluctuation

The AR(5) model for WAR metaphor usage in Global Times’s Chinese source news is formally expressed as y_t = 0.0113 $-$ 0.2055y_t-1 + 0.1565y_t-2 + 0.3152y_t-3 $-$ 0.3915y_t-4 $-$ 0.3470y_t-5 + a_t where.

• •y_t is the value at time t (i.e., the tth week)
• •0.011325 is the intercept/constant term, $-$ 0.2055 is the AR(1) coefficient, 0.1565 is the AR(2) coefficient, 0.3152 is the AR(3) coefficient, $-$ 0.3915 is the AR(4) coefficient and $-$ 0.3470 is the AR(5) coefficient
• •y_t-1 is the value at time t-1 (one week prior)
• •y_t-2 is the value at time t-2 (two weeks prior)
• •y_t-3 is the value at time t-3 (three weeks prior)
• •y_t-4 is the value at time t-4 (four weeks prior)
• •y_t-5 is the value at time t-5 (five weeks prior)
• •a_t is the error term inherent in any statistical model (i.e. the actual value minus the predicted value at time t)

Table 3 below shows the Python output of model parameters, coefficients, and other statistics relevant to the model. Due to the limited space, a similar summary table for TE’s MA(1) model will not be displayed in the following analysis.

Table 3.

Estimated AR(5) model parameters.

Dependent Variable:	WAR metaphor frequency				No. Observations:	30
Model:	ARIMA(5, 0, 0）				Log Likelihood	112.861
Sample:	30				AIC	−211.723
	coef	std err	z	P>|z|	[0.025	0.975]
const	0.0113	0.001	10.192	0.000	0.009	0.013
ar.L1	−0.2055	0.228	−0.902	0.367	−0.652	0.241
ar.L2	0.1565	0.253	0.617	0.537	−0.340	0.653
ar.L3	0.3152	0.163	1.939	0.053	−0.003	0.634
ar.L4	−0.3915	0.271	−1.443	0.149	−0.923	0.140
ar.L5	−0.3470	0.187	−1.857	0.063	−0.713	0.019

Fig. 6 below visualizes the predicted values (orange) under the AR(5) model versus the actual 30 values (blue). By visual inspection, the model seems to capture the main ups and downs in the data. Mean squared error (MSE)26 of this model is relatively low (0.000081), which indicates a good model-to-data fit with the average distance between the observed data and the predicted data rather small. To further evaluate the predictive accuracy of the fitted model, a train-test approach (see Tay, 2022b for its operation on discourse data) is adopted, using the first 27 out of 30 frequency values as training data to fit the model, and withholding the final three frequency values as testing data. Results show that MSE on the testing data (weeks 28–30) is small (0.000127), suggesting that the model has good predictive accuracy with low average distance between the actual testing data and the predicted testing data.

Fig. 6.

Observed vs predicted values for WAR metaphors in Chinese Global Times. (ARIMA models can also make forecasts of future values, which is not the primary objective of the present study and is therefore not demonstrated in Fig. 6 and Fig. 7.)

In specific terms, this model implies that.

• •WAR metaphor usage in the Chinese Global Times is positively correlated in successive news articles, but only up to five weeks apart. In other words, the frequency of WAR metaphors at week t, t + 1, t + 2…t + 5 tend to move in the same direction, but the frequency value at week t does not provide useful information for that at week t + 6 and beyond
• •The more usage of WAR metaphors in the previous five weeks is likely to be followed by a higher (predicted) frequency at week t + 5, and vice versa

As shown in Fig. 1, the Chinese Global Times generally uses far more WAR metaphors than others and the ARIMA model additionally shows rises/decreases within consecutive weeks followed by a sudden movement towards the opposite direction afterwards. In other words, after a few weeks of downward/upward trend of WAR narratives, the Chinese Global Times tends to strengthen/weaken the WAR threat from the coronavirus. It seems that, through such a tightening-and-loosening process, the Global Times was presenting to the Chinese reader that the COVID-19 should be paid meticulous attention overall.

