Category Archives: Chatbots

What can we learn from the ELIZA effect?

Weinbaum’s experiments with ELIZA proved that when we know we aren’t being judged we are happy to talk about anything and even divulge personal information. The ELIZA effect as it is known, addressed the idea that we as humans presume that the behaviour of computers is as analogous as that of humans. Created as a psychotherapy chatbot ELIZA provided a disinhibited low-anxiety environment for patients to talk about their problems. With patients assuming that the computer programme was responding in a purely analogous fashion, and not in the pattern matching way that it actually was.

The ELIZA model has been repeatedly emulated with the creation of chatbot apps that provide virtual friendships and emotional support, such as Woebot, Replika, and Wysa. These therapy bots aim to help people combat depression and loneliness, and feel they have ‘someone’ to turn to. This demonstrates that our willingness to communicate (WTC) is enhanced when the interlocuter we are conversing with is unable to judge us.

This leads me to the main argument of this post. It would appear humans feel more comfortable communicating with chatbots that to date do not possess the AI capacities to fully understand and interpret human emotions. Therefore, the fear of being judged or losing face is drastically reduced. In the language learning classroom, we should therefore try to create a relaxed environment that facilitates learning and help promote WTC so learners feel more comfortable to interact orally and more confident to express their ideas. So while machines endeavour to hone their AI skills to perfectly emulate human behaviour, maybe we as teaching practitioners should try to emulate machine behaviour by encouraging a non-judgemental environment in the language learning classroom that promotes confidence among learners to speak and interact more confidently, especially in online environments where learners appear to feel more reluctant to speak up.

Our expectations of digital language learning partners

My current investigation into speech interfaces as language-learning partners is revealing that one of the main problems companies are facing when developing these tools is that they do not behave how they would like them to. This refers to the product often lacking the necessary layers of programming necessary to design a bot that can fully simulate human-human interaction.

This is does not come as a surprise to me because I firmly believe that until we reach a stage where we fully understand the human brain. I believe we are still quite far from reaching this stage, so I find the efforts to try and build an AI tool that can fully replicate the human brain and fully simulate human-human interaction akin to herding cats. To some extent it can be done, but not quite as we’d like or expect.

From my own experiences of building a conversational bot I appreciate the intricacies of the programming required to build a tool which acts as we would like. It is time consuming, arduous, and extremely challenging to say the least. This is why, I presume, that the English language learning landscape is not flooded with such tools.

I am currently examining learner reactions to spoken digital interface interaction by trying to understand how learners respond and what it is specifically that makes them respond in different ways. My hope is that by better understanding user discourse, it will provide some insights into the characteristics an effective chat interface requires.

Enthusiasm to learn is emotionally driven

Enthusiasm can be displayed in different ways and it can also be present in a learner, but they do not show any visible signs of being enthusiastic to learn, they are simply enjoying the learning and keen to learn.

My current research investigation focuses on how learners demonstrate their enthusiasm when interacting with a speech recognition interface. This includes both linguistic and non-linguistic features. The dataset I am using clearly demonstrates that the psychological state of learners impacts their enthusiasm, and therefore language output and capacity to engage in learning more than any other factor. While this came as a surprise, it aligns with motivation theory and learning which purports that positive emotional and hence psychological states favour learning, and a negative emotional state (anxiety, stress, depression) can adversely affect learning.

I’ve spent a lot of time with humanoid robots, speech recognition interfaces, and autonomous agents and despite their degree of humanness, there is something decidedly safe for me about interacting with a non-conscious being. Maybe that is why Weizenbaum’s research was so successful! The non-judgmental attributes of a machine make the user feel comfortable to interact, and therefore they get more out of the learning experience. This is something I am still investigating, but Buddy, the robot in the image above aims to understand the mood of the use, and then respond accordingly. So empathy is now going beyond human…

Less is more: the argument in defence of HCI for speaking skills

Less is more, or is it?

