Abstract
Chatbot technology is a subfield of Artificial Intelligence (AI) that deals with text-based or speech-based conversational agents. In general terms, a chatbot enables a user to have a conversational interaction with a computer. Chatbots have applications in several fields including trade, tourism, customer care, health services, education, et cetera. This chapter describes two chatbot systems that we are developing for learning Arabic as a foreign language. KalaamBot is a speech-based chatbot that converses with learners and teaches them the language in a conversational setting. KalimaBot is a text-based personal vocabulary assistant that enables students to search for the meaning of words, synonyms, antonyms, and word usage in context. This chapter provides extensive discussion of the several challenges second language researchers and chatbot practitioners encounter when designing chatbots for language learning. Then, it concludes with recommendations and future research.
TopChatbots’ Architectures And Techniques
Typically, the architecture of a chatbot involves three main components: Natural Language Understanding (NLU), Natural Language Generation (NLG), and a Dialogue Management Engine (DME). Some scholars consider the User Interface (UI) as a fourth component, while others do not. In speech-based architectures, two components are added: a text-to-speech engine (TTS) and a speech-to-text engine (STT) (Kulkarni et al., 2019; McTear, 2020; Freed, 2021). Each of these components represents an active research domain that complicates the design of chatbots.
NLU is a branch of AI and NLP that enables computers to understand text or speech. NLU does not only enable computers to understand the meanings but also discern the speaker's intents. In addition, such a system can extract “entities,” like things, people, or places. It involves two tasks: Intent Classification and Entity Recognition. Intent Classification helps the agent to understand the intent behind the user's utterance. In other words, the intent is what the end user intends a chatbot should do for them. The intent is mainly expressed through the verb in a sentence. Entity Recognition identifies the discrete pieces of information received from the user. Entities are typically expressed as nuns in a sentence. For example, consider the sentence: Book flight tickets for me. Book is an intent, while flight and ticket are entities. By identifying intents and entities, this component extracts the meaning from the user input using NLP and NLU.
Key Terms in this Chapter
Conversational AI: The application of machine learning to develop text-based and speech-based chatbots to handle conversations the way humans do.
Thematic Learning: A teaching method that focuses on a specific theme which forms the core of a learning/teaching unit.
Wav2vec 2.0: A transformer-based automatic speech recognition system with great abilities to learn from raw audio data. It is one of the state-of-the-art ASR engines.
Sequence-to-Sequence: A model that takes a sequence of elements (i.e., words, sentences, letters, sounds, time series, et cetera.) and outputs another sequence of elements.
Automatic Speech Recognition (ASR): The ability to take speech as input and produce a transcript as an output. It is also known as speech-to-text (STT).
Learning by Discovery: The students' ability to independently investigate problems and find answers themselves.
Transformers: Innovative neural architecture in natural language processing that tried to solve the problems of long sequences in sequence-to-sequence learning by using the attention mechanism.
Language Proficiency: The ability to use a language as appropriate and accepted by native speakers of that language. This ability can be demonstrated in real-world scenarios.
Reinforcement Learning: A branch of machine learning that trains agents (or bots) to choose the actions that maximize their rewards over time in a certain environment.