Introduction

The notion of sentiment disguise is broadly proposed and discussed in the realm of psychology (Oring 1994; Zhang et al. 2018). It refers to the act of expressing certain sentiments, either consciously or unwittingly, to hide or twist the intended sentiments. Most psychological probes into sentiment disguise focus on the genre of irony, the conscious use of ‘imagination and ingenuity to artfully disguise the expression of a negative sentiment’ (Hao and Veale 2010: 635). Similarly, in the domain of language research, it is observed in many cases as conscious activities, for example, manipulation in news media (Chovanec 2019; Ferrara et al. 2020). However, when translation is involved (Mohammad et al. 2016), it may occur unwittingly, as is representatively observed in the examples in online book reviews of The Happy Prince and Other Tales.

The Happy Prince and Other Tales, a widely acclaimed work of children’s literature, has garnered significant popularity worldwide and has been successfully translated into Chinese (Shen 2014). The reception of The Happy Prince and Other Tales in its Chinese translations reveals instances of sentiment disguise. For instance, common reader comments on the Chinese online book review platform DoubanFootnote 1 include statements such as “The Happy Prince and Other Tales makes me sad/cry every time I read it”. These comments prompt an inquiry into the discrepancy between the seemingly positive connotation of the word “happy” in the book’s title and the audience’s poignant and heartrending response. Explanations for this divergence could be found perhaps from a glimpse of the comments on The Happy Prince and Other Tales on English online book review platform GoodreadsFootnote 2—“some of them (stories in The Happy Prince and Other Tales) don’t have happy endings, and I actually liked those little twists”—highlighting unhappy endings of The Happy Prince and Other Tales. The negatively complied storyline thus might be the reason for the heartbroken emotive reception by the Chinese readers. Nevertheless, this assumption is easily refuted if the last sentence of the book review “I actually liked those little twists” were taken into consideration. It was the readers who had a very negative attitude, claiming that their hearts were broken, while in fact they were not, as they still appreciate the whole story. A more interesting observation of the above book review is that, although the storyline is the same for both Chinese and English readers, there still exhibits a seemingly divergent sentiment reception among Chinese and English audiences. This being the case, there is a possibility that the sentiment, either of the Chinese audience or of the English audience, might be disguised.

The contrast in emotional responses between the Chinese and English readers of The Happy Prince and Other Tales presents an intriguing case for exploring the concept of sentiment disguise in the context of cross-cultural literary translation. This phenomenon challenges the assumption of a direct, universal correspondence between the textual content and the reader’s emotional experience (Rosenblatt 1994), raising questions about the role of cultural and linguistic factors in shaping the translation and reception of emotive narratives.

Sentiment in texts: approaches to sentiment analysis

The field of text sentiment analysis leverages natural language processing, text mining, and computational linguistics to identify and extract subjective information from source materials. Research in this domain has primarily concentrated on three levels of granularity: document-level sentiment analysis, sentence-level sentiment analysis, and aspect-level sentiment analysis (Liu 2012). The purpose of document-level sentiment analysis is to infer the overall opinion expressed in a given text, with the assumption that the document conveys a unified sentiment towards an entity (Glorot et al. 2011; Moraes et al. 2013). In contrast, sentence-level sentiment analysis focuses on the sentiment contained within individual sentences, which may appear in academic papers, discussions, or as isolated micro-texts on social media platforms (Yu and Hatzivassiloglou 2003; Kouloumpis et al. 2011). While document-level and sentence-level analyses provide an overall sentiment direction, they often fail to reveal the specific target of that sentiment. This is because they operate under the assumption that a document or sentence expresses a single, uniform feeling towards an entity—an assumption that is considered rather strong. To address this limitation, aspect-level sentiment analysis adopts a more granular approach, identifying the sentiment associated with each individual entity within the text (Hu and Liu 2004). This finer level of examination allows for a more nuanced understanding of the sentiment distribution. Considering these three levels of sentiment analysis, researchers can gain a more comprehensive and nuanced understanding of the subjective information embedded within textual data, facilitating more robust and insightful interpretations.

Sentiment analysis has evolved through several key trends and application models, including rule-based approaches, machine learning-based methods, and the emerging paradigm of deep learning (see Fig. 1). In rule-based sentiment analysis, the primary focus has been on the use of sentiment-bearing words and their combinations to assess the polarity of phrasal units (Taboada et al. 2011). Early research in this domain discovered that simply counting valence words could lead to inaccurate conclusions (Polanyi and Zaenen 2006; DeSteno et al. 2000). Instead, rule-based sentiment analysis approaches rely heavily on sentiment lexicons—dictionaries that provide sentiment annotations for the words and phrases they include (Joshi et al. 2017). One prominent example of such a sentiment lexicon is SentiWordNet, which is based on the WordNet semantic network (Miller 1995). In SentiWordNet, each synset is assigned positive, negative, or objective scores, indicating its subjective orientation (Esuli and Sebastiani 2006). Other widely used sentiment lexicons include SO-CAL (Taboada et al. 2011), SCL-OPP (Kiritchenko and Mohammad 2018), SCL-NMA (Kiritchenko and Mohammad 2017) and NRC Lexicon (Mohammad and Turney 2013a). For the current study, the NRC Lexicon has been selected as the sentiment analysis tool. This choice is based on the lexicon’s high-quality annotations for both word-sentiment and word-emotion pairings, as well as the consistency it provides in terms of polarity and emotion level sentiment analysis, given that the lexicon was primarily created through manual expert annotation (Taboada et al. 2011). While machine learning methods might adapt to domain-specific patterns over time, the rigor of manually annotated lexicons offers a stable foundation for cross-domain comparability—an advantage for studies prioritizing consistent sentiment and emotion benchmarks over adaptive performance.

