A systematic review on cross-culture, humor and empathy dimensions in conversational chatbots: the case of second language acquisition

Abstract

The advancement of information and communication technologies has led to an increasing use of conversational chatbots in the learning and teaching sector, especially for the second language (L2) acquisition. In the field of second language acquisition, the use of AI chatbots has been explored, mainly studying pedagogical approaches. However, there is a limited study in the development of empathetic strategies for dealing with learners' emotional discomfort, the impact of humor and the consideration of learners' cultural backgrounds. Thus, this study reviews the existing studies on AI second language (L2) chatbots to investigate the development of empathetic strategies for enhancing learners' learning outcomes. To achieve the aim of this study, prior studies from 2012 and 2022 of several popular databases, including Web of Science, ProQuest, IEEE and ScienceDirect are collected and analyzed. This study found that three dimensions such as cultural, empathetic and humorous dimensions have a positive influence on the application of AI L2 chatbots for enhancing learners' learning outcomes. This study also found that the development of an AI chatbot in L2 education has plenty of room for improvement. Several recommendations are made for enhancing the use of AI L2 chatbots which include integrating cross-cultural empathetic responses in conversational L2 chatbots, identifying how learners perceive and react to the learning content, and investigating the effects of cross-culture humor on learners’ language proficiency.

Artificial intelligence; Chatbot; Second language learning; Culture; Empathy; Humor.

1. Introduction

An artificial intelligence (AI) chatbot is a software program which stimulates natural human dialogue through text or text-to-speech functions with the assistance of natural language processing (NLP) using machine learning and commonly deep learning approaches (Zhai et al., 2022). Apart from being efficient, simultaneous and flexible, the increasing popularity of conversational chatbots has drawn great attention to interpersonal communication with users, attending actively to their problems, showing empathy and possessing a well-timed sense of humor. These dialogue-based chatbots integrated with humorous and empathetic dimensions aim to be more emotionally responsive and human-like in various applications such as healthcare, marketing, commerce, retail, education, and learning (Chocarro et al., 2021). Studies show that AI chatbots embedded with humorous and empathetic dimensions have enhanced user satisfaction and task outcomes (Liu et al., 2020; Santos et al., 2020). Popular AI chatbots use emotional and humor algorithms to meet users' needs for companionship, social interaction, and a sense of belonging (Chen et al., 2020). Previous research has shown the effective use of chatbots for developing language skills in one's native and second language (Zhai et al., 2022).

To be able to learn a second language effectively, it is important for L2 learners to master the culture of the particular language of interest. One of the fundamental motivations for learners is to immerse themselves in a foreign culture in order to broaden their worldviews and various life experiences (Zhai et al., 2022). However, limited studies have explored the cultural dimensions when designing an L2 chatbot for facilitating learners to experience flexible and efficient intercultural communications. Previous studies have shown that AI chatbots embedded with humorous and empathetic dimensions can enhance user satisfaction and task outcomes (Liu and Sundar, 2018). Chocarro et al. (2021) state that these dialogue-based chatbots are useful for dealing with emotionally responsive and human-like applications in healthcare, marketing, commerce, retail, education and learning.

Existing computational humor studies primarily comprise of the computer generation of small subclasses of humor (Farah et al., 2021). For instance, the use of virtual social chatbots attempts to mimic the style of human interaction to prevent frustration in a more natural and more enjoyable way (Hong et al., 2014); and the use of automatic humor recognition techniques as a medical assistant by focusing on a humorous or satirical approach to get patients engaged (Fadhil and Schiavo, 2019). However, it seems that most computational humor studies overlooked L2 learners' cultural backgrounds when simulating a sense of humor to facilitate learners' language learning. As humor is universal but culturally tinted (Jiang et al., 2019), humor is significantly intertwined with language and culture, and it needs a significant cognitive effort to fathom its essential meaning behind words and phrases. It impinges on a shared language or set of cultural norms to function (Zhai et al., 2022). Humor as a culture-bound component is an indispensable part of L2 acquisition, and understanding the target language's culture is no exception (Lu et al., 2019). Studies on the use of humor showed that being linguistically competent was not necessarily pragmatically sufficient for L2 learners to perceive humor attached to a specific culture (Florence Ma, 2012; Olajoke, 2013). Specific cultural dimensions and values influence humor perception, humor manipulation and linguistic usage (Archakis et al., 2014). Increasing cultural differences in humor promotes student enjoyment of learning and creative understanding of the target language (Shardakova, 2013). On the other hand, embedding humor in a certain cultural facet improves a learner's linguistic competency (Neff and Rucynski, 2017).

Similarly, existing computational empathetic studies have been carried out to enhance humans' mechanical and computational capabilities in dealing with the detrimental ramifications of social exclusion (Gennaro et al., 2019). Empathetic chatbots have been primarily applied to supportive approaches to examine the user's emotions and generate appropriate responses with high empathic precision (Liu and Sundar, 2018). These studies have focused on three types of empathy: cognitive empathy, emotional empathy, and compassionate empathy (Zhai et al., 2022). However, it seems that these studies have ignored the significance of cross-cultural features in the empathetic modeling design when students experience emotional distress throughout the course of their education. Zhai et al. (2022) argue that it is important for novice teachers to examine students' cultural backgrounds via cognitive or affective empathetic communication. Teachers' empathetic support does not only develop a type of emotional rapport with students by leveraging sustaining cultural practices but also contributes to the dynamics of the learning environment and students' academic success while building connections across disparate social and cultural contexts (Bozkurt and Ozden, 2010; Zhai and Wibowo, 2022).

Based on a review of the existing literature, three research dimensions including empathy, humor and cultural have not been fully explored in AI chatbots for L2 acquisition, as previous studies have only focused on culture, empathy, and humor in hospitality, medical systems, and eCommerce (Wang et al., 2003). Therefore, this paper investigates the current status of conversational L2 chatbots embedded with culture, humorous, and empathetic dimensions. This paper conducts a systematic review to investigate studies about L2 conversational chatbots and identify research gaps in knowledge in embedding cross-cultural, humorous, and empathetic dimensions in the L2 learning environment. Furthermore, this paper analyzed and synthesized previous studies by considering learners’ psychological well-being during their study. Three research questions were therefore formulated concerning L2 chatbots as follows:

RQ1

How do cultural dimensions of conversational chatbots affect L2 learners' learning outcomes?

RQ2

How do empathetic dimensions of conversational chatbots influence L2 learners' learning and retention?

RQ3

How do humorous dimensions of conversational chatbots improve L2 learners' learning and retention?