According to Fig. 4 in Section 5.2, the use of WAR metaphors in the Chinese GT is not based on the actual unfolding of the pandemic in China. For instance, at week 13 when the number of confirmed cases in China reaches its highest point in the year 2020, the WAR metaphor frequency drops to its lowest level. In other words, the tightening-and-loosening strategy is likely devised on purpose, and WAR metaphor use in Chinese GT is not a fair reflection of China’s pandemic situation. Examples 1–4 below will further illustrate this point.27 At week 10 when China has a rather small number of confirmed cases (56 confirmed cases), WAR metaphor usage is far below the series average, and it is mainly for explaining China’s strict pandemic “fight” strategies: e.g., to protect the elderly and the weak (see Example 1). However, the series witnesses an upward trend in the next two weeks (when the number of confirmed cases witnesses an initial rise and a fall afterwards) before a sudden drop at week 13 (when the number of confirmed cases reaches 1,463, the highest in the year 2020). At week 11, the Chinese GT emphasizes that the coronavirus is mankind’s common “enemy” (Example 2a), whereas the US fails to recognize this point (Example 2b). Then, at week 12, the Chinese GT strengthens the severity of this pandemic “fight” in that even Beijing was “attacked” (Example 3a), and Beijing must “fight” (Example 3b). Although Beijing finally won, the rest of China should be even more careful (Example 3c). The newspaper plays down this pandemic “fight” at week 13 by only pointing out that economic resumption relies on a good control of the pandemic situation (Example 4). Such a tightening-and-loosening strategy continues to be seen in the following weeks, which keeps reminding the Chinese reader of the “ferocious” virus “enemy” even on occasions when China temporarily won the “war”, but also occasionally giving them some breathing space from this.

Example 1: Week #10.

ST:	新冠	肺炎	的	大部分	死亡者	是	老人	和	体弱
	xīnguàn	fèiyán	de	dàbùfèn	sǐwángzhě	shì	lǎorén	hé	tǐruò
	corona	pneumonia	AUX	majority	dead people	be	the elderly	and	the weak
	多病者，	而且	穷人	居多，	他们	是	社会	的	弱势
	duōbìngzhě	érqiě	qióngrén	jūduō	tāmen	shì	shèhuì	de	ruòshì
	the ill	and	the poor	mostly	they	be	society	AUX	vulnerable
	群体，	没有	能力	参与	构建	社会	的	抗疫	政策。
	qúntǐ	méiyǒu	nénglì	cānyǔ	gòujiàn	shèhuì	de	kàngyì	zhèngcè
	groups	no	ability	participate in	construct	society	AUX	pandemic fight	policy
TT:	(total omission) (article #46, 26th May)

Example 2: Week #11.

ST2a:	病毒	是	全	人类	共同	的	敌人。
	bìngdú	shì	quán	rénlèi	gòngtóng	de	dírén
	virus	be	all	human	common	AUX	enemy
TT2a:	Viruses are the common enemy of all mankind. (article #48, 7th June)

ST2b:	然而，	美国	就是	这样	尊重	那	200 万	人	染上	新冠
	ránér	měiguó	jiùshì	zhèyàng	zūnzhòng	nà	wàn	rén	rǎnshàng	xīnguàn
	but	US	be	this	respect	those	2 million	people	contract	corona
	肺炎	的	权利	的	就是	这样	尊重	那	11.2 万	人
	fèiyán	de	quánlì	de	jiùshì	zhèyàng	zūnzhòng	nà	wàn	rén
	pneumonia	AUX	right	AUX	be	this	respect	those	112 thousand	people
	在	病毒	攻击	中	自然	死亡	的	权利	的
	zài	bìngdú	gōngjī	zhōng	zìrán	sǐwáng	de	quánlì	de
	AUX	virus	attack	amid	natural	death	AUX	right	AUX
TT2b:	Is this how the US respects the rights of the over 2 million infected people and some 112,000 people who died from COVID-19? (article #49, 11th June)

Example 3: Week #12.

ST3a:	首先，	作为	首都，	北京	的	抗疫	措施	一直	是	全国
	shǒuxiān	zuòwéi	shǒudōu	běijīng	de	kàngyì	cuòshī	yīzhí	shì	quánguó
	first of all	as	capital	Beijing	AUX	pandemic fight	measure	all the time	be	nation
	最严	的	之一，	但	它	还是	被	病毒	攻破	了。
	zuìyán	de	zhīyī	dàn	tā	háishì	bèi	bìngdú	gōngpò	le
	strictest	AUX	one	but	it	still	PASSIVE	virus	breach	AUX
TT3a:	The prevention and control measures in Beijing, China’s capital, were among the strictest in the country, yet still Beijing has been hit. (article #51, 14th June)