Fig. 1: Approaches to sentiment analysis.
Fig. 1: Approaches to sentiment analysis.
Full size image

This figure shows levels and methods of sentiment analysis, from document to aspect level.

The present study adopts a two-pronged approach to sentiment analysis, examining both polarity-level sentiment (e.g., positive, negative, neutral) as well as emotion-level sentiment (e.g., JOY, SADNESS, LOVE). Polarity-level sentiment analysis detects more straightforward positive or negative signals within the text. In contrast, sentiment analysis at the emotion level explores a wider range of human emotions and sensitivities. This includes not only basic feelings such as FEAR, LOVE, and HAPPINESS, but also more complex social emotions like SHAME, GUILT, and ENVY (Smith and Kim 2007). The concept of emotion encompasses a multifaceted set of interactions between subjective and objective factors, mediated by neural and hormonal systems (Kleinginna and Kleinginna 1981). This comprehensive understanding of emotion serves as the foundation for the present study’s examination of sentiment beyond simple polarity, delving into the nuanced landscape of human emotional expression.

The inherent complexity and social attribute of emotions present a significant challenge for linguistic and natural language processing specialists tasked with identifying and analyzing emotional expressions within textual data. While early research in this domain predominantly focused on developing emotion lexicons across languages (Pavlenko 2008; Boster 2005; Chen and Skiena 2014), the field has evolved to incorporate both lexicon-based and machine learning approaches. Traditional lexicon development has typically relied on expert judgments (Asghar et al. 2017) or extraction from authoritative dictionaries (Yekrangi and Abdolvand 2021), with multilingual lexicons often created through translation of emotion terms (Mohammad and Turney 2013a). Recent advances have expanded the methodological toolkit, with supervised learning approaches utilizing sentence-level annotated datasets (Yang and Cardie 2014) and transformer-based models (Bilianos 2022) demonstrating particular promise for capturing contextual emotional nuances. However, lexicon-based methods remain valuable for applications requiring interpretability, cross-linguistic consistency, and theoretical grounding in established psychological frameworks (Ekman 1992; Plutchik 2003).

Previous literature has proposed a variety of emotion taxonomies, several of which have garnered widespread recognition within the field. These include Ekman (1992) classification of six basic emotions (JOY, ANGER, FEAR, SADNESS, DISGUST, and SURPRISE), Parrott (2001) model of five fundamental emotions (JOY, SADNESS, ANGER, FEAR, and LOVE), and Plutchik (2003); Plutchik (1991) taxonomy of eight primary emotions (JOY, SADNESS, ANGER, FEAR, TRUST, DISGUST, SURPRISE, and ANTICIPATION). More recently, the extensive GoEmotions taxonomy has been developed, which identifies over 27 distinct emotional states (Demszky et al. 2020). For the purposes of the current investigation, the emotion taxonomy framework proposed by Plutchik and further developed by Mohammad and Turney (2013b) will be adopted. This taxonomy, constructed through crowdsourcing on Amazon’s Mechanical Turk, includes a 14,200-word-type aspect-level emotion lexicon grounded on Plutchik’s (1991, 2003) model of eight basic and prototypical emotions: JOY, SADNESS, ANGER, FEAR, TRUST, DISGUST, SURPRISE, and ANTICIPATION. This eight-emotion framework provides optimal balance between analytical precision and conceptual clarity for cross-linguistic analysis, avoiding both oversimplification and excessive fragmentation while enabling dimensional analysis of intensity and polarity relationships.

Sentiment disguise and translation

Previous language studies of sentiment disguise focus on the genre of irony, the use of “imagination and ingenuity to artfully disguise the expression of a negative sentiment” (Hao and Veale, 2010: 635), referring to “sentiment-rich viewpoints with concision, sharpness and humor” (Veale and Hao 2010: 765). The present study, which examines the genre of online book reviews, can be seen as an extension of these previous discussions on sentiment disguise in language studies. Like irony, online book reviews often exhibit a concise, sharp, and sometimes humorous style in their expression of sentiment and emotion (Wang et al. 2019). By analyzing sentiment disguise in this context, the current research builds upon the existing understanding of how language can be used to artfully disguise underlying sentiments and emotions.