I was taught from a young age that the wise man is the one who observes and says very little. However, for foreign language learners I think it is quite the opposite, and the more they try to speak and express themselves orally the more they can practise and learn about oral interaction.

My current research is investigating the oral output prompted by interacting with an autonomous agent, and surprisingly I am not finding that the output varies from that of human-human interaction. There are days where participants are motivated and enthusiastic to interact and others where they provide monosyllabic answers.

Where I’m going with this, is that investigating learners interacting with a digital tool has demonstrated to me that in the classroom I often have an expectation of learners to constantly perform, and feel frustrated when they don’t willingly provide output when requested. I am learning that deliberate practice is perhaps not an effective method of language learning and adopting a more laissez-faire approach maybe more appropriate.

So, on the one hand we need learners to speak as often as possible, but on the other hand we can’t expect them to always be willing to speak. For me this highlights the value of human computer interaction (HCI) for language learning and demonstrates that we should lean more heavily on autonomous agents for speaking practise. They provide limitless opportunities, never tire and can be used when learners feel they want to speak, not when they have to.

Hello, can I help you?

Facebook messenger bots can be set up and working within an hour. It is no wonder then that text-to-text chatbots have replaced the automated customer service answer machines in many sectors of industry.

The chatbot can be programmed with a training corpus of customer service complaints in the form of recognisable input data, and possible solution phrases. The algorithms then use key word identification to identify the issue and match it with a suitable response. Given the many experiences of miscommunication with lackadaisical customer service telephone operators, I feel this is a perfect use of chatbot technology.

I have been experimenting with building a bespoke chatbot for my own research purposes, so I can confirm that the practice is comparatively complex compared to the theory of providing an interactional partner for learners of English as a second language. Using the model and frameworks of customer-service chatbots was not possible to modify in my case. I tried using the Dialogue Flow framework provided by Google, which surprisingly provided rather disappointing results.

I feel the fear of a digital world where machines take over from humans is somewhat premature, as there is still a lot of development needed in order to iron out the creases of chatbot technology.

Disinhibition and Human Computer Interaction

For some reason, when we are learning a foreign language, we feel intimidated to speak it. We fear we will be laughed at, won’t say the right thing and won’t be understood or simply lack the confidence to put a voice to the words floating around in our brains forming utterances.

It is clear inhibition to speak is a common problem among language learners for whatever reason. So, I am investigating strategies to disinhibit learners, and to provide them with oral interaction confidence, by engaging with a computer to practice speaking, so they have the confidence to interact with humans.

Human computer interaction (HCI) to practise English conversation offers several advantages compared to practising with a human. The main motivations being:

  •  low inhibition because learners know are they are interacting with a machine that will not judge their performance unless asked to do so.
  • a low-anxiety environment which promotes confidence because of the absence of a human waiting form the next turn.
  • Interaction for as long as the learner wants to practice.
  • Computers do not lose their patience, or tire of conversing or repeating the same conversation pattern.

I therefore strongly believe that HCI is a promising solution for learner disinhibitition.Updates on experiments carried out with chatbots to fulfil this research to follow…

Man or machine?

Man or machine? That is the question! There is an endless flow of information being pushed onto our screens about the danger of robots and machines taking over the world. Martin Ford’s Rise of the robots (2015) presents a blatantly bleak view of automation and the ‘threat of jobless future’ due to the advances of technology.

When it comes to automated customer service agents, I am sure we all have long winded stories of negative experiences. On the flip-side however, I have also had my share of less than favourable customer service experiences with humans. While there is evidence of the frustrations of not being able to interact with a human to resolve customer service issues, there is considerably more evidence which supports the view that the human was unable to resolve the query, and a chatbot could have more than adequately dealt with the matter in a considerably shorter time frame (Xu et al, 2017). Chatbots are also consistently patient and polite; remain unruffled by rude customers, high traffic, or repeated requests for the same information, and never tire (McNeal & Newyear, 2013).

