The agent was made that the interlocutor
The agent was made that the interlocutor
Помогите с английским Rewrite the sentences which use the wrong past tense.
2. Z40, the new drug developed by Pharmatek, marked a turning point in the treatment of cancer.
3. Their competitors failed to see the gap in the market and so missed a great opportunity.
4. The agency was not believing that the machine would save so much time.
5. Zirkon already made good profits when it introduced its new digital camera in 2000.
6. The story goes that he was having the idea for the electric shoebrush while he was washing up.
7. I was planning to visit the International Inventors’ Fair but I did not have time.
Первые три грамматически верны.
А вот дальше будет так:
4. The agency did not believe that the machine would save so much time.
5. Zirkon already made good profits when it had introduced its new digital camera in 2000
6. The story goes that he had the idea for the electric shoebrush while he was washing up.
7. I planned to visit the International Inventors’ Fair but I did not have time
Перепишите предложения, которые используют неправильное прошедшее время.
2. Z40, новый препарат, разработанный компанией «Фарматек», ознаменовал поворотный момент в лечении рака.
3. Их конкуренты не заметили разрыва на рынке и поэтому упустили прекрасную возможность.
4. Агентство не верило, что аппарат сэкономит столько времени.
5. Зиркон уже получил хорошую прибыль, когда представил свою новую цифровую камеру в 2000 году.
6. История гласит, что у него была идея электрической щетки для обуви, когда он мылся.
7. Я планировал посетить Международную ярмарку изобретателей, но у меня не было времени.
B. Make the following sentences negative.
1. It should not be unclear whose job it is.
2. After a short certain period of employment the employee cannot qualify for a pension.
3. She’s a good people person who cannot motivate a team.
4. He can’t have known some of the interns.
5. He cannot be very tired.
6. You should not work more less energetically.
7. You should not take into account everything he says.
8. You must not apologise for the absence.
9. The applicant cannot speak French very well.
10. It may not be true.
11. You may not be late for the interview.
12. You must not go there to see the senior manager.
13. We ought not to hire this man. His qualifications are undisputable.
P. 195
Exercise 1
Complete the sentences using can, could or be able to in the affirmative or negative form.
1. In the mid-1980s only 80% of all teenagers who wanted to work were able____ find a job.
2. Governments will be able_________ take legal actions to restrict or halt business activities of monopolies through anti-trust laws.
3. Though he was a good engineer he _could not/ was not able __ find a job.
4. I _cannot________ understand why you want to meet the manager face-to-face.
5. How long have you __been able to_______ control the collective bargaining process?
6. The manager wasn’t in the office for very long, but we _were able___ speak to him for a few minutes.
7. Be careful when you fill out an application, because this step ___can______ be critical.
8. He always bought a lot of newspapers and thus ____was able to_____ read all ads. It didn’t take him long to find a job.
9. As public job agencies __ could not/ were not able _______ help you, try a private firm.
10. John cannot cope with the increasing public relations work. He needs somebody to help him.
11. The company were looking for a people person who __was able to/could____ join their team.
Exercise 2
If the word or words in bold type are used incorrectly, write the correction in the space.
1. Newspaper ads were able to become a good source of openings, but there were no newspapers in that area. __could have become____
2. He could have got the job, but he didn’t even try to. _____+____
3. The employer wasn’t able to look through all the applications at once._____+____
4. We could have known all about prices for that commodity in the market as we studied the problem closely. __were able to know_______
5. They could have contacted him, but he didn’t leave his name and address. __+_______
6. You were able to read the instructions if your wife didn’t throw them away. ____could have read_____
7. Peter couldn’t have believed that the managing director had really appointed the female applicant advertising manager. ___couldn’t believe______
8. She didn’t speak very frankly, but I could understand that I was quite high on her list. ____+_____
9. He could only brief the agency on the kind of advertising campaign we want because the new advertising manager turned out to be quite efficient. ___+______
10. She could have become a facilitator a year ago, but decided to join a training scheme for a year. _____+____
11. A lot of people thought the manager could resolve issues, but he didn’t. ___(was able to)______
12. The firm could have negotiated salaries and raises, but they didn’t plan to do so. ___+______
P. 197-198
Exercise 1
Complete the sentences using one of the words or word combinations.
