With the exponentially increasing volumes of rich, written textual data generated via consumer–brand interactions in online spaces, harnessing this data to understand and measure its marketing implications has become an exciting opportunity for marketing managers. However, given the increasing sophistication and complexity of written communication in online spaces, it is becoming harder to analyze and comprehend the impact of written consumer communication for brands. Now, a cutting-edge new Journal of Marketing Research article provides an easy-to-use tool that is grounded in nonverbal and sensory marketing theory.
The researchers developed and validated a paralanguage classifier (called PARA) using social media data from over 1.2 million Twitter, YouTube, and Instagram posts. Whereas the currently popular text analysis tools explore verbal aspects of what is being conveyed, this research focuses on the subtleties of how the text is written. “Textual paralanguage” (TPL) is the measure of nonverbal characteristics of a text that extend beyond the literal meanings of words and parts of speech. The PARA software provides output variables at a superordinate sensory level, such as auditory TPL (which includes elements denoting tone, rhythm, stress, emphasis, and tempo of the text), tactile TPL (which includes tactile emojis and emoticons), visual TPL (bodily/nonbodily emoticons/emojis), and aggregate variables (emoji count, emoticons index, and TPL index). The PARA tool uses a rule-based algorithm to analyze textual data.
Implications for Marketing Managers
Feedback forums are rich sources of customers’ opinions and emotions. By applying PARA to measure the intensity of sentiment valence on these platforms, managers can prioritize responses and devise better communication strategies for faster and more effective resolution. Moreover, PARA can also help brands to convey persuasive messages with nonverbal cues that enhance the appeal of their marketing content. Social media is not only a powerful tool for firms but also for individuals such as artists, celebrities, and influencers who seek to attract consumer engagement. PARA can enable such emerging brands to create captivating content with nonverbal cues that inspire people to interact. As the authors suggest, PARA can also serve as a valuable tool for influencer vetting to ensure that the content produced by influencers is consistent with brand communication.
The PARA tool, currently in beta testing, is available here.
Overall, the research shows that paralingual text analysis can help us understand how TPL elements communicate the visual as well as auditory aspects of a text and can be used to study the influence on consumer perception and response. Although extrapolating the exact meaning of the language used in a textual content has been a tough nut to crack, this study provides an easy-to-use tool to understand the tough layer of non-verbal cues in the text.
We interviewed the authors to learn more about this exciting project:
Q. What drove your motivation to analyze auditory, visual, and tactile elements in online textual content? Could you explain the methodology used to develop the PARA tool in layperson’s terms?
A: In this work, we developed a text analysis tool called PARA to identify nonverbal communication cues in text. We realized that many text tools focus on what is said—the verbal content of a message or the words themselves. What makes PARA different is that it identifies 19 features of nonverbal parts of speech—focusing on how something is said—or what’s called textual paralanguage. Simply put, PARA has been built on a set of rules as well as dictionaries that identify keywords.
Q. How is nonverbal communication in online text different from that of traditional face-to-face interactions and are there any potential ethical considerations for using the PARA tool, especially in the context of social media privacy?
A: In-person communication is riddled with nonverbal cues (e.g., facial expressions, vocal tone). In text-based communication, communicators alter the message to add this nuance. People are creative with how they bring the text to life: they might add capitalizations, elongation of words, repeated punctuation, rhythmic punctuation, emojis, emoticons, asterisks, etc. PARA captures these elements.
As with any online content, privacy is always a consideration. It is important to realize that content posted on social media is often public and depends on the privacy settings of a particular user or platform.
Q. Does the textual paralanguage tool capture the subtle usage of linguistic elements (such as the use of wordplay), which are crucial for content success yet marginally captured by traditional textual analysis methods?
A: While PARA does not directly target the use of wordplay, its primary focus is on identifying textual paralanguage in online written communication, which is often overlooked and minimally detected using conventional text analysis methods. In our article, we present empirical evidence highlighting the role that PARA plays in successful online communication. This is demonstrated through its ability to shape content sentiments and predict user engagement, thereby serving as an important driver of content success. It is important to note that while PARA does not explicitly identify wordplay, it excels at detecting the textual paralanguage used alongside it. PARA provides a rich, multidimensional view of online communication by enhancing our understanding of text sentiment.
