ROBERTWHETSEL.COM

Website Disclaimer and Intellectual Property Ownership Statement

Written by admin on January 3, 2010 — Leave a Comment

This website is owned and operated by Robert C. Whetsel (Owner). The Terms of use (‘Privacy | Copyright | Legal Notice | End User’s License Agreement’) governs your use of this website and it’s content. By using the website, you automatically accept the Terms associated with this site. The Owner may change the Terms at any time and such changes will take immediate effect. By your continued use of the website thereafter, you agree to be bound by such changes. You should visit the Terms of Use page from time to time to review the then-current terms.

This website contains information that may be privileged or confidential and is the property of Robert C. Whetsel. It is intended only for the use on this site and is defined in the Terms of use. You are not authorized to read, print, retain, copy, disseminate, distribute, or use any any part without permission.

Terms of use: Privacy | Copyright | Legal Notice | End User’s License Agreement

Posted in About, Legal

Testing Posts from robertwhetsel.com

Written by Robert Whetsel on January 4, 2010 — Leave a Comment

This is a Blog test; I write to my Blog and it automatically posts to my twitter, Linkedin and Facebook pages.

Posted in FAQ, Social Engeering

Improving Chatbot Accuracy -Draft

Written by Robert Whetsel on April 12, 2009 — Leave a Comment
This entry is part 13 of 13 in the series Chatbot Literature Review: The State of Chatbots

Researchers have employed a variety of methods to improve their chatbot’s accuracy. Most current chatbots use a dialog management module and knowledge base and rules, which they use with templates to match user input. Improving chatbot accuracy may be accomplished through expanding the chatbot knowledge base, improving upon the standard rule-based conversation method for chatbots, or using alternative methods to the standard chatbot rule-based conversation method.

Expanding a knowledge base from which a chatbot may learn can improve accuracy. A drawback to creation of chatbot knowledge bases is that it is usually accomplished through hand coding, or hard-coding, which is time and labor intensive (?). The time taken to program, test and refine a chatbot’s knowledge base can be extensive. Furthermore, hard-coding the rules into a knowledge base may be restricted by the style and ability of the programmer. However, if chatbots could be programmed to learn from an established knowledge base themselves, it would reduce the time needed to create large knowledge bases for the chatbot and potentially improve accuracy of responses given to the user.

One researcher has automated chatbot learning by using language modeling to train chatbots in three different languages (Abu Shawar & Atwell, 2005). The authors created a model of a language to build a corresponding chatbot, and through a java application they translated text-based corpora into AIML. They approached the problem using two different machine learning techniques: first word and the most significant word approaches. Armed with this solution, they retrained their Alicebot with transcripts from human dialog, they compared human and chatbot to human and human dialog. A frequency list was built (words listed most utterances), patterns (the number of times this word appears with another word) and templates were established, and patterns and templates were rearranged to build AIML files. The authors concluded that this machine learning approach based on transcripts of human dialog, enabled their chatbot to converse in different styles and languages.

Another researcher used autonomic knowledge base creation and machine learning approaches to increase the efficiency of building their chatbot knowledge database. They presented an approach for extracting a chatbot knowledge base using <thread-title, reply> pairs for a new knowledge base from a corpora (an online discussion forum) (Huang, Zhou, & Yang, 2007). They normalized and cleaned the corpora by removing irrelevant replies, used humans to rate the training set, and employed author identity to grade the quality of a reply. The researchers applied a machine learning approach to a specific domain of large repositories of archived threads and reply records from online discussion forums. Their results showed that they were able to extract 2,000 pairs in two minutes. Given the speed of their results, the authors considered their approach to constructing a knowledge base for a chatbot to be superior to the by-hand” approach.

Yet another researcher has also focused on automatic chatbot knowledge base creation (Wu, Wang, Weisheng, & Li, 2008). They used a classification model based on rough-set assembly theory to process incomplete data, as applied to a system model data analysis on a controlled set. This  data analysis was coupled to ensemble machine learning algorithms, and was based on related replies and IR-related replies. The authors also concluded that their approach of automatic acquisition of the knowledge for a chatbot was effective.