2. Systematic review method

This study adopts a systematic review based on Virtanen et al. (2018), and it aims to set the ground for future research regarding L2 conversational chatbots by integrating humorous interactions and empathetic responses. A systematic review has been conceptualized as a “form of knowledge synthesis that addresses an exploratory research question aimed at mapping key concepts, types of evidence, and gaps in research related to a defined area or field by systematically searching, selecting, and synthesizing existing knowledge” (Colquhoun et al., 2014, p. 1284). A systematic review helps map the landscape of published literature and helpful in identifying research gaps, establishing research agendas, and offering recommendations for policymakers (Tricco et al., 2018).

2.1. Search strategy

This study adopts a search approach to retrieve journals published between 2012 and 2022 in order to examine the advancements of chatbots over the last decade. The databases include academic databases such as Google Scholar, Web of Science, ProQuest, IEEE and ScienceDirect. The following search phrases were used in the search: “artificial intelligence”, “second language”, “cultural backgrounds”, “conversational chatbots”, “empathy” and “humor”. Combined search phrases were used either in pairs or multiple phrases including single or double quotation marks. There are 420 articles from Web of Science, 161 articles from ProQuest, 4 articles from IEEE, and 4 papers from Science Direct.

2.2. Inclusion and exclusion criteria and selection of studies

To be included in the review, these journal articles needed to focus on the following selection criteria: (a) published in English in a full-text article, (b) associate chatbot with L2 learning or any other related second language learning; (c) provides empirical results, and (d) published between 2012 and 2022. In addition, articles were excluded if they did not focus on existing L2 chatbot studies, domains other than L2 acquisition, cultural dimensions, empathetic dimensions, and humorous dimensions. The inclusion and exclusion criteria are summarized and presented in Table 1. This article selection follows four-tier guidelines for systematic review and PRISMA Preferred Reporting Items: identification, screening, eligibility, and inclusion. PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. It is a minimal set of elements for reporting in systematic reviews and meta-analyzes that is based on evidence. The PRISMA statement includes a 27-item checklist and a four-phase flowchart. PRISMA is designed to make systematic reviews transparent with the intention of assisting writers in producing critical appraisal of systematic reviews (Sarkis-Onofre et al., 2021).

Table 1.

Inclusion and exclusion.

Inclusion	Exclusion
(a) Articles must be published in English in a full-text article, (b) Articles must associate chatbot with L2 or any other related additional language learning; (c) Articles must provide empirical results, and (d) Articles must be published between 2012 and 2022.	(a) Articles did not focus on chatbot studies, (b) Articles focused on domains other than L2 acquisition, (c) Articles were not focused on chatbots embedded with culture dimension (d) Articles were not focused on chatbots embedded with empathy dimension (e) Articles were not focused on chatbots embedded with humor dimension (f) Articles were written in other languages.

To ensure the relevance and quality of the articles, the literature that emerged from the databases has been screened twice according to the PRISMA process. The first phase includes examining article titles and abstracts to determine whether they were relevant to the research and met the inclusion requirements. The articles with full text were then selected for the next stage, and then the rest of the retrieved citations were uploaded into the Endnote 9 reference management software, and duplicates were eliminated from the database.

2.3. Data analysis and synthesis

Based on the review process, a total of 48 articles were selected for further examination and evaluation. In this study, a systematic review was conducted in which data were gathered, processed, identified, and summarized. A six-step procedure was employed to determine the recurring dimensions. The first phase involves conducting a thematic analysis of the data to acquire a more in-depth understanding. Then, the initial codes were created. The third and fourth phases entail identifying sub-dimensions and reviewing the sub-dimensions. The fifth phase consists of compiling all of the pertinent concepts. Finally, the data were analyzed to ensure relevance to the study's objectives.

This study has examined the use of chatbots for L2 acquisition in which embedded cultural, empathy and humor dimensions. The research articles retrieved from the databases were screened twice. First, an examination of the abstracts and titles of the retrieved material is carried out in order to determine whether these articles match the minimal inclusion criteria. Second, the entire text of the publications that were included in the study was analyzed and retrieved using the CQUniversity database retrieval tool. The databases yielded a total of 589 articles. After screening and removing duplicates and unsuitable content, the total number of articles was decreased to 313 articles. Then, after the eligibility screening procedure, a total of 48 articles were retrieved for this study. Figure 1 shows the distribution of publications that were published between 2012 and 2022.

Figure 1.

The PRISMA flowchart.

3. Results and discussion

This study investigates the impact of cultural, empathetic and humorous dimensions of conversational chatbots on L2 learners' learning outcomes for improving L2 learners’ efficacy and retention from 2012 to 2022. Based on the systematic review, three main dimensions with related sub-dimensions were identified out of 48 studies concerning different types of influences on L2 chatbots in second language acquisition. They are: (1) cross-cultural dimensions, with the sub-dimensions on cultural awareness and heterogenous cultures; (2) empathetic dimensions, with the sub-dimensions on emotional support and affective strategies; and (3) humorous dimensions, with the sub-dimensions on knowledge base and humorous response as shown in Figure 2. The findings of the search were presented in accordance with the questions raised in the Introduction section.

Figure 2.

Three dimensions of L2 Chatbots for second language acquisition. NOTE: = Cross-Cultural Dimensions. = Heterogenous Cultures. = Emotional Support. = Affective Strategies. = Knowledge Base. = Humorous Response.

The findings indicate that L2 chatbots embedded with cross-cultural dimensions, empathetic dimensions and humorous dimensions are able to (a) foster learners' engagement as well as increase different cultural awareness, (b) detect learners' emotions, provide feedback to reduce learners’ learning anxiety, and (c) establish a healthy relationship with L2 chatbots which are regarded as being humorous and friendly. In addition, several challenges are identified relating to the implementation of L2 chatbot embedded with cross-cultural dimensions, empathetic dimensions and humorous dimensions. These challenges include the need for algorithm design for culture, humor and empathy and the performance of L2 chatbots. Table 2 presents information about the articles that were selected, such as the names of the authors, the research purpose, participants, cross-cultural influence, humor dimension, empathy dimension and research outcomes.

Table 2.

Studies on chatbots.