1. If things are faulty when you first use them, you… go back to the shop
and ask for complete refund. (A) can
2. … I ask you to show me that shredding machine in operation? (B) Could
3. All troubles arose from the fact that Fox & Co…. import their produce. (B) were allowed to
4. Could I look through your magazine? – Of course, you…. (A) may
5. Now everybody… use these copiers in the hall but for some fee.(A) is allowed to
6. Until the 19th century people… travel freely between most countries without a passport.
7. After talking to the director we… make a special bulletin board for ads. (B) were allowed to
8. Applicants… to present details of referees who could attest to the quality of their work.
9. The firm… hire their own expert in PR as it was much too expensive. (B) couldn’t
P. 200-201
Exercise 1
Choose the correct form.
1. You mustn’t/needn’t present the documents in person.
2. Your decision mustn’t/haven’t to depend on your tastes, but of the necessity.
3. You mustn’t/don’t need to look through all these advertisements. Here is a letter from Mr. B., he is hiring you.
4. You mustn’t/don’t have to write to Bell & Co., they have just phoned.
But we must (have to) send our agent there as they insist on it.
5. You mustn’t/don’t need to negotiate with the supplier, he must (has to) provide a replacement by law.
6. Robert didn’t need to sell/needn’t have sold his car when he didn’t have any money, because a few days later he won a lot of money in a competition.
7. – What sort of advertising campaign are you planning? Television? – Well, it mustn’t/needn’t be television – it’s too expensive. But it must/need to be efficient.
9. This is very important. You must (have to) study it carefully and mustn’t/needn’t ignore it
10. You needn’t have sold/didn’t need to sell your house to the first bidder. I know several people who would have paid more for it.
11. Customs needn’t have checked/didn’t need to check all the boxes as they found what they were looking for in the first one.
12. You needn’t have made/didn’t need to make any changes in your office, you wasted your money.
Cambridge B2 First (FCE): Everything You Need to Know
Cambridge B2 First (FCE): Everything You Need to Know
Cambridge B2 First: Everything you need to know
When I moved to Spain I found that there are a lot of English schools and I started wondering why. What I found out is really interesting. Many parents send their children to get extra lessons because at secondary school classes are pretty bad. The teachers are not properly qualified and the children don’t learn as much as they could. On top of that, several states in Spain are changing the law and students have to show B2 level in a second language to graduate from university. Therefore, Cambridge B2 First has become more and more important, not only in Spain, but in many countries in the world. Have a look at this search map for the search term ‘FCE’ which I found online. You can see that people almost everywhere are interested in the exam.
Source: Google Trends
In this article, I’m going to show you what passing the FCE exam means, who should take the exam, who shouldn’t, what the exam looks like and how you can find a preparation centre close to you. This is the perfect place for you to get started if you want to find out more about Cambridge tests like B2 First or if you have already decided that you need to take the exam.
To find general information and motivation to improve your English, check out my other articles here.
What does B2 stand for?
B2 is a level on the CEFR scale designed by the Council of Europe. CEFR is an international framework that describes different English levels and what someone at this level is able to do in that language. In short, B2 means that you are an independent user of the language so you can live in foreign country without the need for much help. First of all, though, let’s have a look at the CEFR to make everything a little bit clearer (You can click here to get more information:
Source: Council of Europe
You can see that B2 is already a pretty high level, but it is often not clear what exactly a student at A2, B2 or C1 can actually do in their second language. To understand the skills that a student at B2 level possesses we can check again with the Council of Europe:
Source: Council of Europe
This is a detailed description of the the different skills reading, speaking, writing and listening at level B2. If you think you can do all of these things without too many problems, you might be able to take the Cambridge B2 First exam with some preparation soon. However, if you don’t think your skills are good enough, yet, you should probably wait a little longer and join classes to get you to the right level before thinking about taking the FCE exam.
If you would like to get the same information in your native language, you can click here.
Should you take Cambridge B2 First?
The FCE exam is very useful in many countries around the world and it can help you to get higher education or better employment. As a result, it has become more and more popular to take the exam in order to stand out from the crowd. Here are some specific reasons why you should take the B2 First:
1. University
If you live in a part of the world where you need to show proof of B2 level in a second language to get a university degree, then, of course, you should do it. In so many different industries and areas, for example marketing and advertising, film and music, communication, information technology and many more, English is still the number-one language in the world. Studying English from an early age and taking Cambridge B2 First can be a big advantage if you want to go to university and, later, get a good job, which leads us to the next reason.
2. Employment
As described above, English can help you obtain a university degree and with higher education come better job opportunities. However, a second language can help even in jobs which don’t require university level education. In our globalised world more and more companies do business abroad and need employees with a good level of English or other second languages. The FCE exam is proof that you have level of English which you can easily communicate in. If I had to choose who to hire for my transportation and logistics company, I would definitely check language abilities. So, if you are looking to improve your career prospects, then yes, take Cambridge B2 First.