ST3b:	中国	的	抗疫	需要	经过	一次	战役	就
	zhōngguó	de	kàngyì	xūyào	jīngguò	yīcì	zhànyì	jiù
	China	AUX	pandemic fight	need	go through	one-time	battle	only
	形成	上	一个	台阶	的	成熟。
	xíngchéng	shàng	yīgè	táijiē	de	chéngshú
	form	up	one	step	AUX	maturity
TT3b:	China should be more mature after each stage of the battle against the epidemic. (article #52, 15th June)

ST3c:	中国	首都	正在	采取	的	行动	决定	了	病毒	终
	zhōngguó	shǒudōu	zhèngzài	cǎiqǔ	de	xíngdòng	juédìng	le	bìngdú	zhōng
	China	capital	PROGRESSIVE	take	AUX	action	determine	AUX	virus	finally
	将	在	这里	被	围剿。	而	那些	没有	采取	相应
	Jiāng	zài	zhèlǐ	bèi	wéijiǎo	ér	nàxiē	méiyǒu	cǎiqǔ	xiàngyīng
	Will	AUX	here	PASSIVE	annihilate	whereas	those	not	adopt	corresponding
	措施	的	地方	就会	自然	承受	相应	的	代价。
	cuòshī	de	dìfāng	jiùhuì	zìrán	chéngshòu	xiàngyīng	de	dàijià
	measure	AUX	place	will	naturally	stand	corresponding	AUX	cost
TT3c:	The measures the capital is taking show that the virus will eventually be curbed here, while places that haven’t taken corresponding measures will have to bear the price. (article #53, 16th June)

Example 4: Week #13.

ST:	抗疫	和	恢复	经济	在	所有
	kàngyì	hé	huīfù	jīngjì	zài	suǒyǒu
	pandemic fight	and	resume	economy	in	all
	国家	都	是	紧密	交织	的
	guójiā	dōu	shì	jǐnmì	jiāozhī	de
	country	all	be	close	interwoven	AUX
TT:	Epidemic fight and economic resumption are closely related in all countries. (article #54, 27th June)

Nonetheless, WAR metaphor usage seems to randomly fluctuate across time in the English Global Times and there is no optimal ARIMA model. As shown in the above examples, the English translated news can have some narratives in the ST wholly omitted (see Example 1) or only those metaphorical words omitted (see Example 2b). The original WAR metaphor can also be translated into a different metaphor, like VIOLENT FORCE (see Example 3a). The English Global Times uses WAR metaphors in a spontaneous and ad hoc manner, which is more grounded in the immediate context and hardly shaped by prior WAR narratives. The contrasting modeling results between the source news and the translated news in Global Times suggest that the GT fabricated an overall WAR scenario for its Chinese readers using a tightening-and-loosening strategy whereas each WAR metaphor use for its English readers is ad hoc, and less likely devised in advance. Obviously, the shift from a modelable series in the source news to a randomly fluctuating series in the translated news is mainly achieved through different metaphor translation methods: omitting, replacing, or paraphrasing the original WAR metaphors in the English translated news texts as shown in the above examples. One possible reason for an overall time-based control of WAR metaphor use in the Chinese source news lies in China’s Covid policy in the year 2020. The GT constantly uses WAR metaphors to legitimize a nationwide “fight” against the coronavirus “enemy”, and therefore make the Chinese people more willing to obey those Covid restrictions. However, there is no need to do so for its English readers who are receptive of a totally different kind of response policy.

6.2. The Economist: MA(1) model vs Random fluctuation

Similar to the Global Times, WAR metaphor usage in the source news of The Economist is modelable, whereas it fluctuates randomly in TE’s translated news. Nevertheless, the two newspapers differ from each other in: a) language of the modelable source news (Chinese vs English); b) the model type: Autoregressive (AR) vs Moving average (MA). In other words, although both the GT’s Chinese reader and TE’s English reader are presented with time-structured WAR narratives of COVID-19, these narratives were not structured in the same way across time. AR and MA models differ mainly in that the former predicts the present value using past values whereas the latter uses past error terms to make predictions. This MA(1) model can be formally expressed as y_t = 0.0008 $+$ 0.3645a_t-1 + a_t where.