Authors	Name of the Chatbot	Research Purpose	Participants	Outcomes
Ahmadi et al. (2017)	An anthropomorphic pedagogical agent	To investigate the viability of agent-based instruction for idiom learning	30 participants	The result revealed that agent-based education was considerably more effective in teaching English idioms
Al-Kaisi et al. (2021)	Alice	To evaluate the efficiency of Alice for students learning Russian as a foreign language	Russian as foreign language students	Alice could be used to teach the Russian voice assistant to international students who are studying Russian
Ayedoun et al. (2015)	Multimodal response generator	To foster L2 learners' willingness to communicate in L2	N/A	The approach decreases students' nervousness while increasing their self-confidence
Ayedoun et al. (2019)	Conversation agent, Jack	To foster learners' readiness toward communication in a second language	40 Japanese university students	Enhancing L2 learners' participation in contact and communication
Ayedoun et al. (2020)	An embodied conversational agent, Peter	To encourage students to converse in the target language.	60 university students	Learners' predicted communication willingness tended to increase with systems
Chen et al. (2020)	Anthropomorphized chatbots for migrants	To comprehend how migrants can be helped in the development of a chatbot that promotes social integration.	3 immigrants	The chatbot is reliable and efficient
Dai et al. (2014)	Multi intelligent agents	To utilize emotional intelligence may aid in detecting the learner's emotional responses and fostering his desire to learn	N/A	Learners' emotional experiences have a substantial influence on their motivation and results
Danilava et al. (2013)	An arrtificial conversational companion	To aid advanced language learners in practicing communication through instant messenger discussions	13 participants of German learners	Need to make the machine seem clever or amusing while maintaining its courteous manner
Dokukina and Gumanova (2020)	Vasya	To teach English via textual instruction and both auditory and visual practice	N/A	Vasya is an excellent tool for English language learners at the elementary level, but it is not enough for those at the intermediate or advanced levels.
Farah et al. (2021)	An ambiguous chatbot	To investigate the usage of a chatbot to help learners' conversational practice with such humor	400 participants	Quantitative research revealed that participant ratings of the chatbot's personality, wit, and friendliness increased significantly when it recognized the humor resulting from the misplaced modifier
Freidkin and Ksenofontova (2021)	Telegram-bot “Pardes”	To increase cognitive activity among students of Hebrew as a foreign language	1000 students from Israel	Lexical competence is enhanced by the introduction via translation, voicing, explanation, drilling, and controlling the process of vocabulary skills
Gayathri (2021)	Automatic mega-agent managed user guide (AMMU)	To utilize AMMU to Improve speaking abilities and lessen speaking anxiety	N/A	AMMU not only enables users to connect with one another verbally but also enhances their listening skills
Goga et al. (2021)	Rose	To propose a platform that combines academics and students to better understand student needs and strengthen the relationship between students and faculty	100 students	The performance of students is one of our platform's primary concerns, and it also seeks to give assistance to the cultural variances that exist
Gonulal (2021)	Intelligent personal assistant (IPA): Google Assistant	To examine language students' responses to IPA's humorous comments	42-second language learners	The findings showed that the students thought that the majority of IPA replies were highly hilarious, and they ranked the IPA's overall sense of humor as being higher than average
Gordon et al. (2016)	The Tega robot platform	To adjust the student's emotional state throughout the tutoring session in order to enhance long-term learning gains	34 children in preschool	The affective reinforcement learning algorithm, which was customized for each child and led to a considerably enhanced long-term pleasant valence, was successful in producing these results
Hong et al. (2014)	The courseware of animated pedagogical agent	To examine the educational effectiveness of animated agent-based instructional materials	63 Taiwanese elementary students	Characters that are animated to look like real people make learning more enjoyable for students
Ismail (2020)	A sensor-free, emotionally intelligent tutoring system	To research how to identify foreign language anxiety (FLA), as well as how to minimize and ultimately overcome FLA	180 L2 learners in America	Utilizing a sensor-free anxiety detector and making adjustments to the system are two potential approaches for mitigating the anxiety experienced by L2 learners
Johnson (2019)	Enskill English's chatbot	To investigate the possibility of utilizing data to monitor the development of learning over time	72 learners	Enskill English helps learn how to speak English, and there are signs that a learner's performance will improve as they use it. Ellie has the potential to be a good learning partner for L2 learners.
Kim et al. (2022)	Ellie	To evaluate the language proficiency, conversational skills, and task performance of students	137 high school students in South Korea	Enskill English helps learn how to speak English, and there are signs that a learner's performance will improve as they use it. Ellie has the potential to be a good learning partner for L2 learners.
Lee et al. (2012)	Mero and Engkey	To test our instructional robots' effectiveness	24 elementary students	There was a significant improvement in oratory ability. In addition, it has shown that the systems enhance students' happiness, curiosity, confidence, and motivation
Liang (2012)	Virtual world ​Second Life	To incorporate pedagogical practices of classroom interactions, online dialogues, and student reflections into an EFL course in order to promote students' comprehension and application of ELF	16 students	The outcomes of conversation analysis offer an in-depth examination of how contextual supports and semiotic resources supported students' usage of ELF
Lin et al. (2015)	Affective tutoring system	To identify students' emotional states during the learning process, the study hoped to enhance students' learning motivation and offer suitable feedback	100 Japanese college students in Taiwan	ATS recognizes the emotions of the learners and delivers relevant feedback to reduce their exam anxiety by providing repeated practice and reducing their communication uncertainty and fear
Litman and Forbes-Riley (2014)	ITSPOKE (Intelligent Tutoring SPOKEn dialog system)	To assess a spoken conversation system that identifies and responds in real-time to user disengagement	N/A	Including emotive aspects in a chatbot improves its performance.
Liu et al. (2020)	chatbot for the field of educational technology (CBET)	To design a chatbot that is domain-specific and sensitive to limited domain inquiries	52 participants	CBET is a very efficient information retrieval tool in a certain area
Lorenzo et al. (2013)	Massively multiuser online learning (MMOL)	To propose the construction of a learner role-play to enhance the foreign language abilities of learners	35 international students at the University of Alcalá	Such a setting encourages an immersive, creative, and collaborative foreign language learning experience
Lubold et al. (2021)	Social robot, Quinn	To investigate how social conversation and entrainment influence self-reported and behavioral rapport responses	48 college students	A social, trainable robot is capable of producing self-reported and linguistic responses that are nuanced and based on individual variations
Luo and Cheng (2020)	A collaborative multimedia teaching ​model	To investigate the use of artificial intelligence technologies in individual foreign language learning	N/A	An interactive platform for teaching foreign languages
Mageira et al. (2022)	AsasaraBot is introduced to the students as the Goddess of Snakes	To teach cultural information to high school pupils in a foreign language	61 Greek students who study both English and French	AI chatbot technology for interactive ICT-based learning is appropriate for the study of foreign languages and cultural material
Mazur et al. (2012b)	A pedagogical conversational agent	To offer a free-flowing discussion system for English as a second language tutoring	N/A	Specific emotionally charged phrases may enhance learning
Mazur et al. (2012a)	Artificial language tutors	The Co-Mix project examines the influence of emotions on problem-solving by focusing on second language vocabulary learning	Various Japanese students	The code-mixing technique is an efficient approach to introducing new vocabulary units
Meng-yue et al. (2020)	Intelligent English culture teaching auxiliary system	To employ chatbot to deepen college-level English culture instruction	N/A	The intelligent English culture teaching auxiliary system is beneficial to improving students' autonomy and interaction of cultural learning
Moussalli and Cardoso (2020)	Echo	To use voice-controlled IPA Echo to promote independent study	11 L2 learners from various backgrounds	L2 students use a range of tactics to prevent communication breakdowns with Echo
Ruan et al. (2019)	BookBuddy	To introduce BookBuddy, a scalable foreign language teaching system that can generate interactive lessons for youngsters depending on their reading material	5 children	Children were very engaged throughout system interactions and preferred speaking English with our chatbot over a human companion
Rzepka et al. (2018)	Radiobot	To show a prototype of the Radiobot system that can construct a verbal conversation automatically and modify its content based on listener feedback.	38 participants	Radiobot can reduce user input load while being adequately interesting
Safitri et al. (2021)	ChatJours	To use ChatJours to facilitate French learning through Telegram	French learners at FFL learners at Universitas Pendidikan Indonesia	The chatbot may be used as a supplementary tool for learning French grammar
Santos et al. (2016)	TORMES	To utilize AICARP to detect learners' emotional states and to react appropriately when L2 learners conduct an oral test.	N/A	TORMES can detect the learner's physiological condition and redirect negative affective states, such as anxiety, into good affective states, such as relaxation
Shi et al. (2020	Empathy chatbot	To propose an empathy-generating L2 learning chatbot based on transfer learning	N/A	Construct and deploy an English Learning chatbot based on the principle of explainable artificial intelligence using ontology graph (OG) and transfer learning
Smakman et al. (2022)	The SAMBuddy Storytelling Cuddle	To determine whether humor, intonation, and gender influence children's trust in social robots	115 children	The results show that robots might be able to build trusting relationships with children and help them feel less stressed
Taguchi et al. (2017)	Scenario-based interactive practice	To investigate the use and efficacy of scenario-based interactive practice in Chinese language acquisition	35 Chinese learners from the USA	The system met the emotional and behavioral needs of each learner to make learning fun
Thompson and McGill (2017)	Jean	To use an e-learning system that detects and reacts to the learner's emotional states	40 adult students	Jean enables the improvement of current e-learning systems for the purpose of providing emotional support
Tran et al. (2020)	PLATICA	To offer immersive opportunities for language practice	N/A	PLATICA can deliver real-time grammatical criticism to help users improve their English proficiency
Troussas et al. (2017)	A chatterbot derived from Alice with an enriched library	To assist children in acquiring English vocabulary	N/A	The chatterbot improves educational outcomes such as motivation and cognitive abilities
Wahyuni (2022)	An EFL chatbot	To cope with the loss of L2 interactions after class,	N/A	Upscaling learners' self-confidence by incorporating a chatbot into the classroom as an alternate method to deal with learners' lack of interactions using English
Wilcock and Yamamoto (2015)	CALL, COGINFOCOM NLP Wikitalk	To employ CogInfoCom channels to increase the learner's interest and engagement.	N/A	The robot's nonverbal behavior increases the learner's involvement in the discourse, and the learner's degree of interest may be assessed by monitoring the learner's nonverbal signals
Wu et al. (2022)	An innovative affective mobile language tutoring system (AMLTS)	To use AMLTS in an effective manner	63 Taiwanese students learn Japanese	AMLTS emotional interaction design significantly promotes learner engagement and performance
Xie et al. (2019)	Open-domain dialog systems	To employ a multi-turn chatbot with the goal of learning and creating emotional reactions that humans are now capable of	4 participants	The chatbot monitors the discussion environment and provides replies that are more emotionally suitable while doing similarly well on grammar
Xie et al. (2021)	Virtual reality (VR) technology	To mix immersion-based English instruction with VR technology in order to increase learning results	106 Chinese Students	The findings indicated that students enjoyed exploring English culture and appreciating its history and customs
Zhang (2021)	CNN interactive agent	The research offers a convolutional neural network-based method for recognizing how learners' attention, memory, reasoning	N/A	This research facilitates the realization of harmonious emotional contact in the innovative learning environment in real-time and without disrupting the learners' usual online learning