Translation has the potential to trigger alterations in sentiment representation (Mohammad et al. 2016), and this process is further complicated by cultural-linguistic asymmetries that may lead to unintentional sentiment disguise. For instance, Venuti’s (2017) framework of domestication—where translators assimilate source-text sentiments to target-culture norms—can dilute or reframe affective cues (e.g., softening irony to align with reader expectations). Conversely, foreignization risks estranging sentiments by preserving source-text features that lack equivalent emotional resonance in the target language. Such shifts are particularly salient in literary translation, where divergent receptions (e.g., Oscar Wilde’s The Happy Prince and Other Tales) may stem from mismatches between intended and perceived sentiment. Existing studies in the reception of literary canon via translation mainly focus on broader prisms, for example, how the canon or non-canon from a foreign culture favorably prospered or regrettably foiled in the recipient culture and literary systems (Rigney 2012; Yu 2020; Liu and Li 2024). However, to address and unravel the divergent sentiment reception from Chinese and English audiences, thereby figuring out what happened in the process of translation, a closer and more refined examination into the sentiment data is required.

Academic interest in the sentiment analysis of children’s literature dates back to the late 20th century (Stevenson 1997). Prior research within this domain has concentrated on three principal thematic strands: the detection and recognition of sentiments expressed within children’s literature texts (Zad et al., 2021; Alm et al. 2005); the socio-cultural perceptions and implications surrounding the portrayal of sentiment in children’s literature (Adukia et al. 2022); and the psychological investigation of emotion-related hypotheses in this literary genre (Jacobs et al. 2020). For instance, Volkova et al. (2010) conducted an experimental study designed to enable even untrained participants to perform sentiment analysis at the sentence level with a high degree of consistency. Meanwhile, Alm et al. (2005) adopted a supervised machine learning approach, utilizing the SNoW learning architecture to empirically examine text-based sentiment prediction. From a socio-cultural perspective, Adukia et al. (2022) investigated the depictions and representations of race and gender in prominent children’s books over the past century, exploring the relationship between the reception of these works and the prevailing local beliefs. Delving into the psychological dimension, Jacobs et al. (2020) focused their attention on the Pollyanna Effect, leveraging the model-based, unsupervised vector space sentiment analysis tool, SentiArt, to generate valuable predictions for future research in the realm of literary sentiment. While these three thematic strands capture the key trends in sentiment and emotion research within children’s literature, a closer examination reveals certain limitations in this body of work. Specifically, the focus has been primarily on the sentence level, without sufficient attention paid to analysis at the document and aspect levels. Additionally, the emotional research in children’s literature has tended to be predominantly empirical, lacking a more descriptive and interpretive approach.

To address these gaps, the current study aims to take a more comprehensive and nuanced approach. Specifically, it will conduct an analysis at the sentence and aspect levels for both Chinese and English online book reviews. The study will also provide detailed and comprehensive descriptions of the polarity and emotion level sentiment features observed. Moreover, the investigation will specifically focus on the following research questions: (1) Is there evidence of sentiment disguise in the reception of Oscar Wilde’s The Happy Prince and Other Tales? (2) If sentiment disguise is identified, how are these disguises realized? (3) What are the underlying grounds for these sentiment disguises?

Method

Data

The dataset in this study involves two parts: the reception dataset and the source dataset. A summary of the dataset is provided in Table 1.

Table 1 Summary of the dataset (E = English/C = Chinese).
Full size table

Reception dataset

The reception dataset utilized in this study consists of excerpted book reviews obtained from prominent social networking platforms. To analyse emotive reception in a comprehensive manner, two sites with sizable user bases that facilitate book ratings and reviews were selected, aligning with the current research agenda. Specifically, the study excerpted 300 English book reviews of Oscar Wilde’s The Happy Prince and Other Tales, comprising a total of 269,565 characters, from the platform Goodreads. Additionally, 300 Chinese book reviews of the work, totaling 87,735 characters, were extracted from the site Douban. Together, these 600 reviews, selected randomly and automatically using tools goodreads-scraper and Beautiful Soup in Python, form the holistic reception dataset for analysis. The selection of 300 reviews from each platform ensures a balanced comparison while maintaining a manageable dataset for analysis, allowing for a robust examination of sentiment dynamics without overwhelming complexity. The choice of these two social networking platforms, with their millions of active users engaged in book evaluation and commentary, ensures that the reception data captured represents a diverse range of emotive responses to the target literary work. This sampling approach provides a robust foundation for the study’s examination of sentiment dynamics, particularly in the context of the observed divergent responses between Chinese and English-speaking audiences. This diversity is particularly relevant given the distinct demographics and cultural contexts of the platforms, which may influence how sentiments are expressed.

Source dataset

The source dataset for this study comprises the original English text of Oscar Wilde’s The Happy Prince and Other Tales, published in 1888, along with its Chinese translation, titled 快乐王子集 (The Happy Prince and Other Tales), published in 2018. The selection of this literary work was based on its popularity and high reputation, particularly within the Chinese context. In China, 快乐王子集 (The Happy Prince and Other Tales) is a recommended reading series for primary school children, curated by the Ministry of Education. The book is published by the People’s Literature Publishing House, one of the most prestigious publishing houses in the country, further underscoring its cultural significance and widespread recognition. The inclusion of both the source English text and its Chinese translation allows for a nuanced examination of the sentiment dynamics that emerge during the translation process.

Analytical procedures

Consistent with the study’s analytical framework, the methodology unfolds in three distinct phases (see Fig. 2): data preparation (setting the data in a standard format, cleaning the data, extracting frequencies, and obtaining the sentiment lexicon); sentiment analysis (on the positive and negative, sentiment diversity) and emotion analysis of specific and salient features in the sentiment results; and keywords analysis and contextualization.