• •y_t is the value at time t (i.e., the tth week)
• •0.0008 is the intercept/constant term, 0.3645 is the MA(1) coefficient
• •a_t is the error term (i.e., the actual value minus the predicted value at time t)
• •a_t-1 is the error term at time t-1 (i.e., one week prior)

Fig. 7 below displays the observed vs predicted plot for WAR metaphor usage in English TE. The model seems to have a reasonable goodness-of-fit by visual inspection, better than that of the Chinese GT. The MSE value is even closer to zero, implying an acceptable distance between the observed data and the predicted data. These suggest that the WAR metaphor data in English TE is more “modelable” than that in Chinese GT. As before, the first 40 out of 43 frequency values were used as training data to fit the model, and the final three frequency values were adopted as testing data to evaluate the predictive accuracy of the fitted model. MSE on the testing data (weeks 41–43) is small (0.000001), which indicates that the model has reasonable predictive accuracy with low average distance between the actual testing data and the predicted testing data.

Fig. 7.

Observed vs predicted values for WAR metaphors in English The Economist.

In broad structural terms, the MA(1) model implies that.

• •WAR metaphor usage is not necessarily correlated in successive weeks. Instead, the present WAR metaphor frequency is negatively correlated with the size of prior error terms, up to one week apart. Meanwhile, the frequency value at week t does not provide useful information for that at week t + 2 and beyond
• •A positive error term (i.e., observed frequency higher than predicted frequency) is likely to be followed by lower (predicted) WAR metaphor frequency at the next week, and vice versa

As shown in Fig. 1, the English TE series witnesses ups and downs in consecutive weeks and the overall frequency is lower than that of the Chinese TE. The ARIMA model additionally indicates that i) there are occasions when the WAR metaphor frequency is unexpectedly high/low for various contextual reasons, engendering a high positive/negative error term (the actual value minus the predicted value) for that week; ii) a larger positive error term will make the frequency value at the next week more likely to drop whereas a larger negative error term will make it more likely to rise. If the frequency is not unexpectedly high/low, it is unlikely to witness a large increase or decrease at the next week. This overall low frequency of WAR metaphors in the English TE could be interpreted in the following ways: first, Western newspapers28 and research communities are quite critical of the use of WAR metaphors to describe crisis situations (as reviewed in Section 1); and second, the UK barely places Covid restrictions and accordingly, the news media may use less WAR frames. Nevertheless, this equilibrium is occasionally faced with an unexpected interruption. Examples 5–6 below will illustrate the unexpected high observed frequency at week 30 and 31 and Example 7 will demonstrate the “restoration” of WAR narratives at week 32.

Example 5: Week #30.

ST:	This starts with the miracle of vaccination, which defends against a pathogenic attack before it is launched. (article #87, 22nd August)
TT:	首先	是	疫苗	接种	这一	奇迹，	它	能	在
	shǒuxiān	shì	yìmiáo	jiēzhǒng	zhèyī	qíjì	tā	néng	zài
	first of all	be	vaccine	inoculation	this	miracle	it	can	AUX
	病毒	发起	致	病	攻击	之前	做出	防御 。
	bìngdú	fāqǐ	zhì	bìng	gōngjī	zhīqián	zuòchū	fángyù
	virus	launch	cause	illness	attack	before	make	defense

Example 6: Week #31.

ST:	Party praise for covid-fighting doctors does little to fix China’s ailing health system. (article #88, 29th August)
TT:	党	表彰	抗疫	医生	对	祛除	中国
	dǎng	biǎozhāng	kàngyì	yīshēng	duì	qūchú	zhōngguó
	the Party	appraise	pandemic fight	doctor	towards	remove	China
	医疗	体系	的	弊病	无甚	用处。
	yīliáo	tǐxì	de	bìbìng	wúshèn	yòngchù
	health care	system	AUX	malady	no	use

Example 7: Week #32.