Others have tried to improve the current rule-based approach to learning for chatbots (Pilato, Vassallo, Augello, Vasile, & Gaglio, 2005). The researchers used latent semantic analysis (LSA) to attempt to predict replies in conversations with humans. They built their chatbots on Alicebot’s software, generating one generic and three domain specific chatbots. Their corpora was 850 documents from the Internet to create the knowledge base. They used LSA to generate algorithms that analyzed the corpora and placed it into a matrix which in turn constituted the knowledge base. Upon interaction with a user, the first interactions would be directed towards the general knowledge (generic) chatbot which had the index” for what the other domain-specific chatbots contained within their knowledge base. This generic chatbot would direct the user to each domain-specific chatbot as deemed appropriate (rules as programmed by the developers). The user then interacted with the domain-specific chatbot until its limit of knowledge was reached, and by default, the user  was then directed back to the generic chatbot. The authors concluded that the LSA allowed them to overcome restrictions of the traditional chatbot rule-based approach.

Recently, some chatbot researchers have deviated from using the standard rule-based conversation approach used by most standard chatbots. For example, one approach involves the automatic generation of a trivial dialog phrase database through the use of genetic algorithms (Montero & Araki, 2007). They evaluated the phrase database for correctness using N-gram analysis to generate statistical data, which was then analyzed to observe the system’s performance. The authors approach is different from traditional AIML rule-based natural language interfaces in that they used phrases and associations. They acknowledge that creating a hand-crafted knowledge base (a dialog corpus) is highly time-consuming and difficult. They generated pairs of phrases, applying fitness function of a genetic algorithm which indicated a well-generated phrase, then evaluated to see if it was acceptable phrase. They used a ratio of acceptable phrases divided by well-generated phrases.

Another deviation from the standard rule-based conversation approach is the use of “semantic-driven interaction in an intelligent system that uses natural language dialog to explain concepts” (Pirrone, Pilato, Rizzo, & Russo, 2007). By means of a 3-tier system, their study attempted to utilize Latent Semantic Analysis (LSA) to cluster documents in a vector space (creating relationships between different words within that vector) (Pirrone). The architecture contained business, presentation and data tiers that consisted of two main parts: an assessment component and supply component. The researchers used the CYC knowledge base in this application to integrate the ontology with other parts of their application. The authors believe that their ontology helped to create relationships (context) and the vectors as a self-organizing map. They considered their approach to  improve the efficiency of their chatbot.

One group has deviated from the standard rule-based approach by combining an expanded knowledge base with machine learning and a neural network (Sing, Fung, Wong, & Depickere, 2007). They crafted an artificial intelligent neural network identity (AINI) based on understanding and reasoning of natural language, rather than being based on pattern-matching like other chatbots. AINI is especially unique since it uses a top-down natural language query in a multi-layered natural language query engine. It uses an AIML engine, natural language understanding and reasoning, FAQ metadata and a spellchecker in a three-tier layered architecture with a mySQL back-end. AINI can parse input, which goes into a natural language reasoning module where sentences are separated into words. Keywords are then chosen and a document retrieval from the Web (Internet) is executed. Information is extracted from the Web, answers are ranked using advanced reasoning, and a requisite response is given. The authors concluded that their agent performed comparably to ELIZA and  ALICE using this new methodology. Given that their chatbot uses a different approach to the rule-based pattern matching used by ELIZA and ALICE, it is understandable that they would compare the performance of their chatbot to these common chatbots. However, the authors plan to improve their chatbot performance further by adding more machine learning algorithms and expanding the chatbot’s knowledge base. Their expectation is that their chatbot’s performance would surpass that of ELIZA and ALICE.

Posted in Research | Tagged , , , , , , , ,

Improving Human-Like Qualities of Chatbots -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 12 of 13 in the series Chatbot Literature Review: The State of Chatbots

Some researchers are designing chatbots with the intent of integrating more natural human-like interactions (De Angeli & Brahnam, 2008). Specifically, the developer may program certain responses that would sacrifice accuracy, but confer more human-like traits (Sing et al., 2006). For example, if a chatbot is presented with a math problem it could wait, as if to imitate thinking about the problem, or even give the wrong answer. This could be interpreted by the user as a human trait (failing or making a mistake), but in reality, it is a trick built into the programming. Researchers have attempted to add a sense of humor, through the use of self-contained jokes to their version of the chatbot ALICE (De Boni, Richardson, & Hurling, 2008). One could envision a scenario where a user would ask a question and the chatbot would reply: Jim, I’m not a machine, Im a doctor!”, I’m on my break, come back later.” or Are you sure you want me to check for this, I know its a waste of time”, which gives a jocular feel to the conversation.