Fig. 2: Sentiment analysis workflow.
Fig. 2: Sentiment analysis workflow.
Full size image

This diagram outlines the workflow for sentiment analysis, from data prep to emotion detection.

First, for the English source text as well as the translated Chinese target texts, the tool Sketch Engine is employed to generate frequency statistics. To enable sentiment and emotion detection, the research has obtained the NRC Word-Emotion Association Lexicon (EmoLex) from the National Research Council Canada. This lexicon, annotated with positive and negative sentiment labels as well as eight basic emotions (JOY, SADNESS, ANGER, FEAR, TRUST, DISGUST, SURPRISE, and ANTICIPATION), was selected due to its high-quality annotations and expert-derived reliability (Taboada et al. 2011).

Second, Sentiment analysis at the polarity level is executed through frequency representation, focusing on top positive or negative words and translation examples. This entails generating word frequency for Wilde’s fairy tales and their reviews using Sketch Engine. Then, this study implement three quantitative metrics to discern similarities, differences, and potential tendencies among translations (see Table 2): the sentiment score (SentiS) reveals raw emotional balance (positive when >0, negative when <0, neutral at 0); the normalized sentiment score (NSentiS) scales sentiment intensity from −1 (completely negative) to +1 (fully positive), enabling cross-text comparison; and the relative sentiment ratio (RSenti) measures emotional dominance while preventing division errors (values > 1 indicate positive emphasis, <1 show negative leaning, and 1 represents perfect balance). These values are calculated according to equations (1)–(3), where k refers to the occurrence of positive words, i represents the occurrence of negative words, n+ is the total number of positive words, n- is the total number of negative words, and TWC is the number of the total word count:

$${SentiS}=\mathop{\sum }\limits_{k=1}^{{n}^{+}}{x}_{k}^{+}-\mathop{\sum }\limits_{i=1}^{{n}^{-}}{x}_{i}^{-}$$
(1)
$${NSentiS}=\frac{{\sum }_{k=1}^{{n}^{+}}{x}_{k}^{+}-{\sum }_{i=1}^{{n}^{-}}{x}_{i}^{-}}{{TWC}}$$
(2)
$${RSenti}=\frac{{\sum }_{k=1}^{{n}^{+}}{x}_{k}^{+}}{{\sum }_{i=1}^{{n}^{-}}{x}_{i}^{-}+1}$$
(3)
Table 2 Interpretation and summary of the formula.
Full size table

For Sentiment analysis at the emotion level, statistical analyses are conducted using the R programming language. Leveraging the Syuzhet package, the study employs an algorithm for emotion extraction to examine emotion distribution and specific emotion words relevant to the translation process. These parameters contribute to establishing a preliminary understanding of emotion representation.

Lastly, keywords analysis coupled with contextualization aims to unravel salient concepts identified in the preceding sentiment analyses. This step aims to answer the second research question dealing with the realization of sentiment disguise in online reviews, possibly unveiling the status and influencing parameters of sentiment disguise in the process of translation. This analysis employs KH Coder (Higuchi 2016), a robust software package specifically designed for quantitative content analysis and text mining (https://khcoder.net/en/). The keyword identification begins with automatic extraction of all words and phrases from the corpus of Goodreads and Douban reviews, followed by language-specific morphological analysis (for supported languages including English and Chinese) to identify base word forms. Subsequent refinement applies multiple criteria, including a frequency threshold, part-of-speech filtering, and stop word removal using built-in dictionaries, with final manual validation to ensure relevance. Contextual analysis involves generating co-occurrence networks to detect semantic term clusters and applying correspondence analysis to visualize relationships between keywords and review documents. The co-occurrence network analysis was conducted using KH Coder’s standard network visualization function (Manual Section A.5). The generated networks employ the software’s default parameters for calculating and displaying word associations, where connection strength is determined by co-occurrence frequency within document subsetsFootnote 3. The correspondence analysis utilizes a contingency table structured as a words-by-documents matrix. The analysis was performed using KH Coder with default settings, which automatically handles the construction of such matrices and applies standardization techniques (e.g., term frequency-inverse document frequency weighting and chi-square transformations) to help ensure robust and interpretable results.

Results and analysis

Polarity-level sentiment analysis

Table 3 summarizes the results of sentiment analysis with sentiment score (SentiS), normalized sentiment score (NSentiS) and ratio of positive and negative words (RSenti) of the source dataset and reception dataset. The two datasets are further differentiated by their subordinate constituents: original and translation, Goodreads and Douban, respectively.

Table 3 Polarity level sentiment scores.
Full size table

A general observation shows an increasing degree of positivity from the source dataset to the reception dataset. Evidence includes an increase in the SentiS value from 808 to 4,190, and an increase in the NSentiS from 0.097 to 0.158. Similarly, the RSenti value rises from 1.778 to 2.286. Contrary to the initially hypothesized heartbroken audience, the reception data present more positive and rich sentiments. These tendencies can be further reinforced by a qualitative look at book reviews, where positive sentiments such as “moved”, “shocked”, “admired”, and “love” appear before negative sentiments like “cry”, “heartbrokenness”, “tears”, and “despair” (see Book review 1 and 2). This pattern can be attributed to the mode of language. Written book reviews tend to convey more complex and nuanced sentiments compared to oral comments (Grosjean 1998). Additionally, the inclusion of stories from The Happy Prince and Other Tales in recommended reading lists for primary school children in China suggests an authoritative recognition and validation by educational authorities. This status as part of the cross-cultural literary canon may contribute to the increase in positive sentiments, a phenomenon that will be further explored in the upcoming keywords analysis and contextualization section.