I think there is a time and a place for everything. But given the inflated lack of patience and the abundance of immediacy that humans expect from the service sector nowadays, I think chatbots are a good option for quick enquiries and to resolve systematic ‘problems’.


From RALL to chatbots

I began the year with a strong desire to continue my research into RALL, and while that is still the case, my research has lead me to investigate the benefits and  pedagogical potential of using chatbot teachers to assist in language learning.

The research examines the use of a speech-to-speech interface as the language-learning tool, designed with the specific intention of promoting oral interaction in English. The computer (chatbot) will assume the role of conversational partner, allowing the learner to practice conversing in English. A retrieval-based model will be used to select appropriate output from predefined responses. This model will then be mapped onto a gamification framework to ensure an interesting and engaging interactional experience.

Speech is one of the most powerful forms of communication between humans; hence, it is my intention to add to current research in the human-computer interaction research field to improve speech interaction between learners and the conversational agent (the chatbot) in order to simulate human-human speech interaction.

So just how should you speak to a chatbot?

So just how should you speak to a chatbot? If you cast your mind back to Tay the chatbot built by Microsoft. She was shut down on the grounds of inappropriateness because she was posting offensive and unsuitable content on her Twitter account. Hardly surprising really considering she was built using corpus from Twitter posts and dialogues, a perfect example of the hunter becoming the hunted.  

The apparent ubiquity of chatbots in the customer service sector is proving to be somewhat beneficial to the companies using them, but less convenient for users. The majority of conversation agents are built using a retrieval-based model, which reply based on a repository of predefined responses from which the most appropriate is selected, based on the input text and context. The output could be limited to as little as three utterances per response. Let’s look at an opening turn to see how this works:

‘Hello, what can I do for you today?’

> No response, delayed response from user, or the chatbot is unable to interpret the user input.

‘I missed that, say that again?’

> No response, delayed response from user, or the chatbot is unable to interpret the user input.

‘Sorry, can you say that again?’

> No response, delayed response from user, or the chatbot is unable to interpret the user input.

‘Sorry, I can’t help.’.

This leads me to believe that we as users need to learn how to speak to an automated conversation agent before determining what we want from it. If we don’t respond, or respond using undecipherable discourse then we are expecting a machine to manage a task that humans would also face problems with interpreting. While considerable research and development is being carried out in the field of intelligent conversational agents, we are still a long way from them becoming an integral part of mainstream customer service interfaces that are able to interpret our utterances and commands to the best of our expectations.

Turn taking and chatbots

Turn taking is a natural part of conversation that we subconsciously engage in order for the discourse to flow. Here is an example:

A: “Good morning”

B: “Morning. How are you? Good weekend?”

A: “Yes thanks, and you? How was Brighton?”

For the Cambridge main suite speaking exams, candidates are assessed on their turn-taking ability under the criteria of ‘Interactive Communication’. In other words, this means the candidates’ ability to:

  • Interact with the other candidate easily and effectively.
  • Listening to the other candidate and answering in a way that makes sense.
  • The ability to start a discussion and keep it going with their partner/s.
  • The ability to think of new ideas to add to the discussion.

Along with the onslaught of technological advances came advance in automated responses from portable digital devices. These conversational agents or dialogue systems are capable of single interactions or up to 6 task-oriented turns. An example of these dialogue agents would be Siri, and an example of a talk-oriented interaction would be: “Siri call Dad”.

Chatbots are not a ‘new’ invention per say. Eliza, created between 1964-1966 at the MIT was a natural language processing computer programme that demonstrated the same characteristics of chatbots day, but on a less sophisticated scale, and with less complex interaction. The aim of chatbot builders is to create natural language processing programmes that replicate human-human interaction by enabling more turns and therefore extended conversations.

The interesting challenge then becomes, how to use each turn as a springboard for the next, and ensure that each one prompts a response that has been pre-programmed, in order not to receive a generic message like: “I’m sorry, but I’m not sure what you mean by that”, when the user is expressing a specific request or a expressing a turn that is not recognised. More about chatbots soon!