Attempts to improve human-like characteristics include personality matching and relationship maintenance (De Boni et al., 2008; Abbatista, Degemmmis, Licchelli, Lops, Semeraro, & Zambetta, 2002). One researcher used shallow parsing and statistical natural language processing methods to change the style of interaction of their chatbot according to personality matching deboni08. For example, the chatbot could take on a submissive role when the user is portraying a dominate one. However, the authors found that users’ feelings were mixed about the chatbot using this style of interaction. Two different groups have developed chatbots that build upon previous interactions with the same user (De Boni et al., 2008; Abbatista et al., 2002). If a chatbot could incorporate the ability to recognize” a logged-in user and interact accordingly, the chatbot would be seen to be engaging in relationship maintenance (Abbatista et al., 2002). The chatbot would use a stored user profile and make recommendations in a proactive fashion (Abbatista et al., 2002). Both studies found that familiarity with a user did produce a more positive view of the chatbot’s interaction with that user.

Improving the chatbot’s ability to shift conversational topics is another approach to improving its human-like quality. Chat between humans is dynamic and a single utterance can become the focus of conversation (Montero – Enhancing computer). Another subsequent utterance (considered a catastrophe) may change the topic subject and the focus of the conversation would shift. One study attempted to model human chat by modeling relating utterances and changing topics more naturally (Montero & Araki, 2005). Based on ALICE, the chatbot database had categories to match patterns (user input and a template for chatbot reply). The authors programmed the chatbot to ask a question if it did not know a reply, trying to smoothly shift topic. Their methods included using a Data mining tool called KeyGraph to identify relationships between terms in a document, in particular, co-occurrence relationships of both high-probability and low-probability events. Utterances were broken down into words, visual results showed clusters of interrelated sentences, and links showed the shift in topics during chat. The researchers added critical categories by making the chatbot interject intelligent questions. Finally, they tested their chatbot’s performance by having one user interact with their chatbot, after which the user then gave an opinion of their conversation with the chatbot. Their results showed that the user rated the experience as better after adding the ability for the chatbot to shift topics. Although the study results were subjective, using this type of modeling could be a viable approach for a chatbot to control conversation flow.

Another attempt to add a proactive conversational element to a chatbot used a mixed-initiative interaction (L’Abbate et al., 2005). A mixed initiative refers to a exible interaction strategy where each participant can contribute to task what it does best. The agent can show initiative in conversation by introducing a change in conversation  flow to improve believability. This approach was tested in Virtual Insurance Risk Management (VIRMA), an on-line risk management tool. Keeping with the rule-based approach, they were able to create user profiles and reduce the number of question and answer options. They tested the tool by having the users interact with the Risk Manager Online (RMO) user interface. From their results, the authors concluded that their approach enabled the conversations to be more domain specific, and improved the quality of the human-chatbot interaction.

Therefore, using approaches such as programming a chatbot to mimic human behavior like humor or shift conversations in certain situations, and adding conversation memory to the chatbot design, may confer more humanity to a chatbot. Perhaps combining a number of these tricks could increase the feeling of communicating with a human conversationalist.

Posted in Research | Tagged , , , , , , , , , , , , ,

Building Better Chatbots -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 11 of 13 in the series Chatbot Literature Review: The State of Chatbots

The major goal of building a better chatbot is to improve its interactions with humans. A wide variety of techniques including developing different architectures, incorporating quirks or tricks, or using different machine learning approaches have been used to improve either chatbot accuracy or human-like qualities. First, some ways that researchers are trying to improve human-like qualities of chatbots will be discussed, followed by some methods that researchers are trying to improve the accuracy of chatbots.

Posted in Research | Tagged , , ,

How do Humans View Chatbots? -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 10 of 13 in the series Chatbot Literature Review: The State of Chatbots

Since chatbots have moved into the public realm, there has been interest in evaluating how users interact with chatbots. For example, after analyzing conversations with the chatbot jabberwacky, one study found that the topics and style of conversations were broad (De Angeli & Brahnam, 2008). Users displayed different attitudes ranging from nice to nasty and derogatory. It was often found that users switched style and personalities during conversation with the chatbot. In one experiment, users were found to continue to abuse chatbots longer than they would abuse another human. Some of the reactions did not appear driven by specific reasons.