3. Challenge yourself
Many people want to see the world, go travelling and meet new people in the process. Others want to be able to watch American or English films in the original version without having to use subtitles or voice over. The FCE exam can be a challenge to prove to yourself that your English is good enough to survive in another country, to enjoy English-speaking art forms or to do whatever else you want to use English for. Taking an exam can be a goal to work towards and to stay motivated and, therefore, push yourself to make progress continuously. If you are one of those people, then, again, take Cambridge B2 First and show the world and yourself how good you are.
Prove to the world who’s the king of the jungle
Why you shouldn’t take the exam
Even though it is a great idea for many people to take Cambridge B2 First, there are also a few reasons not to take it. Let’s have a look at what these are:
1. Your English level
Many people have experience with exams even before they decide to take FCE. At school, university, work or even for your driving licence you need to pass some kind of test before you get what you want. Unfortunately, it is possible to squeeze through many of those tests and barely pass, and as they say, ‘A good horse jumps only as high as it needs to’.
With Cambridge B2 First this is definitely not the case. I have seen a lot of students who, after speaking to me, didn’t take my advice and ended up unhappy with their results. To pass you almost have to be too good for the level. The exam is exhausting, difficult and really tests your English level. If you are not a true B2 level English student with the right strategies, you won’t pass.
For this reason, don’t take the test if you are not sure about your level or if your teacher advises against it. It will be in your best interest.
2. It is simply not important enough to you
If you read this article you probably don’t fall into this group, but some people study English simply for pleasure and without any higher goals. They don’t need or want to challenge themselves too much, but instead enjoy the learning experience and study at their own pace. They might be studying English to be able to order food when they are on holidays or have a conversation with their son or daughter-in-law from the US.
I have absolutely no problem with this group of English students because very often they are more relaxed and enjoy the little success stories more than others. Taking an exam like B2 First, however is probably not the right thing for them. First of all, the exam is not cheap (more about that later) and the stress level is pretty high.
So, for all my casual English students out there, don’t worry about FCE too much. Enjoy the English journey as much as possible and good luck. :). For all the others, keep reading.
What does Cambridge B2 First look like?
Cambridge B2 First is an exam that tests the four language skills reading, writing, listening and speaking as well as grammar and vocabulary. It is broken up in four papers which I’m going to describe below:
1. Reading & Use of English
In the past the two parts were separate, but with the last FCE update in 2015 students take them together in one exam paper. In this part of the exam students have to show how well they can control different grammar and vocabulary and also deal with a variety of texts, for example newspaper articles, magazines or fiction.
There are seven parts to Reading & Use of English for which you have 75 minutes to complete. The time limit is one of the biggest reasons why many students have problems here as you get around 10 minutes for each task. You have to read quite a lot in a short time (ca. 2,200 words) so you have to be ready. Because this paper includes Reading AND Use of English it counts for 40% of your final mark.
The different parts of the Reading & Use of English paper
2. Writing
The challenge in the FCE writing paper is that you have to write quite a lot in a fairly short time. There are two parts and you get 80 minutes to complete them, which means that you have around 40 minutes for each writing task.
The first task is always an essay and the second one a choice from three options. Possible types of writing tasks in the second part are articles, reports, reviews, emails or letters.
Each writing task has to be between 140-190 words long, which makes 280-380 words in total. This doesn’t sound like a lot, but there are also specific requirements for each task. Students have to show that they can answer complex questions, choose the correct layout, use appropriate language, organise their text in a logical way and show a wide range of grammar and vocabulary.
For me, writing is one of the best papers in Cambridge B2 First to get good marks. You can find lots of example tasks online and practise each one of them until your fingers bleed and you run out of ink in your pens.
Quick note: If you need some writing motivation, check out this article and if you want to see an example, you can download it here from the Cambridge website (no virus, I promise).
3. Listening
The listening paper in Cambridge B2 First is another difficult one that quite a few students hate. The problem, however, is not so much the level of difficulty itself, but rather the fact that you listen for 35-40 minutes without breaks. For a lot of people it can be exhausting and with time, attention and focus become more and more difficult to keep up.
In the FCE listening test there are four parts. Each part is a little bit different in what kind of recording you listen to and what exactly you have to do and, as with the other parts, you can have a look at an example by clicking here. You listen to each recording twice so you don’t have to panic if you can’t hear an answer during the first round.