ST:	Before covid-19 struck, for example, flexible-office companies (including the troubled WeWork) had a tiny global market share of under 5%. (article #89, 10th September)
TT:	在	新冠	疫情	出现	之前，	灵活	办公	的	公司	（包括
	zài	xīnguàn	yìqíng	chūxiàn	zhīqián	línghuó	bàngōng	de	gōngsī	bāokuò
	AUX	corona	pandemic	appear	before	flexible	officing	AUX	firm	include
	麻烦	缠身	的	WeWork）	在	全球	市场	中	所	占
	máfán	chánshēn	de		zài	quánqiú	shìchǎng	zhōng	suǒ	zhàn
	trouble	plagued	AUX	WeWork	in	global	market	AUX	AUX	account
	的	份额	很	小，	不到	5％。
	de	fèné	hěn	xiǎo	bùdào
	AUX	share	very	small	less than	5％

Unlike GT and NYT whose highest and lowest values are quite repetitive, TE has unexpectedly high values at week 30 and 31, both of which occur in late August, a time when China and Russia developed covid vaccines and put them into emergency use (e.g., for frontier doctors and nurses as well as the military personnel).29 At this time, the modelable Chinese GT did not witness a high frequency of WAR metaphors (at week 19 of the series). It is assumed that with an increasing familiarization with coronavirus and the development of vaccines, the GT newspaper led by the Community Party of China downplays its WAR narratives to temporarily give the Chinese people breathing space from the “terror” of this pandemic “war”, though Covid restrictions were not loosened in throughout the year 2020. Nonetheless, the English TE’s unexpected high frequency at week 30 and 31 seems more explainable. At week 30, WAR metaphors are used for science popularization, as well as health and policy communication, describing how a covid vaccine works (see Example 5), which may encourage people to participate in the COVID-19 vaccination program. At week 31, they are utilized for sarcastic purposes, illustrating how China’s pandemic “fight” achievements were built on the sacrifices of doctors and nurses (see Example 6). Further analyses can focus on these two points, thereby revealing the newspapers’ function in science/health/policy communication and its political nature. Conversely, at week 32 (in early September), the English TE reduces its WAR narrative and begins to report the pandemic’s positive impact: it expediates the transformation of workplace. However, on occasions when the predicted frequency is quite close to the actual frequency (e.g., week 7, 9 and 38), there are no sharp increases or decreases at the next week. In fact, according to Fig. 5 in Section 5.2, such an unexpected use of WAR metaphors in TE targeting its English readers is fabricated on purpose (e.g., for a promotion of the government’s vaccination program or other political reasons) and has little relation to the pandemic severity in the UK.

Like GT, WAR metaphor usage in TE’s translated series is non-modelable, which indicates that the Chinese TE uses WAR metaphors spontaneously, more grounded in immediate contexts. Metaphors are sometimes retained in the translated text where the vaccine-supported “defense” against the coronavirus “attack” is explained to both English and Chinese readers (see Example 5) and the recognition of Chinese doctors’ “fight” efforts is seen in both the source news and the translated news (see Example 6). Metaphors can also be omitted, such as Example 7, where the “strike” launched by coronavirus against mankind is only expressed to the source news readers. The non-modelability of WAR metaphors in the TE’s translated news seems to indicate that the processing of WAR metaphors in translation is not guided by an overall time-based strategy. Instead, whether a WAR metaphor should be retained, omitted, paraphrased or created in the translated news is on a case by case basis.

6.3. The New York Times: Random fluctuation vs Random fluctuation

Unlike GT and TE, who have their source news series modelable and translated news series non-modelable, The New York Times, in its both source (English) and translated (Chinese) news, describes the pandemic situation as a WAR in a spontaneous, ad hoc manner. In other words, WAR metaphor usage in both Chinese and English NYT is not structured in a way such that prior practice predicts the present, but instead randomly fluctuates across time.

As shown in Fig. 1, the two non-modelable series share similar “shapes” with minor observable differences. All this indicates the fact that readers of different language versions of reports may have slightly different experiences of the same global event. Moreover, given the unpredictability of the COVID-19 situation, such non-modelability in discourse seems to reflect the corresponding background reality of COVID-19. Conversely, the modelability of GT’s and TE’s source news series is more likely to result from certain intended discourse manipulation: use WAR frames in a time-structured manner to “construe” the pandemic event. However, this does not mean that the NYT does not use discourse to “construe” social reality, as WAR metaphor itself is already a way of conceptualizing the pandemic event. It is also possible that The New York Times uses metaphors other than WAR and even discursive features other than metaphor in a time-structured manner, which is beyond the scope of the current study. Examples 8–10 below illustrate how NYT uses WAR metaphors in immediate contexts to “construe” social reality.

Example 8: Week #16.