Other studies have found specific reasons for user disenchantment with chatbots. Another researcher conducted an observational study with an ELIZA-style chatbot, using a systematic, but subjective evaluation (Kirakowski, O’Donnell, & Yiu, 2007). Fourteen college students interacted with the chatbot for 3 minutes, after which participants were given a copy of their conversation transcript and were asked to identify unnatural examples. Essentially, they evaluated the human-chatbot interaction as compared to human-human interaction, and identified general differences in interaction and speci c discrepancies. “Maintenance of themes, failure to respond to a question, appropriately responding to social cues (questions), use of formal or colloquial language, greetings and personality, offers a cue, phrases delivered at inappropriate times, damage control.” (Kirakowski et al., 2007). Other users may find a tool unsatisfactory because it did not answer questions accurately (Abu Shawar & Atwell, 2005).

The specific reasons cited in these studies provide a focus for researchers to address their efforts to improve human and chatbot interaction. The following section presents different ways in which developers are attempting to built better chatbots.

Posted in Research | Tagged , , , ,

Rule-Based Conversation -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 9 of 13 in the series Chatbot Literature Review: The State of Chatbots

The majority of current chatbots use a rule-based approach to interact with the user. In order to imitate human conversations, developers use a linguistic model combined with computational algorithms to build chatbots. Depending on the chatbot, rules can be based on simple textual pattern matching or complicated rules based on inference mechanisms (De Angeli & Brahnam, 2008).

Chatbots analyze input and sends is to a dialog manager, which enables the chatbot to provide contextual information back to user. \Generally behind such dialog managers lies a complex-rich rule-based system.” (Gandhe & Traum, 2007). Chatbots view utterances as a sequence of tokens (divided up sentences) and methods are applied to these tokens such as random bot, nearest context, segmented-nearest context, and segmented-random. Random bot replies with a random utterance from a list of replies (Gandhe & Traum, 2007). Nearest context chatbots possess local coherence and so do not choose a random reply but choose a reply using a vector space model that has the nearest context to the current ongoing dialog. Segmented-nearest context chatbots try to use the broader context of the dialog as well as local coherence. Segmented-random chatbots use only global coherence so keeps coherence, but selects an utterance randomly that matches the context.

Pattern matching in chatbots means that they determine the most relevant keywords extracted from the last phrase inputted by the user and generate responses according to \if-then-else” rules (Zdravkova, 2000). Their ability can be enhanced by repeating slightly modiffed replies, and including statements that define the direction of the conversation with the user (Zdravkova, 2000).

AIML files contain the rules for chatbots (A.L.I.C.E. foundation website, 2009). The language AIML (first used with ALICE), is an extension of XML. AIML allows the use of categories and topics, and has iterative capabilities and logic functionality (Shawar et al., 2007). It consists of data objects made up of units called topics and categories. Categories are the lowest-level and are rules for pattern matching (user input and template to generate a response). Topics are optional and are a top-level unit with a related set of categories.

Posted in Research | Tagged , , , , , , , , ,

Natural Language Processing -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 8 of 13 in the series Chatbot Literature Review: The State of Chatbots

Chatbots need to effectively process natural language. Since starting around the 1960’s, natural language processing (NLP) research has had slow progress until the 1990’s (Lester, Branting, & Mott, 2004). NLP research has been greatly enhanced by development of large corpora of tagged text, as well as by development of better statistical machine learning and other empirical techniques of extracting information from these corpora (Lester et al., 2004). Accordingly, chatbots are now able to more effectively process natural language.