The different parts of the Listening paper
4. Speaking
The fourth and last part of Cambridge B2 First is speaking. In my opinion, the speaking exam, like writing, is another good way to get a lot of marks as it is always the same structure. This means that you can practise a lot to become better quickly and feel confident and relaxed on the day of the exam.
The FCE speaking paper is done in pairs or groups of three and takes around 14-15 minutes in total. In some parts you speak alone, but in others you interact with your partner(s). There are two examiners in the room with you – one is the interlocutor (That’s the person who asks you all the questions) and the other one stays in the background and takes a lot of notes to make sure that you get the right marks.
The different parts of the Speaking paper
There are a lot of example speaking exams online for free. You can watch them on YouTube to get a good idea of how it all works. Here’s one for you:
For tips on how to improve your speaking and great ways to practise, read my article here.
Are there other B2 First exams?
There are two more FCE exams which are very similar to the traditional one. The papers are the same, but they might have different content or a different mode of delivery.
1. B2 First for Schools
This version of the exam is basically the same as the one described in this article. The only difference is that the contents for preparation and in the exam itself are more suitable for teenagers and schoolchildren. You can find more information on Cambridge B2 First for Schools here.
2. B2 First (computer-based)
For a few years now Cambridge English has been offering the B2 First exam in a paper-based and a computer-based version. The contents are exactly the same, but the delivery method is obviously different. Have a look at this tutorial video to see what the computer-based version looks like so you can make an informed decision.
Where can I find a test centre for B2 First?
There are a lot of centres all over the world where you can sign up for the exam and also book classes to prepare. Very often these centres have something in the window or on their website that identifies them as an official preparation centre:
These schools usually offer good-quality classes to get you ready for the exam, but do your research in your area and ask the schools what their classes look like. Many schools offer free trials where you can have a look and see if you like it or not.
On the official website of Cambridge English there is also a search tool to find test centres for Cambridge B2 First in your area. Click here to get to their site.
Enjoy the journey
I know that this is a lot of information, but Cambridge B2 First (FCE) is an important exam for many people. If you decide to go on this journey and prepare for it, you need all the information you can get. I wish you all the best and success.
sharon’s@home
A. USE
We use the passive voice when we are interested in the activitity rather than in who or what does it.
e.g. This television has been made in Japan.
The important thing is that the televison has been made, not who made it.
e.g. The deficit should be reduced.
In this phrase we give importance to the deficit.
The passive is used when the agent is unknown or not important.
e.g. Thousands of people are feared dead.
We may not know how or where these people died, however, in this phrase we put emphasis on the possible number of people that are feared dead.
e.g. A Japanese nuclear plant was damaged yesterday.
We might not know why until we read the whole news report.
Note: The passive is often used when the action is more important than the agent as in newspaper reports, headlines, advertisments, formal notices and in scientific experiments or processes.
B. FORM
In passive sentences we use:
To be ( in the correct form ) + the past participle of the verb.
e.g. They will resolve the problem in the Nuclear plant.
The problem in the Nuclear plant will be resolved.
e.g Technicans are repairing the Nuclear Plant.
The nuclear plant is being repaired.
1. PASSIVE WITH AN AGENT.
e.g. A team of nuclear experts solved the problem.
The problem was solved by a team of nuclear experts.
e.g. A Tsunami hit Japan.
Japan was hit by a Tsunami.
2. PASSIVE with the preposition “with” for contable and uncontable objects.
Be careful! There are always exceptions! We will use with + instrument or material to say what instrument or material the agent used.
e.g. Fortunately the houses were built with wood.
e.g. Sushi is made with raw fish and other products.
e.g. Paella is made with lots of ingredients.
e.g. He was hit with a handbag.
If the passive agent is a person that performed the action we use by. Take a look at the exception.
e.g. The houses were destroyed by a tidal wave.
e.g. Sushi is made by the best Japanese chefs.
e.g. The best paella in town is made by Sharon.
e.g. He was hit by a car.
Thus, when we are talking about natural disasters or transport, we will use “by”.
3. PASSIVE WITH MODALS VERBS
When we have a modal verb (can, must, will, should, might……..) in the active sentence, we have to leave it in the passive too. So the form will be as follows:
Modal + Be + Past participle of the verb.
Let’s see an example.
e.g. Sharon should contact Kaoru. (active)
Kaoru should be contacted. (passive)
4. PERFECT PASSIVE MODAL VERBS
Modal + Have + Been + Past participle of the verb.
e.g. Ventura must have bought the newspaper that I like today.
The newspaper that I like must have been bought today.
