Conversational AI: Dialogue Systems, Conversational Agents, and Chatbots by Michael McTear

This book has appeared in the series Synthesis Lectures on Human Language Technologies: monographs from 50 up to 150 pages about specific topics subjects computational linguistics. The intended audience of are researchers and graduate students NLP, AI, related fields. I define myself as a linguist; my review is perspective “random” linguistics researcher wanting learn more this topic or looking for good guide teach course dialogue systems. found very easy read interesting therefore believe that McTear fully achieved his purpose write “a readable introduction various concepts, issues technologies Conversational AI.” He succeeds remarkably well staying right level technical details, never losing giving an overview, reader does not get lost numerous details algorithms. Additionally, people who experts could still be useful because its bibliography exceptionally complete: large number early works recent studies cited commented through whole book.The structured into six chapters. After introduction, there two chapters types systems: rule-based systems (Chapter 2) statistical 3). followed by chapter evaluation methods 4), after which neural end-to-end reviewed 5). ends with challenges future directions research AI 6). it was meaningful distinguish three systems, but modular It might, at first, seem strange system placed between reader, discussion comes around place book, helps better understand difference sequence In review, will discuss one same order they appear first chapter, explains clearly what cases can introduced perform tasks. sketches historical present context domain illustrates different existing many examples. introduces subject linguist, have admit would like seen linguistic description how human characterized.The second provides detailed complete overview work field shows evolved. particular, diagram explanation architecture were helpful such and, again, plenty examples illustrate make read. However, if shortened—it actually 180 instead usual Technologies—I should slightly fewer systems.From third smooth transition: Thanks comprehensible Chapter 2, framework adapted become system. Moreover, text reinforcement learning used management everything nicely illustrated clear examples.The fourth discusses evaluated training data collected. particularly comparison (e.g., using Amazon Mechanical Turk workers) automated metrics interesting. liked ethical stake when collecting amounts crowd-sourcing platforms. But, marginal comment put aside, complete, also concise descriptions all subcomponents evaluated.The fifth presents understanding type (be data-driven). explanations word embeddings recurrent networks rather successful: They mechanisms these architectures clear. Throughout especially advantages disadvantages explained. up-to-date bibliographic references impressive will, opinion, advanced readers well. true enumeration available corpora data.The last example: multi-modality, problem sparseness, handling discourse phenomena, involved AI. Although find interesting, their variety makes 6 eclectic cannot shake impression serves catch-all been addressed elsewhere book. think possible introduce discussions earlier For example, problems anaphora, presented maybe even discussed 4 (about evaluation). issues. most bots having female voices sexist (because bot assisting function) time speech generation 2; gender-specific biases result biased train (in addition, small ethics missed On hand, protection customer privacy issues: People provide personal some “talking speaker type” Alexa Google Home people’s homes may exposed sensitive information. other question whether always refer bot, letting them speak real human. throughout just paint vision development systems.In conclusion, McTear’s beginning research, examples, accessible reading non-experts (provided knowledge NLP). highly recommend search comprehensive topic.