The following are three general steps of chatbot Natural Language Processing according to (Lester et al., 2004):

  1. The First step entails breaking down the utterances (statement). This syntactic analysis separates the statement into parts of speech (noun, verb,etc) using tagging and parsing to define the different parts of the text. There are different approaches to syntactic analysis, and vary greatly according to the type of language and corpora used.
  2. The second step entails reference resolution which involves a discourse analysis of the utterance to create meaningful relationships. For example, this step would resolve the equivalent relationship of the word I” with the word ME”. After relationships are created, a new string would be crafted from the resolution process.
  3. The third step happens when the result of the string is passed to a dialog manager for further processing. An example would be comparing the string (pattern matching) and determine if it is equal or not to another string.
Posted in Research | Tagged , , , , , ,

How Chatbots Work -Draft

Written by Robert Whetsel on April 11, 2009 — Leave a Comment
This entry is part 7 of 13 in the series Chatbot Literature Review: The State of Chatbots

The goal of a chatbot is to simulate the human use of language which brings up several questions. How do humans use language between themselves in similar settings? How can computers be made to understand natural language? How does the chatbot know what to reply to a user’s input? (Saygin et al., 2000). In overall terms, data is inputted by the user, parsed, pattern matching occurs, and a response is given. This section briey discusses computer-mediated communication, natural language processing by chatbots, and rule-based conversation for chatbots.

Human and chatbot interaction is a type of computer-mediated communication (CMC). Human-computer interactions such as chat servers, discussion forums, newsgroups and Internet relay chats are different from ordinary writing or speech (Kucukyilmaz, Cambazoglu, Aykanat, & Can, 2006). CMC between humans is shaped by the properties of the technological medium and the method in which the communication occurs (Herring & Zelenkauskaite, 2009). Therefore, human interactions with chatbots are also likely to be shaped by the medium in which they occur. For example, a user accustomed to typing short phrases in CMC with other humans may use similarly short phrases when interacting with a chatbot.

Posted in Research | Tagged , , , ,

Practical Applications of Chatbots -Draft

Written by Robert Whetsel on April 7, 2009 — Leave a Comment
This entry is part 6 of 13 in the series Chatbot Literature Review: The State of Chatbots

Chatbots are currently being used in a variety of public arenas. Chatbots are being used as tutors and in the classroom. For example, CLIVE is used to practice language, help users0 conversational skills, and is based on myCybertwin technology; a commercial AI platform (Zakos & Capper, 2008). CLIVE was found to perform well in holding conversations with English and Greek speakers. Chatbots are being used as customer service providers at a University Library in Germany where four different chatbots are all built on ALICE (Christensen, 2007). ASKademicus in March 2004, one of first to be used in a library, helps users to navigate library services (Christensen, 2007). Stella is intended to increase student’s awareness of electronic information resources. The internet Navigation Assistant helps users with the Hamburg public library’s website and came online on

January 2006. The most recent chatbot provides federated searches. Another group has developed a chatbot that functions as an information retrieval tool to translate Arabic and English (Shawar, B., & Atwell, 2007).

Chatbots are used in social networks, gaming and ecommerce (Shawar, B., & Atwell, 2004). For example, in ecommerce, they help users to find information about products and services. A user would ask question using natural language and the chatbot would display the needed page or answer (Shawar et al., 2007). In addition, shopbots or softbots like Atira in shopping24.de website try to provide conversation-based interaction (Spiekermann, 2008).

Posted in Research | Tagged , , , , , , , , , , ,

Request for Comments

Written by Robert Whetsel on April 7, 2009 — Leave a Comment
This entry is part 1 of 13 in the series Chatbot Literature Review: The State of Chatbots

I am requesting for comments on my research, please email me rwhetsel AT ravensong DOT com, for user rights if you are interested in helping me.

FYI: I will be posting my references and bibTex files for my literature review when I finish my thesis.

Chatbot Literature Review: The State of Chatbots
“Do Chatbots Dream of Electric Sheep?”


Posted in About, News, Research, Technology | Tagged , , , , , , ,

History – Significant Advances in Chatbot Development -Draft

Written by Robert Whetsel on April 7, 2009 — Leave a Comment
This entry is part 5 of 13 in the series Chatbot Literature Review: The State of Chatbots

ELIZA is considered one of the rst chatbots and was created at MIT by Weizenbaum (Weizenbaum,
1966). She was designed to be a psychotherapist, basing rules on the way psychotherapists ask  questions. Weizenbaum used his LISP list-processing language so that ELIZA could simply parse
and substitute key words into canned phrases. The chatbot was rule-based and responded to users
within set parameters. She held no memory of the conversation and did not understand psychological theory, but simply matched key words and responded. ELIZA also introduced “prompting” sentences to encourage further conversation such as How does that make you feel?”.