5. VERBS WITH TWO OBJECTS
Some verbs have two objects. For example; Offer.
e.g. Kaoru offered Sharon the job in Japan.
(the two objects are Sharon and the job)
Therefore we have two possible ways of putting this phrase in the passive.
e.g. The job was offered to Sharon. (by Kaoru) e.g. Sharon was offered the job. (by Kaoru)
It is more usual for the passive sentence to begin with the person and we put the person who performs the action into the passive only if it adds information.
Other verbs like offer which can have two objects are:
ask tell give send show teach pay
Here are some examples of passive sentences with these verbs:
Kaoru gave me two hours to make my decision. (active)
I was given two hours to make my decision. (passive)
Two hours were given to me to make my decision. (passive)
Has anyone shown you the newspaper today? (active)
Have you been shown the newspaper today? (passive)
Has the newspaper been shown to you today? (passive)
6. STRUCTURE WITH INTRODUCTORY IT: IT IS SAID THAT…
These structures are rather formal but can also be used.
a) It is + passive + that + clause. (impersonal construction)
Many people believe the climate is changing.
e.g. It is believed that the climate is changing.
They claim Tokyo is one of the most expensive cities in the world.
b) Subject (person) + passive + to infinitive. (personal construction)
Many people believe the climate is changing.
e.g. It is believed to be changing.
They claim Tokyo is one of the most expensive cities in the world.
e.g. It is claimed to be one of the most expensive cities in the world.
Other verbs can be used in these passive patterns:
agree, allege, announce, assure, believe, consider, decide, discover, expect, feel, find, explain, hope, know, mention, propose, report, say, suggest, suppose, think, understand
Cooperation in Online Conversations: The Response Times as a Window Into the Cognition of Language Processing
Measuring the cognitive cost of interpreting the meaning of sentences in a conversation is a complex task, but it is also at the core of Sperber and Wilson’s Relevance Theory. In cognitive sciences, the delay between a stimulus and its response is often used as an approximation of the cognitive cost. We have noticed that such a tool had not yet been used to measure the cognitive cost of interpreting the meaning of sentences in a free-flowing and interactive conversation. The following experiment tests the ability to discriminate between sentences with a high cognitive cost and sentences with a low cognitive cost using the response time of the participants during an online conversation in a protocol inspired by the Turing Test. We have used violations of Grice’s Cooperative Principle to create conditions in which sentences with a high cognitive cost would be produced. We hypothesized that response times are directly correlated to the cognitive cost required to generate implicatures from a statement. Our results are coherent with the literature in the field and shed some new light on the effect of violations on the humanness of a conversational agent. We show that violations of the maxim of Relation had a particularly important impact on response times and the perceived humanness of a conversation partner. Violations of the first maxim of Quantity and the fourth maxim of Manner had a lesser impact, and only on male participants.
1. Introduction
While considerable progress has been made in the two fields of syntactic (see Socher et al., 2010; Chen and Manning, 2014, for examples) and semantic processing (see Berant and Liang, 2014; Pasupat and Liang, 2015, for examples), one aspect of natural conversations is often forgotten: pragmatic processing (see Jokinen and De Smedt, 2012; Jacquet et al., 2019, for reviews). Understanding the structure of an utterance and its semantic content is not enough to have a complete understanding of the utterance itself within its context. Indeed, there can be vast differences between what is said and what is meant in conversations between humans as Grice (1975), and later Sperber and Wilson (1995) noted.
On this distinction between what is said and what is meant, Grice (1975) introduced the Cooperative Principle along with its maxims to describe various expectations that allow conversation partners to infer the meaning of an utterance through the intention of its speaker. The Relevance Theory (Sperber and Wilson, 1995) later updated Grice’s original principles and offered a more in-depth, more unified explanation of the processes involved in inferring what is meant from what is said (and from what is not said).
While this makes for amusing conversational agents that are interesting to play with for a little while, this lack of relevance makes them unusable in more serious environments, where a lack of pragmatic reasoning cannot be hidden behind a shallow but seemingly witty reply. During a conversation with a customer service agent, for example, users will expect the agent to be helpful, not witty (Chakrabarti and Luger, 2015).
Despite the importance of pragmatic processing in how believable a conversational agent is, evaluating its quality within conversations is difficult to automate. Unlike syntactic processing, there are no well-defined rules that are stable enough to base it on, such as grammar. Unlike semantic processing, there are no direct and stable associations between certain speech features and implied meaning.