Based on the data analysis, the following research questions can be addressed:

RQ 1 - Effects of cultural dimensions of conversational chatbots on L2 learners’ learning outcomes.

Cultural dimensions in learning increase students deeper understanding of their own culture as well as other. It develops students' awareness of the social norms, values, and behaviors that are associated with each culture. In addition, the student will be able to recognize and explain cultural differences with sensitivity and self-assurance (Kramsch, 2013).It is a learned system of common knowledge, attitudes, actions, beliefs, and customs among a group of individuals from many nations or cultures (Kramsch, 2013). Culture is responsible for the formation of beliefs, the transmission of ideas, and the dissemination of social values and practices. Language, an essential component of culture, is the medium through which all of these qualities are transmitted (Kramsch, 2013). There are two elements found which are associated with cultural dimensions in the conversation design which affected L2 learners’ learning outcomes: cultural awareness and heterogenous culture.

3.1. Cultural awareness

Cultural awareness is sensitivity to the similarities and contrasts that exist between two distinct cultures, as well as the application of this sensitivity in efficient communication with members of another cultural group (Luo and Cheng, 2020). Several articles have reported that L2 chatbots embedded with cultural components were an effective tool to assist the learning process as shown in Table 3. Johnson (2019) introduced a virtual cultural awareness trainer, Enskill, a cloud-based interactive platform for L2 learners to practice their speaking skills of a second language. Enskill provides learners with an immersive learning experience which also increases learners' cultural awareness. The author claims that learners' L2 competence increases through role-plays with Enskill, and Enskill is a great language learning tool that helps learners understand English phrases in real-life situations. Lee et al. (2012) developed a chatbot which can retrieve language learning data from the internet and generate L2 feedback to assist L2 learners in comprehending the content. The findings show that the interactions between the chatbot and L2 learners were enjoyable and fruitful, and learners’ cognitive abilities also increased. Luo and Cheng (2020) examined the use of an L2 chatbot in foreign language learning where the authors found that individual L2 learners with various cultural heritages, thinking patterns and growth backgrounds respond differently to the same linguistic form. Mageira et al. (2022) presented a report on field experiments with an L2 chatbot, AsasaraBot, and their findings showed that the majority of the participants has a positive attitude towards AsasaraBot.

Table 3.