Book review 1

最近收拾书柜, 又翻回小时候读的《快乐王子》, 感动、震惊、怀念、欣赏……我不知如何描述, 种种情绪姑且化为一句话——我只想流泪。 (I recently cleaned up the bookcase and turned back to the Happy Prince and Other Tales I read when I was a child. I was moved, shocked, missed, and admired… I don’t know how to describe it. All the emotions are temporarily reduced to one sentence - I just want to cry).

Book review 2

少年时读原文, 优美流畅的语言, 细细密密的织出了自然世界的和谐美丽, 织出了西班牙宫廷生活的阴郁生活, 织出了国王秘密的伤逝爱情, 织出了小公主童年的空虚寂寞。现在成年了, 读来, 既感到了爱情的悲悯, 又感到了生活的难解。不明白为什么王尔德的笔下会有那么多的心岁, 眼泪, 阴沉, 绝望, 而在他笔下这些往往衍生出更多的爱与希望。 (When I was a teenager, I read the original text, and the beautiful and fluent language carefully woven the harmony and beauty of the natural world, woven the gloomy life of the Spanish court life, woven the king’s secret sad love, and woven the emptiness of the little princess’ childhood. lonely. Now that I am an adult, when I read it, I feel the pity of love and the incomprehension of life. I don’t understand why there are so many heartbrokenness, tears, gloom and despair in Wilde’s writings, and these often lead to more love and hope in his writings.)

Translation (C) is the most salient when examining the frequency of sentiment scores further. It has the lowest SentiS (281), NSentiS (0.062), and RSenti (1.542) values, indicating a loss of positive sentiment in translation. This observation aligns with the findings of previous studies on sentiment analysis of Oscar Wilde’s stories in Chinese translation, which have similarly reported a growing prevalence of negative sentiments, accompanied by a decline in the frequency of positive sentiments (Liu 2023).

This observation prompts a closer examination of the translation mechanisms that contribute to this loss of positive affect. A micro-level analysis of specific translation examples offers insights into the phenomenon. In Translation Example 1, the original text maintained a relatively neutral tone, with the Mathematical Master simply inquiring about the possibility of knowing something unseen. However, the translated version introduced the adjective “stereotyped” to characterize the Master, imbuing the portrayal with a negative connotation that depicts the character as rigid and unyielding. This subjective lexical choice by the translator serves to reinforce the negative sentiment towards the character.

Similarly, in Translation Example 2, the translation altered the sentiment polarity from the source text. Whereas the original Rocket character expressed concern over a couple potentially losing their only son, the translated version has the Rocket stating “I never said they encountered misfortune”, shifting the sentiment to a more negative framing. These translation-induced changes in wording and narrative framing can be attributed to the translators’ decisions to add contextual information and objective details, particularly tailored for a child readership, in an effort to enhance comprehension. However, these interventions have inadvertently resulted in a dampening of the positive sentiments present in the original texts.

Translation example 1

Original: “How do you know?” said the Mathematical Master, “you have never seen one.” [Neutral].

Translation:“你们怎么知道的?”刻板的校长发问道, “你们又从来没见过天使什么样子。(“How do you know?” asked the stereotyped headmaster, “You have never seen what an angel looks like.”) [Negative].

Translation example 2

Original: I never said that they had, replied the Rocket; “I said that they might.” [Neutral].

Translation: “我从来没有说过他们赶上了倒霉的事, ”火箭答道, “我只是说他们也许会有不幸。” (“I never said they encountered misfortune,” answered the Rocket, “I only said that they might be.”) [Negative].

Context: But they have not lost their only son, said the Roman Candle; no misfortune has happened to them at all……. If they had lost their only son there would be no use in saying anything more about the matter. I hate people who cry over spilt milk. But when I think that they might lose their only son, I certainly am very much affected.

While the reception dataset exhibits increased positivity, the Translation dataset stands out with the lowest sentiment scores, indicating a significant loss of positive affect. This highlights the presence of sentiment disguise in the reception of The Happy Prince and Other Tales. Translation emerges as a potential influencer, with translators’ lexical choices and narrative framing inadvertently obscuring or inverting the original texts’ emotional qualities. This discovery necessitates further investigation as the study progresses to discuss the complex interplay between translation practices, readership reception, and cross-cultural literary canon formation.