As a result, there is no gold standard to evaluate the quality of a conversational agent, and even less so to assess the quality of its pragmatic processing, even though many different evaluation methods exist for conversational agents in general (Paroubek et al., 2007; Hung et al., 2009; Ptaszynski et al., 2010; Chakrabarti and Luger, 2015; Meira and Canuto, 2015).
We argue that the Turing Test, already well known in computational sciences as a suggested method to test the (supposed) intelligence of a machine in textual conversations through a comparison with a human, can, in fact, be enough to detect flaws in pragmatic processing during such conversations when it is instead seen as a humanness testing environment. We also argue that the discriminating power of the Turing Test in this context can be made more specific, at the level of the utterance, if it is associated with a recording of the response times (RT) to indicate how difficult inferring the utterance’s meaning was.
To test this idea, we used three of Grice’s maxims commonly violated in conversations with artificial agents. These maxims are the first maxim of Quantity, the maxim of Relation and the fourth maxim of Manner, which all have been shown to have different effects on the humanness of a chatterbot (Saygin and Cicekli, 2002). We hypothesized that conversational agents violating the maxims would be identified as being more machine-like (Especially for violations of the maxim of Relation) than conversational agents producing more typical utterances.
We also hypothesized that the RT would increase following utterances requiring a high cognitive effort to infer their meaning. In particular, we expected that the responses of the participants would be delayed the most following violations of the maxim of Relation, a lower or absent increase following violations of the maxim of Quantity, and an intermediary increase following violations of the maxim of Manner.
2. Theoretical Background
2.1. Pragmatics of Conversation
Language is a code encrypting concepts into symbols that are the words of a sentence. However, this code does not contain all the information necessary to retrieve the complexity of the concepts it references. Indeed it only includes the information required to bring the mental representations of the conversational partner which is in the position of receiver slightly closer to those of the conversational partner in the position of the emitter. This operation gives the emitter the opportunity to save processing time (cognitive effort) and to only spend it on encoding information that they believe their partner does not already have. The details that are not important to reach the goal of the conversation can be left out, letting the receiver fill in the gaps with their knowledge, and in doing so avoiding the need to deal with redundant information. This balance between a cost (processing time) and effect (the modification of the receiver’s mental states) has been described in the Relevance Theory (Sperber and Wilson, 1995, 2015).
Relevance Theory considers relevance to be the result from the interaction between an utterance’s effect on the reader’s mental representations, and the processing cost that was required to infer its meaning, using contextual clues along with the utterance itself. The higher the processing cost, the lower the relevance, and the higher the cognitive effect, the higher the relevance of the utterance. As a consequence, optimal relevance is reached whenever an utterance gives a high cognitive effect for a low processing cost. It is also important to consider that, according to this theory, participants in a conversation will by default assume the relevance of an utterance, and that it is worth processing to recover its implied message.
In many cognitive tasks, the behavior of participants could be considered to be biased when compared to formal logic. Relevance Theory, by giving good predictions of these behaviors, instead revealed that people are not incoherent in the way they reason as they use contextual information in addition to the information explicitly provided to them, which makes it difficult to reason without a concrete context. The literature is quite abundant in this area and includes the logic of connectors (see for examples Politzer, 1986; Sperber et al., 1995; Noveck, 2001), Piaget’s inclusion task (see for example Masson et al., 2016b; Politzer, 2016), bias in probability judgment (see for examples Hilton, 1995; Baratgin and Noveck, 2000; Baratgin and Politzer, 2006, 2007, 2010), and decision making (see for examples Bless et al., 1998; Bagassi and Macchi, 2006; Masson et al., 2016, 2017a).
Relevance Theory was initially inspired by Grice (1975)’s work, before unifying it and expanding it. When trying to describe what conversations are about, Grice noted that:
“Our talk exchanges do not normally consist of a succession of disconnected remarks, and would not be rational if they did. They are characteristically, to some degree at least, cooperative efforts; and each participant recognizes in them, to some extent, a common purpose or set of purposes, or at least a mutually accepted direction. This purpose or direction may be fixed from the start (e.g., by an initial proposal of a question for discussion), or it may evolve during the exchange…But at each stage, some possible conversational moves would be excluded as conversationally unsuitable” (Grice, 1975, p. 45, emphasis in the original).
It is following this description that he proposed the Cooperative Principle, defined as:
“Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged” (Grice, 1975, p. 45).
In this principle, Grice explained that participants in a conversation have expectations regarding the shape and content of their partner’s utterances. He categorized them into four different maxims and their sub-maxims.