For example, consider this dialogue:
Person A: “Why did the bicycle fall over? 🚲😉”
Person B: “Hmm… I’m not sure, why did it fall over? 🤔”
Person A: “*drumroll* Because it was two-tired!!!”
In this instance, while PARA would not directly recognize the wordplay (‘two-tired’), its ability to capture textual paralanguage cues (e.g., emojis, emphasis, vocalizations, etc.) helps inform us about the playful, joking context in which the wordplay is used, enhancing the holistic understanding of the text.
Q. How does the tool treat the following categories: (1) only text, (2) only emojis, and (3) text + emojis? Can the tool be used to analyze sentiments in a specific linguistic category? If so, how should researchers deal with these limitations?
A: PARA is a linguistic tool designed primarily to detect and classify textual paralanguage. The paralinguistic features it detects include not only emojis but also alphanumeric (keyboard-based) nonverbal cues, with 15 features that can be operationalized using text alone. In our article, we demonstrate that disregarding these textual paralanguage features can result in missing significant explanatory and predictive power when analyzing sentiment valence and intensity. This suggests that PARA is an indispensable tool for sentiment analysis, regardless of whether the data comprises only text, only emojis, or a combination of the two. Moreover, when used in conjunction with traditional sentiment methods such as VADER or LIWC, PARA can enhance the accuracy of sentiment analysis. This is because it captures variations in sentiments that might be overlooked when relying solely on text-based verbal cues. Of course, with any methodology, there are limitations. PARA uses human-generated rules and dictionaries based on training data to detect textual paralanguage. Such rule-based detection may not be exhaustive, and there are likely rare and nuanced forms of paralanguage that remain undetected, which could be further investigated in conjunction with generative AI. In addition, the features detected by PARA are rooted in the English language. There are many future research opportunities to expand PARA to detect nonverbal features in other languages.
Q. Given the highly subjective nature of emotions, how do you approach the potential ambiguity of nonverbal cues in the text, such as emojis, and the challenge of accurately determining their precise emotional implications?
A: In our study, we focus primarily on sentiment valence (i.e., the positivity/negativity of a message) and sentiment intensity (i.e., the degree of positivity/negativity of a message). PARA is not designed to detect specific emotions or make emotional inferences from emojis. In other words, our approach allows us to identify the influence of nonverbal cues across various contexts and social media platforms, independent of the specific emotion being expressed.
A natural next step is to further investigate whether certain nonverbal features convey specific emotions better. For example, certain paralinguistic features may better indicate happiness, whereas others convey pride. It is important to consider the impact of each nonverbal feature in different contexts, particularly how verbal and nonverbal text features interact to produce emotional effects. We consider this to be an important area for future research.
Q. Could you share any specific applications or cases with marketing researchers and professionals where PARA can effectively predict consumer engagement? What is the easiest way for marketing researchers to access these tools?
A: In our article, we demonstrate the effectiveness of PARA in predicting consumer engagement using social media data from platforms such as Twitter, YouTube, and Instagram (N = 1,241,489 posts). PARA proves valuable in predicting consumer engagement metrics, including the number of likes and retweets, even after controlling for a wide range of verbal features using existing text tools. PARA will be particularly relevant when analyzing text content that has been generated on mobile devices. Text from cell phones as opposed to computers is more likely to include paralanguage because individuals have easy access to emoji keyboards, and on phones, people generally write more brief, more emotional, and more “gist-based” content. Textual paralanguage is particularly relevant for marketing managers involved in content creation and consumer listening. Whether replying to customers on Twitter, managing a brand’s Instagram page, or composing emails to notify customers of important updates, marketing managers should consider the way content is written as it significantly impacts consumer interpretation and engagement. To access the tool, visit https://www.textualparalanguage.com/. You can download the PARA software for free, and it is compatible with both Mac and Windows systems. Researchers or practitioners interested in the PARA Python source code can contact us for access (email@example.com).
Read the Full Study for Complete Details
Read the full article:
Andrea W. Luangrath, Yixiang Xu, and Tong Wang (2023), “Paralanguage Classifier (PARA): An Algorithm for Automatic Coding of Paralinguistic Nonverbal Parts of Speech in Text,” Journal of Marketing Research, 60 (2), 388–408.
Go to the Journal of Marketing Research