Emotion-level Sentiment Analysis

The overall emotion level sentiment distribution of the sentiment analysis is exhibited in Fig. 3. The results indicate that the English-language corpora (Goodreads and Original) exhibit significantly greater emotional richness—measured as the relative frequency distribution of emotion-bearing words (i.e., the proportion of each emotion category within the total emotion-related lexicon)—compared to the Chinese datasets (Translation and Douban). This disparity is particularly evident in three key affective dimensions: JOY (Goodreads: 17.99% vs. Douban: 11.74%; Original: 15.54% vs. Translation: 9.75%), TRUST (Goodreads: 16.72% vs. Douban: 8.27%; Original: 14.90% vs. Translation: 7.98%), and ANTICIPATION (Goodreads: 12.22% vs. Douban: 7.41%; Original: 12.80% vs. Translation: 7.40%). This finding aligns with arguments made in previous studies that different languages possess distinct sentiment and emotion systems (Russell, 1991; Vaid, 2006). Consequently, the act of translation may exert a significant influence on the reception of cross-cultural literary canons, providing grounds for the possibility of occasional sentiment disguise.

Fig. 3: Emotion level sentiment distribution.
Fig. 3: Emotion level sentiment distribution.
Full size image

This chart visualizes the distribution of eight basic emotions (e.g., joy, sadness, anger) across different datasets, including original and translated texts from Goodreads and Douban.

Moreover, the Goodreads and the Original within the English dataset, together with the Translation and the Douban within the Chinese dataset, present a similar corresponding and consistent distribution pattern across emotion categories. If the previous polarity-level sentiment analysis was a tentative testification of sentiment disguise, the emotion level sentiment distributions and receptions appear more dependent on the specific texts (language) and expressions received and accepted by the readers. For example, the decrease of ANTICIPATION in Translation (compared to Original) might be causally linked to a lower occurrence of ANTICIPATION in the Douban compared to the Goodreads. Similarly, the slightly higher percentage of ANGER in Douban (compared to Goodreads) could be partially attributed to the increase of ANGER in the Translation compared to the Original.

With another look into the distribution by emotion category, we could observe that the most frequent emotions are JOY and TRUST, which is in agreement with previous sentiment findings that the values concerning degree of positivity are always above zero. Considering the previous claim that the Chinese audience of The Happy Prince and Other Tales portrayed themselves as heartbroken, we expect there to be an increase in SADNESS in the reception dataset. However, the stability/steadiness noted in the emotion SADNESS among all four datasets manifests rather contradictory evidence, that is, SADNESS remains almost unchanged across datasets. Nevertheless, if we combine the distribution pattern of emotions JOY and SADNESS, a conclusion could be made that, despite the unchanged trend of SADNESS, the reduction in JOY would indirectly reinforce the image of a less cheerful audience in the Chinese reception dataset. The conclusion is then supportive of the image representation of the heartbroken audience in China.

TRUST is another salient emotion distributed in different datasets together with JOY as discussed above. Differently, TRUST in Chinese exhibits a rather considerable reduction from the Translation dataset to the Douban dataset. According to the emotion taxonomy by Plutchik (1991), TRUST is derived from one of the basic “adaptative behavior” (p 57) summarized as Incorporation, that is, “the act of taking in or ingesting food” (p. 61), and is regarded as “a basic prototypic pattern of behavior indicating acceptance of stimuli from the outside world into the organism”. There seems to be no persuasive and self-evident reasons for the reduction in such a basic and somehow positive emotion. Conducting a targeted analysis of the top TRUST-associated lexical items presented in Table 4 provides a potential new interpretive prism. Several religion-related terms, such as “天使” (angel), “宗教” (religion), and “God”, emerge within this TRUST-centric word cluster. This suggests that the reduced TRUST levels manifested in the Chinese datasets may be linked to a loss of religious connotations and denotations during the translation process.

Table 4 Top TRUST words among different datasets.
Full size table

To further explore and substantiate this hypothesis, a granular examination of the translation process and the introduction of specific linguistic examples is warranted. It appears that certain religious elements (e.g., “thee”, “hath”, “art”, “thou”) present in the source material may have undergone a dilution or obfuscation of their sacred meanings in the Chinese translations. This linguistic and semantic transformation could potentially contribute to the considerable reduction in TRUST observed between the Translation and Douban reception corpora.

Translation example 3

Original: “Who hath dared to wound thee?” cried the Giant; “tell me, that I may take my big sword and slay him”.

Translation: “谁竟敢伤害你? ”巨人大声责问道, “快告诉我, 看我拿上我的大剑, 去把他砍了。” (“Who has dared to wound you?” cried the Giant; “tell me, that I may take my big sword and slay him”).

Translation example 4

Original: “Who art thou?” said the Giant, and a strange awe fell on him, and he knelt before the little child.

Translation: “你是谁? ”巨人问道, 一种莫名的恐惧传遍他全身, 他在小男孩儿面前跪下了。(“Who are you?” said the Giant, and a strange awe fell on him, and he knelt before the little child).

According to the Collins dictionary, the pronoun “thee” is an “old-fashioned, poetic, or religious word for ‘you’ when you are talking to only one person”. Similarly, the expressions “hath”, “art”, and “thou” carry distinct religious-tinged connotations. However, in the translated Chinese versions, the use of the standard pronoun “你” (you) and the verb “是” (are) represents more common, everyday language devoid of any overt religious denotations. Alternatively speaking, in the process of rendering into a recipient culture (Chinese) where religious elements may be subject to additional editorial consideration in publications (Aikman 2006), especially in readings for children, the religious elements together with their related emotions are dropped. Elucidated from the literary canon prism where canon formation is generally accepted as a “collective cultural process of value conferral upon works of literature” (Baker and Saldanha 2020: 52), the influencing factors of canonization contain not only pure linguistic lens into the quality of the text but also social-cultural or historical stances like race, gender, or regional diversity (Cawelti 1997; Lauter 1991). In brief, the religious-colored expressions are filtered to meet the general social-cultural standards in the recipient culture, which subsequently leads to the observed reduction of the receptive emotion TRUST.