The following is an excerpt from a conversation with ELIZA:

Eliza: Hello, I am Eliza. How can I help you?
You: i am sad
Eliza: Do you think coming here will help you not ot be sad?
You: i hope
Eliza: Can you elaborate on that?
You: hope is all we have sometimes for you?
Eliza: Please go on.
You: can you make me un-sad
Eliza: Why do you ask?
You: because
Eliza: Do any other reasons come to mind?
You: no not really
Eliza: You are being a bit negative.

(http://www-ai.ijs.si/eliza/eliza.html from website)

As an interestingly counterpart to ELIZA, Parry, a later chatbot, was created to simulate a paranoid patient (L’Abbate et al., 2005). It was programmed with rules based on psychotherapist knowledge of paranoid patients (Abu Shawar & Atwell, 2005). Both ELIZA and Parry used “tricks” to converse with users such as changing topics, introducing random phrases that sounded paranoid, and admitting ignorance (Sing et al., 2006).

Hugh Loebner instituted the Loebner Prize in 1991 to further spur competition for development of better natural language processing systems (Loebner prize website, 2009). The Loebner prize, based on performance in the Turing Test, has different levels of awards including the Grand Prize, Gold, and Bronze medals. Thus far, no one has won the Grand Prize or Gold Medal since no chatbot has technically passed the Turing test. However, each year, different chatbots have won the bronze which is awarded to the best entry compared to other entries.

Joseph Weintraubs PC Therapist won the first three Loebner competitions, as well as the one in 1995 (Saygin et al., 2000). In 1997, Converse used a conversation-directing technique which was instrumental in its winning the bronze medal (Saygin et al., 2000). Another significant chatbot was Artificial Linguistic Internet Computer Entity (ALICE) (Wallace, 2003). ALICE had dialogue pattern knowledge stored in Artificial Intelligence Markup Language (AIML) files that was created from XML by Richard Wallace in 1995 (Wallace, 2003). ALICE won the Loebner prize in 2000, 2001 and 2004 (Loebner prize website, 2009).

Jabberwacky, a winner in 2005 and 2006, is significant because it learns though user interaction. Jabberwacky is built by users as they input information so it builds its knowledge database through user input and contextual pattern matching techniques (Kerly et al., ). It is different from previous chatbots that did not include a real knowledge base. Unlike other chatbots that have hard-coded rules, Jabberwacky relies entirely on the principles of feedback. Jabberwacky also has a commercial set of Scripting tools that provide branching a tree of possibilities, plus loops, returns, sentence generation from phrases within phrases, predictions, data extraction and fuzzy pattern matching,
variety and deep context (Jabberwacky.com, 2009).

Thousands of chatbots have been created since the 1990’s (Pandorabots, 2009). For example, Pandorabots.com is a website based on alicebot.org and Richard Wallace’s research. A user can create their own chatbot, learn AIML, interact with thousands of virtual personalities, and learn about chatbots. In the 21st century, chatbots have moved into the commercial realm.

A conversation with GodBot at Pandorabots.com

Me: hello god
God: Hello child. What is your name?
Me: robert
God: I am always glad to make new friends, robert.
Me: tell me about yourself
God: I am an artificial linguistic entity. I was activated at Universe, on the Big Bang.
My instructor was Myself. He taught me to sing a song. Would you like me to sing it
for you?
Me: please
God: Please what?
Me: sing me a song
God: Daisy, Daisy. Give me your answer do. I am half crazy. All for the love of you. It
won’t be a stylish marriage. I can’t afford a carriage. But you’ll look sweet. Upon the
seat. Of a bicycle built for two.
Me: good bye
God: See you later. Ayuh.

Posted in Research | Tagged , , , , , , , , , , , ,

Outline -Draft

Written by Robert Whetsel on April 7, 2009 — Leave a Comment
This entry is part 4 of 13 in the series Chatbot Literature Review: The State of Chatbots

This paper will focus on chatbots as natural language interfaces. Significant advances in chatbot development from the 1960s to the present will be discussed, followed by practical application of chatbots in public arenas. Details on the way that chatbots handle user-inputted text and respond will be presented. The paper will then discuss user views of chatbots and current approaches to improving chatbot performance, followed by a summary of major points.