ST:	Though the new enemy is a virus, even less susceptible to verbal and physical firepower than terrorists, the Trump administration appears to be setting its target on a foreign power: China, where the outbreak appears to have started but which is hardly responsible for the United States being the most infected country in the world. (article #48, 12th May)
TT:	虽然	新的	敌人	是	一种	病毒,	比起	恐怖分子，	它
	suīrán	xīnde	dírén	shì	yīzhǒng	bìngdú	bǐqǐ	kǒngbùfènzǐ	tā
	although	new	enemy	be	a type of	virus	compare to	terrorist	it
	更不	容易	遭受	言语	和	身体	上	的	火力
	gèngbù	róngyì	zāoshòu	yányǔ	hé	shēntǐ	shàng	de	huǒlì
	less	easy	suffer	verbal	and	physical	AUX	AUX	firepower
	威胁，	但	特朗普	政府	似乎	正在	将	目标	对准
	wēixié	dàn	tèlǎngpǔ	zhèngfǔ	sìhū	zhèngzài	jiāng	mùbiāo	duìzhǔn
	threat	but	Trump	government	seem	PROGRESSIVE	put	target	towards
	一个	外国	势力:	中国。	疫情	似乎	源自	中国,	但
	yīgè	wàiguó	shìlì	zhōngguó	yìqíng	sìhū	yuánzì	zhōngguó	dàn
	a	foreign	force	China	pandemic	seem	from	China	but
	美国	成为	世界	上	感染	最	严重	的	国家,
	měiguó	chéngwéi	shìjiè	shàng	gǎnrǎn	zuì	yánzhòng	de	guójiā
	US	become	world	AUX	inflection	most	severe	AUX	country
	恐怕	难以	归咎	于	它。
	kǒngpà	nányǐ	guījiù	yú	tā
	afraid	difficult	blame	AUX	it

Example 9: Week #19.

ST:	But America’s impatience, its unwillingness to do what it takes to deal with a threat that can’t be beaten with threats of violence, runs much deeper than one man. (article #55, 10th June)
TT:	但是	美国	太	没有	耐心，	不愿意	采取	必要	的	行动
	dànshì	měiguó	tài	méiyǒu	nàixīn	bùyuànyì	cǎiqǔ	bìyào	de	xíngdòng
	but	US	extremely	have no	patience	unwilling	take	necessary	AUX	action
	来	应对	一个	无法	以	暴力	威胁	击败	的	威胁
	lái	yìngduì	yīgè	wúfǎ	yǐ	bàolì	wēixié	jībài	de	wēixié
	to	cope with	an	unable	use	violent	threat	beat	AUX	threat
	这	远非	一个	人	的	行为	所	能	解释。
	zhè	yuǎnfēi	yīgè	rén	de	xíngwéi	suǒ	néng	jiěshì
	this	beyond	one	person	AUX	action	AUX	can	explain

Example 10: Week #25.

ST:	Covid-19 was supposed to be China’s Chernobyl. It’s ended up looking more like the West’s Waterloo. (article #65, 14th October)
TT:	新冠	病毒	曾	被	视为	中国	的	切尔诺贝利	结果
	xīnguàn	bìngdú	céng	bèi	shìwéi	zhōngguó	de	qiēěrnuòbèilì	jiéguǒ
	corona	virus	once	PASSIVE	regard as	China	AUX	Chernobyl	end
	看	上去	更	像	是	西方	的	滑铁卢。
	kàn	shàngqù	gèng	xiàng	shì	xīfāng	de	huátiělú
	look	AUX	more	like	be	West	AUX	Waterloo

In general terms, Examples 8–10 present the NYT’s dissatisfaction with the American government’s (i.e., the Trump Administration) performance in coping with the coronavirus “enemy”. One reason is that Trump focused more on the conflicts with China than on this real health threat (see Example 8). Another is that the United States, as a whole, lacked determination and patience and was unwilling to take necessary measures (see Example 9). All this led to the US’s “waterloo” (see Example 10). Behind the scene, there was the 2020 United States presidential election and the NYT framed this pandemic WAR in favor of the Democratic Party (see previous research mentioned in Section 2 and Example 8 for the NYT’s Democratic partisanship). Such ideological underpinnings are more related to the immediate context and are hardly affected by the actual unfolding of this pandemic event. However, unlike the NYT that only construes this pandemic “war” in immediate context, the GT and the TE also exert much control at a macro level.