Cross-cultural influence in L2 chatbots

Approaches	Authors and Year	Name of the Chatbot	Cross-Cultural Influence
Cultural Awareness	Kim et al. (2022)	Ellie	Ellie's utterances are also culture-coded (as someone who grew up in San Francisco, California, in the United States)
	Mageira et al. (2022)	AsasaraBot is introduced to the students as the Goddess of Snakes	cultural content learning
	Wahyuni (2022)	An EFL chatbot	The culture embedded in the target language
	Xie et al. (2021)	VR technology	English culture
	Freidkin and Ksenofontova (2021)	Telegram-bot “Pardes”	Israel culture - Geographical position of Israel.”
	Safitri et al. (2021)	ChatJours	The design includes the culture of a certain society
	Luo and Cheng (2020)	A collaborative multimedia teaching ​model	Dominance driven by European and American cultures; need to incorporate mother tongue
	Meng-yue et al. (2020)	Intelligent English Culture Teaching Auxiliary System	English culture
	Moussalli and Cardoso (2020)	Echo	Culture
	Johnson (2019)	Enskill English's chatbot	Enskill, immersive training systems, is embedded with and cultural features
	Taguchi et al. (2017)	Scenario-based interactive practice	Embedding language within culturally-rich interactions
	Troussas et al. (2017)	A chatterbot derived from Alice with an enriched library	Engage learners in cultural learning
	Wilcock and Yamamoto (2015)	CALL, COGINFOCOM NLP Wikitalk	The cultural difference needs to be considered
	Lorenzo et al. (2013)	MMOL	Spanish culture and history
	Lee et al. (2012)	Mero and Engkey	Providing cultural and contextual information
	Mazur et al. (2012a)	Artificial language tutors	Use “culture” and “sightseeing for cross-cultural communication
	Mazur et al. (2012b)	A pedagogical conversational agent	Medium for cross-cultural communication
Heterogenous Culture	Goga et al. (2021)	Rose	Cultural difference support
	Chen et al. (2020)	Anthropomorphized chatbots for migrants	Reflect on German culture

3.2. Heterogenous culture

Heterogenous culture refers to disparities in cultural identity based on factors such as class, ethnicity, language, customs, religion, and a feeling of location, among many others Goga et al. (2021). Two studies have pointed out that L2 chatbot increased learners' cultural awareness and engagement in learning a new language while in a foreign country rather than their own. Chen et al. (2020) found that when migrants from various cultural backgrounds entered Europe and formed a heterogeneous group, the heterogeneous nature of L2 migrants requires a chatbot designer to achieve a comprehensive understanding of migrants' needs where different voices can be heard. Thus, the authors believe that this type of L2 chatbot design serves as a social integration tool. The L2 chatbot can help migrants adjust to a new culture, social science and classic anthropology of the host country. Meanwhile, Goga et al. (2021) developed a chatbot, Rosa, to decrease L2 learners' dropout rate while studying abroad. The authors argued that it is important that an L2 chatbot is able to improve learners' second language knowledge considering L2 learners’ socio-economic status and support cultural differences.

Nineteen studies reported that L2 chatbots embedded with cultural components have shown to be an effective tool to assist the learning process. From Johnson's (2019) Enskill chatbot to Mageira et al.’s (2022) AsasaraBot, all of the L2 chatbots offered insights into the contribution to the cross-cultural communication in the target language. Language learners found that the experience with L2 chatbot fruitful and enjoyable. Some of the L2 chatbots integrated with cross-cultural elements even take into considerations of learners' socio-economic status and support cultural differences. There are two sub-dimensions included cross-cultural dimensions: cultural awareness and heterogenous cultures.

However, it is found that none of the nineteen studies have considered learners' own cultural backgrounds, which play a role in the formation of the learner's identity and cognitive characteristics with which an L2 learner adopts. Benattabou (2020) states that language learning is a manner of expressing one's culture and identity. An approach to L2 learning includes intercultural language learning which entails developing with L2 learners an awareness of their own culture, and identification of others. As to cognitive characteristics, Mehrotra and Yilmaz (2015) argue that the cognitive characteristics coexist in language proficiency development: one is an analytical procedural system, rule-based; the other is a declarative system, example-based. Adult learners are more characterized on the former, whereas children are defined by the latter. Due to various cognitive characteristics, young language learners have an advantage under natural acquisition situations, since their capacity to learn is instinctively stronger than their adult counterparts. These cognitive characteristics impact how a learner consumes information, perceives and organizes it in their brains.

Thus, it can be observed that cultural dimensions such as cultural awareness and heterogenous culture of conversational chatbots can have a positive impact on L2 learners' learning outcomes. The findings of L2 chatbots embedded with cultural awareness showed that the interactions between the chatbot and L2 learners were enjoyable and fruitful, and learners' cognitive abilities also increased. The findings also showed that L2 chatbots can be integrated with heterogenous culture to improve learners' second language learning while considering L2 learners’ socio-economic status and support their cultural differences.

RQ2 - Influence of empathetic dimensions of conversational chatbots on L2 learners’ learning and retention.

The display of empathy is a significant aspect of the attitudes of communicators, and empathy plays an essential function in intercultural communication (Jiang and Wang, 2018; Zhai and Wibowo, 2022). By reducing tension and stress, humor fosters language learning and makes the learning process efficient and enjoyable (Azizinezhad and Hashemi, 2011). In learning, empathy is the degree to which an instructor makes an effort to understand students ‘personal and social circumstances, to care and concern students’ positive and negative emotions, and to respond compassionately without losing focus on students' academic growth (Jiang and Wang, 2018). There are two factors associated with empathetic dimensions in this study to influence L2 learning and retention: emotional support and affective strategies.

3.3. Emotional Support

Emotional support is the demonstration of caring and sympathy for another individual. Communication may be verbal or nonverbal (Ismail, 2020). Four studies have examined the emotional support of L2 chatbots on language learners' emotional states as well as their learning curves as shown in Table 4. Ismail (2020) introduced an L2 chatbot to detect and provide appropriate responses to reduce L2 learners' foreign language anxiety, diminish their negative feeling and eventually improve learning outcomes. The author believes that it is a viable strategy to use of this sensor-free anxiety chatbot to alleviate anxiety. Similarly, Lin et al. (2015) developed an affective tutoring system to identify L2 learners' emotional states and provide appropriate emotional support during learners' learning process. The finding showed that affective factors such as anxiety and mood swings have a direct effect on the learning of a second language. Mazur et al. (2012b) used the emotional classification developed by Nakamura (1993) to identify L2 learners' emotional states, and the L2 chatbot is designed to foster learners’ interaction with empathy. The authors underline the significance of emotional identification of a chatbot by offering an appropriate response to improve learning.

Table 4.

Empathetic dimensions embedded in L2 chatbots.

