Chatbots' Greetings to Human-Computer Communication

• Chatbots are defined as systems that seek to mimic conversation rather than understand it and freely engage across conversational topics.Source: §1 · p. 2
• More than 225,000 botmasters have built over 250,000 chatbots on Pandorabots, producing more than 3 billion interactions by February 2015.These figures indicate the community’s scale and proximity to real users.Source: §1 · p. 2
• The paper reviews chatbot development from pattern-matching templates and prepared small-talk stores to architectures incorporating learning processes.It also considers how chatbot concepts and tricks can contribute to the illusion of intelligence.Source: §1 · p. 2
• Scientific documentation is limited for many chatbots, making their underlying technologies difficult to uncover and increasing reliance on web-page references.Source: §1 · p. 2
• The paper covers chatbot history, platforms and enrichment methods, techniques supporting the illusion of intelligence, potential contributions to human-computer communication, and future challenges.Source: §1 · p. 2

2.1 Early days

• Eliza used regex-based pattern matching and rephrasing rules, including wildcard substitution and content-free fallback remarks, to simulate conversation.Its implementation illustrates how predefined rules transformed user input into apparently relevant responses.Source: §2.1 · p. 3
• Eliza’s role-playing produced the “Eliza effect,” as users believed they were conversing with a human despite the system’s limited understanding.This impression exceeded its designers’ expectations and demonstrated the persuasive power of simulated understanding.Source: §2.1 · pp. 3–4
• Parry extended Eliza-like architecture with conversation knowledge and a state of mind, making outputs depend on beliefs, desires, and intentions; psychiatrists guessed its transcripts no better than chance.Its performance suggested that emotional behavior could be easier to imitate than intellectual capability, while its limited knowledge prevented modeling a real person.Source: §2.1 · pp. 4–5

2.2 The chatbots’ competitions

• Turing-inspired competitions, especially the Loebner Prize, institutionalized chatbot evaluation, but their reliance on deception generated controversy about whether conversational success demonstrates intelligence.Eugene Goostman convinced 33% of human judges in a 2014 U.K. test, while no Loebner contestant had won the prize outright since 1991.Source: §2.2 · pp. 5, 7
• Despite criticism of the tests, chatbot competitions contributed strongly to advances in the field and helped make competing systems among the most popular chatbots.The Loebner Prize also produced influential systems such as PC-Therapist, Jabberwacky, and A.l.i.c.e.Source: §2.2 · p. 7, §2.3 · pp. 7–8

2.3 Other distinguished Chatbots

• Jabberwacky and Cleverbot emphasized learning from users’ conversations, whereas A.l.i.c.e. used pattern matching to sustain general conversation and supported a large community through Pandorabots.Jabberwacky’s publicly described learning mechanism remained unavailable, while A.l.i.c.e. was designed to avoid detection without adhering to a specific role or topic.Source: §2.3 · pp. 7–8

2.4 The chatbot next door

• Chatbots moved beyond competitions into e-commerce, education, medicine, and multilingual platforms, while modern systems added faces, speech, bodies, and emotional expression.Elbot’s use on IKEA’s sites exemplifies this transition from contest systems to real-life services.Source: §2.4 · pp. 8–9

3 Building chatbots

• Chatbot platforms combine scripting languages with engines that map user utterances to appropriate answers.Source: §3 · p. 9

3.1 Scripting languages/platforms

• AIML encodes knowledge as input-matching rules and output templates, supported by freely available interpreters and A.l.i.c.e.’s files.Source: §3.1 · pp. 9–10
• AIML 2.0 is designed for easy modification and deployment, allowing non-experts to create and extend bots through Pandorabots.Source: §3.1 · p. 10
• ChatScript provides an open-source scripting engine with an ontology and is presented as addressing maintainability problems attributed to AIML.Source: §3.1 · p. 10

3.2 Building chatbots by chatting

• Chatbots can learn by storing previously unseen interactions, reusing them with later users, and treating acquired replies as future responses.Source: §3.2 · pp. 10–11
• The learning mechanisms of Jabberwacky and Cleverbot are not publicly available, and unsupervised learning may produce unexpected or undesirable results.Source: §3.2 · p. 11

4 Towards the illusion of intelligence and/or the art of scripting

• Botmasters script responses and conversational cues to maintain an illusion of understanding when universal plausible answering is impossible.Source: §4 · p. 11

4.1 Giving the bot a personality

• A simple assigned personality can justify contradictions, ignorance, topic shifts, or incongruous answers by making them fit the chatbot’s character.Source: §4.1 · p. 12

4.2 Directing a conversation

• Directing conversation helps avoid unexpected interactions, using questions or predefined topics to keep users engaged and the exchange manageable.Source: §4.2 · pp. 12–13

4.3 Paying attention to small talk

• Small talk establishes rapport, avoids silence, and supports trust, cooperation, and human-like interaction through neutral or canned exchanges.Source: §4.3 · pp. 13–14

4.4 Failing like a human

• Simulated delays and errors challenge expectations that computers should be fast and accurate, making interaction more like ordinary human communication.Source: §4.4 · p. 14

5.1 Some works that merge both communities

• Researchers combine chatbot resources with conversational agents, including AIML-based domain knowledge, corpus-trained systems, and personality-based recovery strategies.Source: §5.1 · pp. 15–16

5.2 Main challenges

• Chatbot logs provide large collections of real requests and replies that could support human-computer communication research.Source: §5.2 · pp. 16–17
• 72% of Edgar’s out-of-domain requests were answered after using the Subtle corpus, and about 65% of those answers were considered appropriate.Source: §5.2 · p. 17
• Open challenges include filtering unwanted answers, detecting paraphrases and semantic relations, selecting suitable responses, and ensuring contextual sense.Source: §5.2 · p. 17
• The paper identifies substantial room for improvement because chatbot-community corpora remain insufficiently explored.Source: §5.2 · p. 17

6 Conclusions and Future Challenges

• The paper’s central conclusion is that chatbot tools, corpora, and know-how should not be neglected by human-computer communication researchers.Source: §6 · p. 18
• Personality, conversation direction, small talk, and human-like failure are presented as features that create the illusion of intelligence.Source: §6 · p. 18
• The chatbot community’s main richness is its immense collection of interactions, most of which represent real human requests.Source: §6 · p. 18
• A major future challenge is automatically integrating these resources while avoiding contradictory answers and choosing responses appropriate to a chatbot’s character.Source: §6 · p. 18