To put it another way, GT and TE attempt to employ a two-layer manipulation when framing the pandemic as a “war” for their source news readers: first, construe this pandemic “war” from a political perspective (ironically mocking the US’s “failure” or China’s relentless “exploitation” of doctors and nurses); and second, construe this pandemic reality from a broad time-structured perspective (an overall tightening-and-loosening strategy vs an unexpected use of WAR frames from time to time). Nonetheless, in the translated news, GT and TE tend to ignore the second layer, and only use WAR frames in an immediate, ad hoc manner. Unlike GT and TE, NYT construes this pandemic “war” using strategies that are more spontaneous and contextualized for its both source news and translated news readers. In a sense, the present results have further strengthened the point that translation is of vital significance in journalistic studies not only because it is nowadays an actual way of news production (i.e., a newspaper publishes translations of a single source article as its multilingual versions) but also because it relates to ideological affinity, gatekeeping, and power relations (Valdeón, 2022). In other words, when journalists decide “what to translate” and “how to translate”, they are selecting or constructing a social reality for readers with different socio-political backgrounds. It actually makes a difference if they use time-structured discourse for one reader group but do not do so for another reader group.

7. Conclusion

The study has demonstrated that time series analysis is useful when comparing cross-linguistic and cross-national news discourse, especially in terms of their respective ways of using metaphors in source news and translated news to manipulate ideology for different target readers. Specifically, if metaphors are assumed to reveal hidden opinions and ideologies in news discourse (as suggested by the reviewed studies in Section 1), when a series of metaphors is modelable, it indicates that a metaphor-led ideology manipulation is also structured by time (i.e., the two-layer manipulation summarized in Section 6.3). On the other hand, when the series is non-modelable, the ideology may still be manipulated as metaphors themselves are already ways of manipulating recipients’ beliefs and opinions, but the manipulation is not time-structured (i.e., the one-layer manipulation summarized in Section 6.3). Subsequently, specific details of the fitted model further reveals the specific temporal dynamics of this manipulation. These models can then be scrutinized using qualitative methods to inspect time series patterns in context. In fact, it is the qualitative interpretation that makes the time series analysis of discourse data intriguing. In other words, ARIMA modeling outcomes only provide a focal point for the later qualitative analysis that reveals the specific strategies of manipulation of beliefs and opinions and the possible consequence or impact of such manipulation.

Following this analytical reasoning, this study conducts both quantitative and qualitative analyses of WAR metaphors across week intervals and concludes that the Global Times and The Economist manipulate COVID-19 as a WAR in a broad time-structured manner for their source news readers but in an immediate ad hoc way for their translated news readers. By contrast, The New York Times constructs this pandemic WAR randomly across time for both its source and translated news readers. Moreover, the AR(5) model for the Chinese Global Times series and the MA(1) model for the English The Economist series suggest that the two newspapers actually use different time-structuring strategies when framing this pandemic WAR for their source news readers. The Chinese Global Times adopts an overall tightening-and-loosening strategy whereas the English The Economist witnesses an unexpected use of WAR metaphors from time to time. These cross-linguistic and cross-national differences are related to their ideological underpinnings as well as the Covid policies in the three countries. The findings have further highlighted the complicated nature of news discourse in mediating political stances, ideologies, and opinions towards world events for different target readers through a translation practice. More importantly, the findings have deepened our understanding of WAR metaphors. In spite of the long-standing criticism of WAR metaphors in a pandemic situation (as reviewed in Section 1), the three newspapers use WAR metaphors for different reader groups to varying degrees and for various political and policy reasons. Moreover, the use of WAR metaphors can be either ad hoc or time-structured. Special attention should be paid to the fact that the translation of these WAR metaphors can actually change these time-structuring patterns. From this aspect, the current research may enrich existing studies on the translation of Covid metaphors.

Beyond theoretical contributions as such, the study also provides a different approach to news discourse analysis, which actually serves as a complementary tool rather than a substitution for the more usual content and thematic analysis. Specifically, auto-ARIMA that selects the best model based on a minimum AIC value from a variety of candidate models has its own strength, and can sometimes offer alternative solutions when a manual observation of (P)ACF spikes (a method mostly adopted in previous research) is less straightforward. A mixed-method approach with TSA providing focal points for later qualitative analysis may lead to new insights. However, the study is limited in terms of time span (only week intervals in 2020), metaphor type (only WAR metaphor) and the scope of newspapers. Future research can take a longer-distance view on various discursive features across more varied newspaper types, thereby unveiling more hidden patterns underlying the complicated relations between (translated) news discourse and ideology manipulation. Moreover, this TSA method can be adopted in similar studies using social media data to investigate, from a (non)modelability perspective, whether the construction of discourse in newspaper has any effect on how people talk and think about the same world event on social media platforms.