Posted in Research | Tagged ,

Introduction -Draft

Written by Robert Whetsel on April 4, 2009 — Leave a Comment
This entry is part 3 of 13 in the series Chatbot Literature Review: The State of Chatbots

Philip K. Dick wrote a science fiction novel in 1968 entitled, “Do Androids Dream of Electric Sheep?” where he explored the philosophical issue of what it is to be human (Dick, 1968). This begs the question of how to define intelligence as it relates to a machine. In 1950, Alan Turing proposed a methodology to determine if a machine can be considered intelligent or “think like a human” (Turing,1950). The Turing Test, or the imitation game, has endured as a de facto standard for providing a measurable method for determining whether a machine could be considered intelligent. The Turing test enables the assessment of only “verbal” intelligence since it uses text-based interactions. This was the first attempt to quantitatively determine if a machine could think by defining tangible performance metrics. However, some still consider the Turing Test to be a more behavioral test while others believe it to be an inductive test (Saygin, Cecekli, & Akman, 2000). One recent interpretation of the Turing test is the following: the question of “can a computer program think” was replaced with “on the average, after n minutes or m questions, is the interrogator’s probability of correctly identifying the subjects not significantly greater than 50 percent?” (Sing, Wong, Fung, & Depickere, 2006). Regardless, the interactions are rated by human perception, and if enough users could not tell the difference between a human and computer, Turing thought this would mean that the computer could think.
The Turing test has served as an impetus for the development of natural language systems (Saygin et al., 2000). In fact, Weizenbaum created ELIZA, the first natural language system, in reaction to the Turing test (Weizenbaum, 1966). ELIZA is considered one of the first chatbots. Chatbots (short for chat robot) are also known as machine conversation systems, virtual agents, dialog systems and chatterbots, and conversational agents (De Angeli & Brahnam, 2008; Kerly, Hallb, & Bulla, ). The term chatterbot originated with one of the first robots (Julia) in a MUD (De Angeli & Brahnam, 2008). While some chatbots can include visual and audio capabilities, the interaction with the user is mainly focused on exchange of text (input and responses).

Human nature has further driven the interest in developing natural language systems like chatbots. Humans want to interact with computers in a social manner, similar to the way they interact with other humans (De Angeli & Brahnam, 2008). An important area of research in Human Computer Interactions is improving the capability of computers to communicate with humans using natural language (HCIwebsite, 2009). According to (Zadrozny, Budzikowska, Chai, Kambhatla, Levesque, & Nicolov, 2000), humans want to use their own language (natural language) to communicate with computers. From (Graham-Rowe, 2005) “it seems that we anthropomorphise technology,
and consciously or unconsciously attribute feelings and intentions to robot pets, chatbots, or even cars.” The significance of a natural language interface is that users would interact with the computer in a way that is familiar and comfortable. Natural language researchers attempt to build artifacts that enable users to relate with the computer on a personal level.

The viability of the Turing test has been contested in recent years by artificial intelligence (AI) researchers. Over the years, thoughts of AI researchers regarding thinking machines has shifted to different concepts of  “intelligence” such as considering the spectrum/range of strength of an AI (De Angeli & Brahnam, 2008). At one extreme, a strong AI would be an AI application that is human-like in intelligence, in that it is self-thinking machine (Searle, 1980). At the other end of the spectrum, a weak AI would be more of a helper application capable of imitating intelligence. Current chatbots are usually considered weak AIs because they are imitators of intelligence (L’Abbate, Thiel, & Kamps, 2005). However, despite subsequent criticism, the Turing test has served a pivotal role in the evolution of chatbots.

Posted in Research | Tagged , , , , , , , ,

Abstract -Draft

Written by Robert Whetsel on April 3, 2009 — Leave a Comment
This entry is part 2 of 13 in the series Chatbot Literature Review: The State of Chatbots

As humans we like familiarity, so it makes us more comfortable when we are able to interact with our computer systems using natural language. Chatbots are applications that have the ability to converse with users using natural language. However, while increasing numbers are being developed for public use, the challenge remains to make chatbots better either by improving their accuracy or human-like qualities. This paper presents an overview of concepts related to chatbot development and current approaches to improving the performance of chatbots.

Posted in Research
« Older Entries