Overall, the identified sentiment disguise is perhaps achieved via subtle emotion alternations. One the one hand, while polarity level sentiment reception is not profoundly affected by the translation, emotion level sentiment distributions and receptions are more dependent on concrete texts (languages) and expressions the readers receive and accept. On the other, the power of translation is becoming even more salient in the translation, especially in this case of the literary canon where the religious-colored expressions are filtered to meet the general social-cultural standards in the recipient culture.

Keywords Analysis and Contextualization

To address the third research question and elucidate the underlying grounds for the observed sentiment disguise, this section will employ a multipronged analytical approach involving (1) correspondence analysis to identify key conceptual patterns and associations, (2) a co-occurrence network to illuminate the clustering and interrelationships of salient thematic and semantic elements raised by reviewers across the Douban and Goodreads datasets, as well as (3) a close, qualitative examination of selected review excerpts to provide situated, contextual insights into the specific concepts, interpretations, and emotional responses articulated by the readerships.

The correspondence analysis, aiming to visualize and reveal the relative relationships between categories, may provide sentiment clues for further problematization and justification. The readers from Goodreads focus on the short story titles (see Devoted, Friend, Rose, Selfish, Nightingale, Rocket in the upper cluster of Fig. 4), characters (swallow, Happy Prince, Hans), the author (OSCAR, WILDE) and the most intensive cluster in the lower left section of Fig. 4 expressing feelings or emotions related to The Happy Prince and Other Tales. The concepts discussed are mostly closely related to and within the frame of the story settings (i.e., bird, garden, and so on), in line with the argument that reviews are “subset of characters and their relationships” (Shahsavari et al. 2020: 277).

Fig. 4: Correspondence analysis of the Goodreads dataset.
Fig. 4: Correspondence analysis of the Goodreads dataset.
Full size image

This plot displays the results of a correspondence analysis on the Goodreads dataset, showing how words and concepts (e.g., characters, themes) are distributed across dimensions based on their co-occurrence patterns.

However, reviews of the cross-cultural literary canon in Chinese dataset Douban entail a broader range of concepts, from the literary canon per se to the projection of the real society, resonating with their own situation. Evidence comes from Fig. 5 that except the plot-related words (夜莺 ‘nightingale’, 王子 ‘prince’, 燕子 ‘swallow’, 巨人 ‘giant’, 灵魂 ‘soul’), author-related words (王尔德 ‘Wilde’, 作者 ‘authour’), and emotion-related words (美丽 ‘beauty’, 悲剧 ‘tragedy’), there are words describing the actual society like现实(reality) in the correspondence analysis. Examples could be found from the reviews within actual context:

Fig. 5: Correspondence analysis of the Douban dataset.
Fig. 5: Correspondence analysis of the Douban dataset.
Full size image

This plot presents a correspondence analysis for the Douban dataset, highlighting key terms and themes in Chinese related to the reviewed literary works.

Book review 3

小学一年级的时候, 中央台播《快乐王子》的动画片, 看完我就大哭起来。说不出来的难过, 好伤心。那个时候它还不叫CCAV, 那个时候我还不知道王尔德是谁。我想王子为什么要让自己变成那般丑陋残缺, 小燕子为什么不飞到温暖的南方? 若干年后, 当我在历史课本上读到谭嗣同, 在《门》中读到老舍投湖, 都不能抑制同样的泪水。他们都拥有纯白的灵魂。(When I was in the first grade of elementary school, CCTV broadcasted the cartoon Happy Prince and Other Tales, and I burst into tears after watching it. Unspeakable sadness, so sad. It wasn’t called CCAV at that time, and I didn’t know who Wilde was at that time. I think why did the prince make himself so ugly and mutilated, and why didn’t the little swallow fly to the warm south? A few years later when I read about Tan Sitong in my history textbook and Lao She committed suicide by throwing himself into a lake in The Gate, I couldn’t hold back the same tears. They all have pure white souls).

Based on Book review 3, the memories and tears triggered by the literary canon The Happy Prince and Other Tales connect the western fairy tales to eastern cartoons, the fictional characters to living human beings, and Victorian Britain (when Wilde lived) to Late Qing China (when Tan Sitong and Lao She lived). The process of Wilde’s canon formation, in this case, transcends cultural boundaries thus legitimating the afterlives of the originally cannoned or non-cannoned literature into a different culture (Li 2013). Each time literature crosses cultural boundaries, its literary value is reproduced in the form of translation if the boundaries of the notion translation are to be expanded (Kolbas 2018). Both translation and literary scholars have scrutinized the inherent dynamics in the process of literary works becoming canonized via translation and agreed that literary quality is now becoming merely one of the influencing factors on the basis of the eloquent argument that “judgments of quality are extremely context-bound” (Baker and Saldanha, 2020: 54). Consequently, the concepts from book reviews of canonized cross-cultural literature The Happy Prince and Other Tales echo the idea of contextualization by translation and literary scholars alike. Conclusions could therefore be made that the main concepts of emotive reception are expanded via the cross-cultural translation activity.