2.1.1. Quality
The first of these maxims is the maxim of Quality: “try to make your contribution one that is true.” It is subdivided into two sub-maxims: (1) “do not say what you believe to be false,” and (2) “do not say that for which you lack evidence.”
Grice considered the maxim of Quality to be the one on which all three others depend, and it can even be argued that the entire Cooperation Principle relies on the receiver of the utterances considering that the emitter believes that what is being said contains a valid information. Otherwise, the receiver would not even try to infer a hidden meaning. In consequence, while this maxim does not depend on the other three, the opposite cannot be said to be true, as all three heavily rely on the maxim of Quality being respected (Benton, 2016).
One can also argue that this maxim might heavily depend on the occidental interpretation of what a lie is, as it uses the term “try [to make your contribution one that is true],” yet in Mopan culture, for example, falsehood is not considered to be depending on the mental state of the emitter and whether they believe that their utterance is true or not: regardless it will be deemed to be a violation of conversational Quality if the information in and on itself can be considered to be false (Danziger, 2010).
It is also worth considering the fact that the sub-maxim of truthfulness (“do not say what you believe to be false”) is often violated (or at the very least suspended) in cases like metaphors and irony, while the maxim of Quality itself is not, since these sentences still do imply a true information despite what is being said being factually wrong. Thus, they cannot be considered to be actual lies, which are violations of the sub-maxim of truthfulness and of the maxim of Quality while making the interlocutor believe that they are in fact being respected (Wilson, 1995).
Despite its importance, it cannot be said that its violation produces an actual effect on the humanness of an interlocutor during a conversation (Saygin and Cicekli, 2002). In consequence, this maxim will not be the main focus of our study.
2.1.2. Quantity
Grice’s second maxim is the maxim of Quantity, which explains the expectations regarding the actual amount of information contained within an utterance and is defined by two sub-maxims: (1) “Make your contribution as informative as is required (for the current purposes of the exchange),” and (2) “do not make your contribution more informative than is required.”
The first sub-maxim of Quantity is quite often violated, and what its violation implies is usually simple to interpret. For example, it is very commonly used when the emitter of an utterance attempts to deceive the receiver to keep some pieces of information hidden to them (McCornack, 1992). It can also be used to imply that the emitter does not know the answer to a question, and potentially does not care about it either, for example:
1. Where does Marc live?
2. Somewhere on Earth.
In this case, 2 answers with an information that was already well known to 1. As a consequence, this utterance has a low relevance, since it has no cognitive effect, other than potentially changing the perception of the state of interest of 2 in the conversation. Though a violation of the sub-maxim does not necessarily imply a lack of interest. For example:
1. Where does Marc live?
2. Somewhere in France.
In this case, 2 likely does not give enough information (unless the context is specifically to be talking about countries), but this answer might simply be the result of not wishing to violate the maxim of Quality: not giving more information than what they know to be true.
Engelhardt et al. (2006) used an experimental protocol similar to the one used by Tanenhaus et al. (1995) consisting in showing pairs of participants some items to move, and where they can potentially be moved to. One needed to describe which spot each item should be moved to, while the other needed to do the corresponding action of moving the item accordingly. The study showed that participants preferred to over-describe the items to be moved (and in doing so violated the second sub-maxim of Quantity) while trying to avoid under-descriptions, which caused ambiguity. Violations of the maxim in the instruction they received had a direct effect on participants with the task of moving the item: the under-description visibly confused the participants, as was shown through the investigation of their ocular fixations. Confusion has also been observed when over-describing since participants needed more time to understand the movement that needed to be done. Similar results had already been published before by Spivey et al. (2002).
Horn (1984) interpreted the first maxim of Quantity to be the result of the principle of economy. Indeed, the speaker tends to only be giving the information that he or she must (“make your contribution necessary”), and no more. At the same time, they also tend to be giving as much information as possible that facilitates the listener’s task by improving the clarity of the meaning of the utterance (“make your contribution sufficient”), and no less. He argues, like Grice hinted at, that the second maxim of Quantity is more related to not giving irrelevant pieces of information, rather than actually giving as little information as possible. This is entirely coherent with the results of the previously described experiment, and more generally with Relevance Theory itself.
In the context of AI in conversations, the over-description is often seen by participants as a somewhat mechanical and artificial behavior, yet the lack of information, which can cause ambiguity and violates the first maxim of Quantity, is actually considered to be more human-like as it can be interpreted to be a sign of disinterest (Saygin and Cicekli, 2002).