Approaches	Authors and Year	Name of the Chatbot	Empathy Dimensions
Affective Approaches	Wu et al. (2022)	An innovative affective mobile language tutoring system (AMLTS)	The stress relief module The emotion recognition module
	Ayedoun et al. (2020)	An embodied conversational agent, Peter	Backchannels emotion-sensitive support
	Xie et al. (2019)	Open-domain dialog systems	The emotional dialogue model of Asghar The Emotional Conversational Device
	Thompson and McGill (2017)	Jean	Affective stack model
	Taguchi et al. (2017)	Scenario-based interactive practice	Integration of affective components (such as motivation, drive, and fun)
	Santos et al. (2016)	TORMES	Ambient Intelligence Context-Aware Affective Recommendation System (AICARP)
	Ayedoun et al. (2015)	Multimodal response generator	Affective-cognitive context
	Litman and Forbes-Riley (2014)	ITSPOKE (Intelligent Tutoring SPOKEn dialog system)	ITSPOKE uses affect-adaptation
	Dai et al. (2014)	Multi Intelligent Agents	A neural mechanism model for analyzing the learners' emotional characteristics
	Hong et al. (2014)	The courseware of animated pedagogical agent	Social interaction influences affective factors
	Lee et al. (2012)	Mero and Engkey	Affective factors are embedded
	Mazur et al. (2012b)	A pedagogical conversational agent	Emotions and words with certain emotional load
Emotional Support	Shi et al. (2020)	Empathy chatbot	Fine-tuned GPT-2 model
	Tran et al. (2020)	PLATICA	the EmpatheticDialogues dataset comprises 60,000 utterances
	Ayedoun et al. (2019)	Conversation agent, Jack	Affective backchannels
	Gordon et al. (2016)	The Tega robot platform	Affdex mobile SDK to detect learner's facial expression

3.4. Affective strategies

Affective methods are learning techniques focused with negative and positive emotion management (Wu et al., 2022). Thirteen studies have examined the affective strategies of L2 chatbots, whereby L2 chatbots help learners cope with language learning anxiety and foster engagement. Ayedoun et al. (2019) introduced a dialogue management model based on communication strategies and affective backchannels (DiMaCA), which are able to train the L2 chatbot's ability to carry out affective conversations with L2 learners and foster L2 learners' willingness to interact in English as a second language. The authors claimed that DiMaCA was able to alleviate expected anxiety and proved to be effective in fostering L2 learners' confidence. Gordon et al. (2016) presented an L2 chatbot that is able to affectively interact with children as a long-term companion. As the L2 chatbot is affectively sensing, its affective feedback significantly maximizes L2 learners' long-term learning gains. Wu et al. (2022) developed an AI chatbot system that can detect the emotional shifts of a student via text exchanges. This system assists instructors in monitoring and preventing students from developing psychological issues, such as foreign language learning anxiety and stress.

Due to the subjective nature of emotions, however, AI chatbots tend to be prejudiced and lack of cultural sensitivity (Purdy et al., 2019). Tran et al. (2020) propose an immersive method to engage L2 learners in informal English discussions with a chatbot, PLATICA, which is able to recognize learners’ emotions. Similarly, Shi et al. (2020) created an English-learning chatbot employing ontology graph (OG) and transfer learning. To mimic human communication, the authors included paired datasets into a deep learning model that incorporates the semantic expression in NLP and empathetic personality matching which was taken from XiaoIce. However, the project made little progress when OG was used to explain (output) phrases in natural English.

All seventeen studies embedded with empathetic functions reported that L2 learners' foreign language anxiety was reduced, their negative feeling was diminished and their learning outcomes was improved. Mazur et al. (2012a) used the emotional classification developed by Nakamura (1993) to identify L2 learners' emotional states, and the L2 chatbot was designed to foster learners' interaction with humor. The authors underline the significance of emotional identification of the chatbot by offering an appropriate response to improve learners' willingness to interact and learn. Yet, out of seventeen studies, only one article (Zhang, 2021) presented an empathetic detecting algorithm. In Zhang's study, the author proposed a convolutional neural network, which comprises of three convolutional layers, three pooling layers, and one connected layer to detect L2 learners' emotions and generate appropriate responses to enhance learning. The author claims that the key to creating harmonious emotional connection in an intelligent learning environment is the identification of learner emotions, and only by identifying learner emotions can L2 chatbots be altered to accomplish self-adaptation and personalized learning.

These approaches, however, are limited in scope to address emotion recognition in L2 acquisition. Although AI chatbot can evoke an empathetic response, no study was able to make definitive conclusion on the empathy in a conversation chatbot. This qualitative synthesis of empathy should be interpreted with caution due to the inconsistent use of outcome ratings across the analyzed studies. For example, two studies (Ayedoun et al., 2019; Lin et al., 2015) reported that L2 chatbots can detect emotional fluctuations, or mood swings experienced by learners in text conversations. These systems help teachers monitor students’ psychological behaviors and take appropriate precautions, and yet Taguchi et al. (2017) reported that AI chatbots tend to be prejudiced and unable to understand cultural nuances. Even though, Wu et al. (2022) developed a new system for text-based sentiment analysis based on a study conducted by Ptaszynski et al. (2010), and Wu et al. (2022) aimed to identify the emotions of L2 learners from texts containing emotional distress. These emotional systems can only handle first-level questions with predictable responses and with no personality traits. To solve this issue, Shi et al. (2020) developed an English learning chatbot using ontology graph to simulate human dialogue. It inherited the semantic expression in NLP and the empathy matching personality developed by XiaoIce. Unfortunately, this project has made little progress in explaining (outputting) sentences in natural language using OG. According to Gennaro et al. (2019), conversational chatbots embedded in empathetic responses cause palliative effects on these introverted participants. Wu et al. (2022) also found that the absence of non-verbal elements, such as paralinguistic elements, gestures, and facial expressions, puts an undue burden on written disclosure to attain more creative clarity.

It can be seen that emotional support and affective strategies of conversational chatbots can have a positive impact on L2 learners' learning outcomes and retention. The findings from this study showed that affective factors such as anxiety and mood swings have a direct effect on the learning of a second language and L2 chatbots embedded with emotional support fostered learners' interaction and improve learning outcomes. The L2 chatbots were able to detect and provide appropriate responses to reduce L2 learners’ foreign language anxiety, diminish their negative feeling and eventually improve learning outcomes and retention. Similarly, L2 chatbots embedded with affective strategies can detect the emotional shifts of a student via text exchanges. This system assists instructors in monitoring and preventing students from developing psychological issues, such as foreign language learning anxiety and stress.

RQ3 - Effects of humorous dimensions of conversational chatbots for improving L2 learners’ learning and retention.

Humor dimensions highlight the various ways in which individuals communicate with others and cope with daily stress through the use of humor. The inclusion of humor in the course material increased students' engagement with the material. Humorous components have a crucial function in aiding students' recollection, enabling discovery, leading them to other sources, supporting the sharing of information with others, promoting self-regulation, requiring cognitive effort, and facilitating learning (Farah et al., 2021). In this systematic review, two approaches are used in the studies of conversational chatbots to enhance learners’ learning outcomes: knowledge base and humorous response.