As the concept and noteworthiness of translation has been foregrounded, the co-occurrence network (see Fig. 6) further explores the main concept “翻译” (translation) that might lead to divergent emotive receptions in different readers. It is observed from the concepts in Fig. 6 that the readers’ attitudes towards translation are largely negative: 1) words associated with translation errors (错误‘error’,失‘lost’); 2) words associated with translator’s inappropriate styles (文绉绉‘genteel’, 蹩脚‘crappy’, 失色‘eclipse’, 拙劣‘clumsy’, 简陋‘shabby’); 3) words associated with translation as a painful activity (苦功夫‘hard work’, 再创作‘recreation’). Despite the rare occurrences of some positive words 雅(elegant) and 诗画 (poetic), the necessity of recontextualizing these words could not be neglected, and the contextualized book review 4 is shown as follows:

Fig. 6: Co-occurrence network of selected word “翻译” (translation).
Fig. 6: Co-occurrence network of selected word “翻译” (translation).
Full size image

This network visualizes concepts related to “translation,” revealing a mostly negative reception through terms linked to errors, awkward style, and difficulty, with rare positive terms requiring contextual interpretation.

Book review 4

译文和原文给人的感觉差距竟然如此之大, 巴金先生翻译的反而更适合小孩子阅读, 而失了原文那种诗画般的、古老的、娓娓道来的行文风格。(The difference between the translated text and the original text is so great that Mr. Ba Jin’s translation is more suitable for children to read, but loses the poetic, ancient and eloquent style of the original text).

Surprisingly, the observed positive word 诗画 (poetic) is negated by the negative indicators lose, hence, the meaning expressed is completely controversial and contradictory to what was previously regarded as a positive one. Book review 5 further confirms the existence of negative sentiments held by readers towards translation, while the specific words they resort to are seemingly positive:

Book review 5

私以为外文著作一经译著, 腔腔调调总会发生变化, 其中的情感也不如起初那样纯粹。你总不能指望野外的玫瑰经花匠采摘后放入一尊玻璃皿后依然保持鲜活的夜色。翻译的确是个苦功夫, 文化的差异, 思想的迥同, 很难保证译文还是原本鲜活的颜色而不是译者夹杂个人情感的再创作。无论如何, 读者若是能够体会到哪怕是一点点的关于美的遐想与纯净的怜悯, 译者的功夫也没有白费。王尔德的著作译者众多, 也有出版商做出的噱头。如果有条件的话, 还是去看看原著吧。(I personally think that once a foreign language work is translated, the tone of voice will always change, and the emotion in it is not as pure as it was at first. You can’t expect a rosette in the wild to stay alive at night after being picked by a gardener and placed in a glass dish. Translating is indeed hard work. Due to cultural differences and very similar ideas, it is difficult to ensure that the translation is still the original vivid color rather than a re-creation of the translator’s personal emotions. In any case, if the reader can experience even a little bit of reverie about beauty and pure pity, the translator’s efforts will not be in vain. Wilde’s work has been translated by many translators, as well as gimmicks made by publishers. If you can, go and read the original).

This is similar to the idea that readers of books translated from other languages are uncertain about their assessments because the readers often cannot or do not read the original texts. And “if they dislike something about the book, the translation is often painted as the possible scapegoat” (Kotze et al., 2021: 20). Consistent with the argument that translation is generally scapegoated, the negative attitudes towards translation observed in the co-occurrence analysis partially contributed to the increased SADNESS and heartbroken image of the audience as discussed at the beginning of the paper. In summary, the sentiment disguised in the reception of Wilde’s children’s literature could at first be accounted for from the cross-cultural perspective and the translational perspective as well.

Conclusion

This study has investigated the phenomenon of sentiment disguise in the specific cross-cultural case study of the reception of Oscar Wilde’s children’s literature. Through an examination of readers’ perceptions and emotive responses to both literary canon and its translations, this research has identified potential indicators of sentiment disguise in the reception of Oscar Wilde’s children’s literature that warrant further investigation. This observation was achieved through a meticulous analysis and comparison of sentiment values and distributions. Subsequently, the realization particulars of sentiment disguises are probed into from emotion level sentiment analysis, for example, distribution by category, top words inventory, and exemplary explanations. This sentiment disguise is achieved when emotions are altered in the process of translation and changes in socio-cultural standards. Expositions stemming from the keywords and contextualization analysis tentatively provide arguments from the cross-cultural and translational perspectives.

While this study has provided valuable insights into sentiment disguise in the context of literary reception, it is important to note that the findings may not fully generalize to other genres beyond the literary domain. The examination of a specific genre, such as children’s literature, could be subject to unique dynamics that may not translate directly to alternative forms of cultural expression, like speeches or political discourse. Therefore, future research would greatly benefit from expanding the investigation of sentiment disguise into a wider range of cultural artifacts. Cross-genre explorations would not only validate the insights from this study but also uncover new dimensions and nuances that could further elucidate this phenomenon. Pursuing these avenues of investigation would contribute to a more robust and generalizable understanding of sentiment disguise in cultural transmission and reception.