Funding

This work is partly supported by a Hong Kong Polytechnic University Faculty Reserve Grant (1-ZVY8) awarded to the second author.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Biographies

Yufeng Liu is a PhD candidate at the Department of Chinese and Bilingual Studies, The Hong Kong Polytechnic University. Her research interests lie in the intersection of metaphor, translation and communication studies. She has published articles in journals such as Discourse & Society and Social Semiotics as well as some book chapters published by Routledge and Springer. https://orcid.org/0000-0002-3423-664X

Dennis Tay is a Professor at the Department of English and Communication, The Hong Kong Polytechnic University. His research interests include cognitive linguistics, healthcare communication, and the statistical modeling of discourse. He is Co-Editor of Metaphor and the Social World, Associate Editor of Metaphor and Symbol, and an Academic Editor of PLOS One. https://orcid.org/0000-0001-9484-6266.

Appendix A. Keywords for generating concordance lines in AntConc

Keywords in English Texts	Keywords in Chinese Texts
Covid*30	covid*
	冠状* (guānzhuàng, corona)
	新冠* (xīnguān, novel corona)
*virus*	病* (bìng, virus/disease/pathogen/case)
disease*
pathogen*
case*
pneumonia*	肺炎 (fèiyán, pneumonia)
*demic*	流行* (liúxíng, pandemic/epidemic/plaugue)
plague*	*疫* (yì, pandemic/epidemic/plague/virus fight/vaccine)
vacci*
influenza*	流感* (liúgǎn, flu/influenza)
flu
infect*	传染* (chuánrǎn, infection/infectious disease)
crisis*	危机* (wēijī, crisis)
outbreak*	爆发/暴发 (bàofā, outbreak)

³⁰* means any letter(s). For instance, *demic* can refer to “pandemic” and “epidemic”.

Appendix B. Python code

Most Python code of the study is listed on the Appendix of Tay (2022b, pp. 71–73) book chapter. We added an auto-ARIMA feature and the train-test approach to the analysis and the corresponding Python code is as follows:

#auto-ARIMA
from pmdarima import auto_arima
import time
t0=time.time()
model_name='ARIMA'
arima_model=auto_arima(series, start_p=0, start_q=0,
max_p=20, max_q=5,
seasonal=False,
d=None, trace=True,random_state=12345,
error_action='ignore',
suppress_warnings=True,
stepwise=True)
#train-test approach
train_series=series.iloc[0:len(series)-3]
model=sm.tsa.SARIMAX(train_series, order=(0,0,1),trend='c').fit()
print(model.summary())
predict=(model.get_prediction(start=1,end=len(series)))
predictinfo=predict.summary_frame()
forecast=(model.get_forecast(steps=6))
forecastinfo=forecast.summary_frame()
fig, ax=plt.subplots(figsize=(15, 5))
forecastinfo['mean'].plot(ax=ax, style='k--',label=“forecast”)
plt.plot(series, label=“observed”,color='dodgerblue')
plt.plot(predictinfo['mean'], label=“predicted”,color='orange')
ax.axvspan(len(train_series),len(train_series) + 3,color='red',alpha=0.2,label='train-test')
ax.fill_between(forecastinfo.index, forecastinfo['mean_ci_lower'], forecastinfo['mean_ci_upper'], color='grey', alpha=0.2, label=“95% CI”)
ax.set_ylabel('WAR metaphor frequency', fontsize=12)
ax.set_xlabel('Interval', fontsize=12)
ax.set_title('English TE series',fontsize=12)
plt.setp(ax, xticks=np.arange(1, len(series) + 4, step=2))
plt.legend(loc='best',fontsize=10)
plt.xticks(fontsize=6)
plt.show()
#MSE score of testing data
from sklearn.metrics import mean_squared_error
mse=mean_squared_error(series.iloc[-3:], predict.predicted_mean.iloc[-3:])
print('mse: %f' % mse)

Data availability

Data will be made available on request.