We investigate the task of building open domain, conversational dialogue systems based on large dialogue corpora using generative models. Generative models produce system responses that are autonomously generated word-by-word, opening up the possibility for realistic, flexible interactions. In support of this goal, we extend the recently proposed hierarchical recurrent encoder-decoder neural network to the dialogue domain, and demonstrate that this model is competitive with state-of-the-art neural language models and backoff n-gram models. We investigate the limitations of this and similar approaches, and show how its performance can be improved by bootstrapping the learning from a larger question-answer pair corpus and from pretrained word embeddings.

/pdf/building-end-to-end-dialogue-systems-using-generative-2hd29gvys6.pdf

Building end-to-end dialogue systems using generative hierarchical neural network models

We investigate the task of building open domain, conversational dialogue systems based on large dialogue corpora using generative models. Generative models produce system responses that are autonomously generated word-by-word, opening up the possibility for realistic, flexible interactions. In support of this goal, we extend the recently proposed hierarchical recurrent encoder-decoder neural network to the dialogue domain, and demonstrate that this model is competitive with state-of-the-art neural language models and back-off n-gram models. We investigate the limitations of this and similar approaches, and show how its performance can be improved by bootstrapping the learning from a larger question-answer pair corpus and from pretrained word embeddings.

Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models

This paper introduces the Ubuntu Dialogue Corpus, a dataset containing almost 1 million multi-turn dialogues, with a total of over 7 million utterances and 100 million words. This provides a unique resource for research into building dialogue managers based on neural language models that can make use of large amounts of unlabeled data. The dataset has both the multi-turn property of conversations in the Dialog State Tracking Challenge datasets, and the unstructured nature of interactions from microblog services such as Twitter. We also describe two neural learning architectures suitable for analyzing this dataset, and provide benchmark performance on the task of selecting the best next response.

/pdf/the-ubuntu-dialogue-corpus-a-large-dataset-for-research-in-139jgm1iui.pdf

The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems

Real conversations with artificial intelligence

Chatbots are computer programs that interact with users using natural lan- guages. This technology started in the 1960’s; the aim was to see if chatbot systems could fool users that they were real humans. However, chatbot sys- tems are not only built to mimic human conversation, and entertain users. In this paper, we investigate other applications where chatbots could be useful such as education, information retrival, business, and e-commerce. A range of chatbots with useful applications, including several based on the ALICE/AIML architecture, are presented in this paper.

Chatbots: are they really useful?

We present machine-learning techniques used to generate an Arabic chatbot, which accepts user input in Arabic and generates replies extracted from the Qur'an. A system to learn conversational patterns from a Corpus of transcribed conversation has been used to generate a range of chatbots speaking different languages including English, French and Afrikaans. We review aspects of the Arabic language, which pose problems for chatbot-learning, and we discuss the revised process to handle Arabic training text and input/output. In principle, the Qur'an provides guidance and answers to religious and other questions; so we used the Qur'an as a training corpus for our chatbot. As the Qur'an is not a transcription of a conversation, we adapted the learning process to cope with the structure of the Qur'an in terms of sooras and ayyas. The resulting system accepts user input in Arabic, and answers with appropriate ayyas from the Qur'an.

An Arabic chatbot giving answers from the Qur'an

A chatbot is a conversational agent that interacts with users through natural languages. In this paper, we describe a new way to access information using a chatbot. The FAQ in the School of Computing at the University of Leeds has been used to retrain the ALICE chatbot system, producing FAQchat. The results returned from FAQchat are similar to ones generated by search engines such as Google. For evaluation, a comparison was made between FAQchat and Google. The main objective is to demonstrate that FAQchat is a viable alternative to Google and it can be used as a tool to access FAQ databases.

FAQchat as in Information Retrieval system

The Quran is held by Muslims to be a single-authored text, the direct words of God (Allah), conveyed by the angel Gabriel to Mohammed 1355-1378 years ago, and later transcribed verbatim to be used as the sole authoritative source of knowledge, wisdom and law. The challenge for computer scientists is to represent this knowledge, wisdom and law in computer systems: to build intelligent systems which can answer any question with knowledge from the Quran, and can help society, both Muslim and non-Muslim, to understand and appreciate the Quran.

Understanding the Quran:a new grand challenge for computer science and artificial intelligence

CALL can be a route to learner autonomy, allowing students to use PC-based software to learn individually, without need of class or teacher. However, language is above all a medium for communication, implying a dialogue of two or more participants: although CALL can provide exercises and lessons in grammar, vocabulary, writing skills etc, it might seem that conversation practice still calls for a teacher, or at least fellow students. A chatbot is a program which can chat in natural language, on a topic built into the chatbot’s internal linguistic knowledge model. Many chatbots exist, with different “knowledge” programmed by the chatbot builder. A chatbot may appear to be a suitable partner for conversation practice; for example, the speak2me.net website includes Lucy, who “chats” in the language and style of a nice, polite British young lady. However, language teachers will know that conversation practice is normally on a specific topic, to learn topic-specific vocabulary and language. Lucy is nice to chat with initially, but doesn’t adapt to different topics or lessons. Furthermore, Lucy helps autonomous learners of English, but not other languages. We have developed algorithms for adapting or retraining a chatbot with a corpus, to chat in the language and topic of the training corpus. An attraction of the corpus-training approach is that in principle any corpus, in any language and on any topic, can be used; so we have gone on to test this principle, by using domain-specific corpora to train chatbots to chat on specific topics such as the Qu’ran , Computing Frequently Asked Questions, and non-English language corpora such as the Corpus of Spoken Afrikaans. Language learners and teachers have given evaluation feedback, indicating that these adaptive chatbots offer a useful autonomous alternative to traditional classroom-based conversation practice.

/pdf/fostering-language-learner-autonomy-through-adaptive-1rtd08nmnq.pdf

BA Abu Shawar

Papers

Chatbots: are they really useful?

An Arabic chatbot giving answers from the Qur'an

FAQchat as in Information Retrieval system

Understanding the Quran:a new grand challenge for computer science and artificial intelligence

Fostering Language Learner Autonomy Through Adaptive Conversation Tutors