3.5. Knowledge Base

In general, a knowledge base is a centralized depository for information. Examples of knowledge bases include a public library, a database of relevant material on a certain topic (Liu et al., 2020). Two studies have integrated humor algorithms in L2 chatbots to improve language learning (see Table 5). Farah et al. (2021) integrated a humorous algorithm into an L2 chatbot which entails ambiguous grammatical constructs, misplaced modifiers, to harness humor for L2 learners during their study. The authors took reference to the ambiguity concept of humor, and they believed that anthropomorphic conceptions of characteristics such as friendliness and humor can improve L2 learners’ language proficiency. Liu et al. (2020) included humor into a built-in knowledge base in order to make the chatbot more user-friendly for L2 learners where the authors claimed that by extending the capabilities of the open-domain DeepQA agent, it could respond appropriately to inquiries from restricted domains. The results show that the chatbot algorithm is effective in information retrieval of second language learning.

Table 5.

Humorous approaches embedded in L2 chatbots.

Approaches	Authors and Year	Name of the Chatbot	Humor Dimension
Humorous Response	Smakman et al. (2022)	The SAMBuddy Storytelling Cuddle	Humor variant dialogue
	Al-Kaisi et al. (2021)	Alice	Embedded with a sense of Russian humor
	Gayathri (2021)	Automatic Mega-agent Managed User Guide (AMMU)	Emotion detection: Jokes
	Gonulal (2021)	Intelligent personal assistant (IPA): Google Assistant	humorous interactions with IPAs
	Lubold et al. (2021)	Social robot Quinn	Humorous rapport
	Dokukina and Gumanova (2020)	Vasya	The style in which Vasya teaches grammar is really approachable, and she has a fantastic sense of humor as well.
	Rzepka et al. (2018)	Radiobot	Add humor generation
	Ahmadi et al. (2017)	An anthropomorphic pedagogical agent	Funny virtual character
	Danilava et al. (2013)	An Artificial Conversational Companion	Need to make the system humorous
	Liang (2012)	Virtual world ​Second Life	humorous communication
Knowledge Base	Farah et al. (2021)	An ambiguous chatbot	Make use of the humor that may be derived from ambiguous grammatical constructions in order to bolster anthropomorphic impressions of characteristics.
	Liu et al. (2020)	chatbot for the field of educational technology (CBET)	Add humor corpus into the knowledge base

3.6. Humorous Response

A computational sense of humor would entail not just the capacity to recognize when others are making jokes but also the ability to respond (Gayathri, 2021). Ten studies laid emphasis on humorous responses of L2 chatbots to enhance language learning. Dokukina and Gumanova (2020) introduced a chatbot named Vasya. Vasya is humorous and able to explain grammar in a way that is straightforward and easy to understand. The authors suggested that Vasya is very helpful for L2 learners at a beginner's level, however it gets too complicated at an advanced level, which needs a human instructor to step in. Gayathri (2021) developed an L2 chatbot named English master, Ammu, using Google Dialogflow. Ammu can identify L2 learners' emotional states, modify its responses and integrate some forms of humor to enhance L2 learning. The author claimed that Ammu can establish a healthy relationship with the L2 learners. Similarly, Gonulal (2021) examined intelligent personal assistants (IPAs)’ sense of humor, in which fifteen humor-inducing utterances were embedded to facilitate L2 learners' speaking and listening skills. The results showed that L2 learners highly prefer intelligent personal assistants which are humorous and helpful in improving learners' overall language skills.

Twelve studies find that AI chatbots with a good sense of humor are becoming the preferred learning choice for L2 learners to improve their learning outcomes (Lubold et al., 2021; Rzepka et al., 2018). Mazur et al. (2012b) argued that the originality of this type of chatbot is able to perform various functions with different modules. Humor modelling, for example, is designed to generate simple verbal jokes to improve learners’ attitudes toward the system and motivate them. However, the authors did not elaborate on how to perform humor dimensions and reduce vocabulary ambiguity. Dokukina and Gumanova (2020) developed a chatbot named Vasya where they believe that it has a good sense of humor. Vasya can explain grammar with complex and sloped approaches. However, it is only suitable for the beginner level L2 learners, and at advanced levels, the user interface is more complex for learners to handle. Therefore, a human teacher needs to step in and support learners when advanced levels are operated.

So far, it has been found that no study on L2 conversational chatbots entailing how to generate culturally related linguistic humor to enhance the learner's attitude towards the system, motivates and improves the learner's performance. In addition, no research has been done to characterize the nature of computer interface humor mechanisms by first reflecting the cultural background of the target language and then providing contextual recognition strategies that guide conversations as needed. It is important to embed culturally related humor as studies on class-room language teaching suggest that implementing culturally related humor in learning can help increase learners' emotional intelligence (Chabeli, 2008). For example, Slaski and Cartwright (2002) discovered that learners, who achieved higher scores in the emotional intelligence scale, experienced less subjective stress, maintained better health and well-being, and demonstrated better outcomes. Furthermore, when culturally related humor is applied, individuals with high traits of emotional intelligence display sound stress management skills and an ability to appraise (Slaski and Cartwright, 2002).

It can be seen that humorous dimensions derived from knowledge base and humorous response of conversational chatbots have a positive effect in improving L2 learners' learning and retention. This is supported by a few studies (Azizinezhad and Hashemi, 2011; Renninger and Hidi, 2015) which suggested several benefits for students when cross-culture humor is applied in learning. Learners experience less anxiety towards the studied subjects, and more self-motivation is observed when humor is integrated into learning (Renninger and Hidi, 2015). Anderson (2011) finds that cross-culture humorous examples in the study enhance the effectiveness of students’ learning as learners are exposed to a gratifying environment that lengths their attention span and provides a different perspective that enhances retention of the content. This enhancement in learning promotes a positive and creative learning environment that engages learners in the learning process (Anderson, 2011).

It is believed that implementation of culturally related humor dimensions in L2 conversational chatbots develops the total power of the brain. McGhee (1983) claims that the perception of humor engages both hemispheres in the brain at the same time, the left hemisphere is the “logical brain”, which involves in language competence and logical analysis, and detailed joke recognition, and the right hemisphere is the “creative brain” which involved in imagining and comprehending humor. Learners engaged in convergent thinking tasks have displayed this proclivity of activating both hemispheres of the brain over a creation task, and learners involved in cross-cultural humor activities are measured similar brain activity to those taking part in convergent thinking (McGhee, 1983). Thus, the involvement of humor can be a cognitive process that engages both hemispheres of the brain, and when humor modelling is embedded in conversational L2 chatbots, it enhances learners’ outcomes of cognitive, emotional, academic and psychological competence.

When the culturally related humorous function is enabled in conversational chatbots, it will benefit learners’ learning outcomes. Chabeli (2008) discovers that by embedding culturally related humor in class, learners tend to encounter cognitive disequilibrium by adopting unusual ways of examining information of cross-culture humor from some antagonist viewpoints. These request learners to develop a new strategy or modify its existing schema. Under a particular circumstance, this type of disequilibrium creates a kind of constructive bewilderment that can benefit learning. Cognitive disequilibrium is the force of strains people experience when exposed to contrasting perspectives that they cannot solve (Chabeli, 2008).

In its essence, culture comprises of an acquired ability that fuses cognitive or affective aspects of communication (Wang et al., 2003). Feshbach (2009) argue the importance for novice teachers to explore students' cultural backgrounds via cognitive or affective empathetic communication. Teachers' empathetic support does not only develop a type of emotional rapport with students by leveraging sustaining cultural practices but also contribute to the dynamics of the learning environment and students’ academic success while building connection across disparate social and cultural contexts (Bozkurt and Ozden, 2010) (See Figure 3).

Figure 3.

Contributing components of humor and empathy in learning.

This paper considers both theoretical and practical implications. For theoretical implications, this study may serve as a theoretical basis for learning new languages and other educational subjects. This study has offered a comprehensive understanding of the challenges and benefits posed by AI chatbots in L2 acquisition. This evaluation offered a chance to conduct more research that would contribute to the enhancement of the overall learning experience for students.

Practically, this study illuminates the practical constraints of adopting a conversation system in L2 acquisition. Before building a chatbot to maximize the results for learners, it is essential to take into consideration of learners' cultural background, empathetic support, humorous responses, and skill-based cognitive learning methodologies. Researchers may use the study's results to develop successful tactics for chatbots. The findings are also valuable for language teachers in better identifying cross-cultural circumstances surrounding the usage of new technologies in order to give appropriate means of increasing the learning experiences and outcomes of their students.

It can be seen that there have been a wide variety of studies on the pedagogical use of chatbots for L2 acquisition. To further understand the challenges on the effective use of chatbots for L2 acquisition, future research studies should boost learners’ learning experiences from four macro skills such as speaking, listening, reading and writing. The scope of this research is constrained to an examination of chatbots in L2 acquisition. Future research should focus on age, gender, and ethnography will aid in the development of effective learning strategies.

4. Conclusion

This paper investigates the current status of conversational L2 chatbots embedded with culture, humorous, and empathetic dimensions. Despite increasing attention to neural conversation models, there is a lack of research providing a context-aware empathetic and humorous dimensions when learners show a lack of interests and encounter emotional discomfort during learning, and there is a lack of research that considers the cultural background of L2 learners to facilitate learner language learning.

Various approaches have attempted to develop empathetic chatbots to elicit appropriate emotional responses to the emotional state of L2 learners (Fryer et al., 2020; Santos et al., 2020). However, most approaches are limited to emotion recognition, only a few can evoke culturally-related empathetic responses, and the answers are predictable as they can only handle first-level questions, not the complicated ones.

Existing humor dimensions have been designed to generate simple linguistic jokes to improve learners' attitudes towards the system and increase learners' motivation. However, most studies did not elaborate on how to execute the humor dimensions and reduce the lexical ambiguity. Moreover, the existing conversational chatbots embedded with humorous modelling are only suitable for L2 learners at a beginner level. At the advanced level, its user interface becomes rather complicated to execute for the learners. Thus, further study is needed to test the efficacy of computational empathetic responses and humorous interactions in chatbots to detect how learners’ emotion evolvement and humorous interactions lead to better outcomes.

4.1. Limitations

One of the study's limitations is that the information and analysis of existing studies are confined to solely L2 acquisition. Furthermore, the research is constrained to reviewing literature from 2012 to 2022. Future research should evaluate the utilization of alternative technologies to enhance other sectors, such as the medical and hotel industries.

There is also limited literature considering learners' cultural backgrounds when offering individual feedback. The importance of diversity in feedback cultures is generally overlooked. Feedback contexts and cultures are intertwined, research on student feedback cultures is broadly driven by cultural variation, and it depicts students' struggles with learning, critical thinking, and academic writing. Thirdly, studies have found that learners experience frustration and anxiety during learning (Ayedoun et al., 2019; Divekar et al., 2021), yet there is no empathetic solution given to alleviate those concerns. In general, the above studies have shown that chatbots have a significant positive impact on learners’ speaking training.

4.2. Recommendations

This study has examined the use of chatbots for L2 acquisition. The following recommendations are provided for improving learners’ acceptance and experience in L2 chatbots.

Firstly, to advance the application of L2 chatbot on empathetic modelling, it is recommended that chatbot developers (a) integrate cross-cultural empathetic responses in conversational L2 chatbots, and (b) identify how learners perceive and react to the learning content. These chatbots embedded with cross-cultural empathetic modelling can intervene to further mitigate learners’ L2 anxiety and stimulate an individual learning momentum and outcome as shown in Figure 3. This is supported by Zhai et al. (2022) who claim that learners are more likely to adopt a specific chatbot which can provide proper interactions, and understand their thoughts and feelings.

Secondly, it is recommended that chatbot developers focus on identifying the effects of cross-culture humor on learners' language proficiency. When exploring learners' perception of cross-cultural humor in class, Thomas et al. (2015) found that learners' performance were improved, and positive outcomes were reported when humorous methods, entailing effective learning, was used in the study. The authors confirmed that one of the prominent effects of introducing cross-cultural humor into the classroom has strengthened learning and information retrieval skills. The authors argued that a learner's cognitive and affective states are indispensable when culture-related humor is employed appropriately during learning. Meanwhile, Safitri et al. (2021) found that learners are inclined to increase self-motivation and be actively involved in their learning process when they are subjected to humorous learning content.

Thirdly, chatbot developers should eliminate any potential for bias based on gender throughout the design process. In addition, it is important that these developers exercise care while training the bot to ensure it exhibits the desired behavior. There is a possibility that the chatbot may exhibit racist or sexist behavior or use inappropriate language if it has not been properly programmed. According to Jeon (2021), biases ultimately lead to decreased trustworthiness in chatbots adoption intention.

Finally, chatbot developers should acknowledge the mandatary importance to privacy when designing and developing an L2 learning chatbot as personal information is normally collected through a chatbot. Developers must be aware of the basic human rights to privacy and freedom, especially with regard to the protection of personal data from unethical behavior. Unregulated data storage, authentication performed without the user's consent, and the authorization and encryption of personal data performed without the user's permission will result in the user's mistrust of the chatbot.

Declarations

Author contribution statement

Chunpeng Zhai; Santoso Wibowo: Analyzed and interpreted the data; Wrote the paper.

Funding statement

This project has received an internal grant from Research Training Program (RTP) Living Stipend Scholarship (980025753) of Central Queensland University, Australia.

Data availability statement

Data included in article/supp. material/referenced in article.

Declaration of interest's statement

The authors declare the following conflict of interests: The author whose name is listed immediately below certify that he as NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in this manuscript.

Additional information

No additional information is available for this paper.