According to the Relevance Theory, in the case of the first sub-maxim of Quantity, the emitter of the utterance gives little information, preventing the production of any cognitive effect on the mental representations of the partner. It is possible that they might try to infer another meaning though, which might require a certain cognitive cost, but we expect that it would not be very high, especially since a more likely explanation can be that this person is bored, rude, or not talkative (Saygin and Cicekli, 2002).
In our preliminary experiment, no significant detrimental effect of violations of the first sub-maxim of Quantity could be observed on the humanness of the conversation partner violating it, which was coherent with this observation. In this study, we will continue to be focusing on this sub-maxim as a reference for a low machine-like effect.
2.1.3. Relation
The third maxim of Grice’s Cooperative Principle is the maxim of Relation: “be relevant.” Grice explained it with the following example:
“I expect a partner’s contribution to be appropriate to immediate needs at each stage of the transaction; if I am mixing ingredients for a cake, I do not expect to be handed a good book, or even an oven cloth (though this might be an appropriate contribution at a later stage)” (Grice, 1975, p. 47).
It is in fact as an attempt to further study this maxim that Sperber and Wilson (1995) suggested the Relevance Theory as a more general explanation of Grice’s Cooperative Principle.
Because conversation partners will always assume by default that an utterance is somehow relevant, apparent violations of this maxim have a very noticeable effect. People will indeed believe that the emitter of such utterances is not comfortable with the topic and want to switch to another one. In the context of an artificial partner, this will usually be understood as a lack of comprehension of the sentence the violation was in response to, and as a consequence, it will be qualified to be very machine-like (Saygin and Cicekli, 2002).
In our preliminary experiment, violations of this maxim had a significant effect on the humanness of the emitter of such utterances and on the response times of the utterances following it, especially with female participants. We will continue to use violations of this maxim in our experiment as a reference for a strong machine-like effect.
2.1.4. Manner
Finally, Grice’s fourth maxim, the maxim of Manner, is defined through four sub-maxims: (1) “avoid obscurity of expression,” (2) “avoid ambiguity,” (3) “be brief (avoid unnecessary prolixity),” and (4) “be orderly.” The maxim of Manner refers to the structure of the utterance itself and on the inferences resulting from it. Among its sub-maxims we find “be orderly” which relates to the order of the pieces of information given during the utterance. It is not always clear which order should be considered the correct one though. It could very well be placing causes before consequences like in the following example:
1. They took a shower and went out.
2. They went out and took a shower.
In this case, it is clear that 1 and 2 do not have the exact same meaning. Yet the effect of order is not always as visible, as in the following case:
1. It is raining outside; she took her umbrella.
2. She took her umbrella; it is raining outside.
Blackmore and Carston (2005), in their paper on the and connector, suggested that keeping a chronological order is not necessarily required in some situations of causality, and that different orders might implicate, among other things, an aspect of surprise, as in their example:
1. Paul can’t spell and he is a linguist.
2. Paul is a linguist and he can’t spell.
In both cases, the second part following the and is the part triggering the effect of surprise, as in 1 Paul is not expected to be a linguist since he cannot spell, and in 2 he is not expected not to be able to spell since he is a linguist. Wilson and Sperber (2012) suggested that the meaning inferred from an utterance with structures similar to these ones is produced from a call to our knowledge on the probabilities of a causality between the different elements of the utterance.
In our experiment, some focus will be given to the fourth sub-maxim of Manner, which was not studied in our preliminary experiment.
2.2. Turing Test
Unfortunately, it is quite difficult to study pragmatic aspects of language in an ecological setting. Studies are indeed only qualitative since producing operational protocols for a quantitative study can be quite complex due to the inherent variability of conversations (Cohen, 1971; Alba Juez, 1995; Blackmore and Carston, 2005; Herring, 2013, for examples). We believe that the recent advances in the field of AI offer a new framework in which original methods can be designed to investigate the nature of conversations. Artificial conversational entities are indeed quite common now making them potentially useful tools to study human behavior and, specifically for our study, participants can now readily expect having to interact with them.
The Turing Test (Turing, 1950) is one of such protocols involving an AI. It consists in having participants acting as judges, attempting to find the machine between two interlocutors. Since the task essentially revolves around the participant’s ability to compare two interlocutors within an interactive textual conversation, it can be used to explore the conversational features that might be expected when conversing with a human. Because of this, we suggest that involving an actual AI can become superfluous in this context, as participants only need to believe that they will be conversing with one.
Turing (1950) described the idea of a test able to answer the question “Can machines think?” In his paper, he suggested a test analogous to a gender guessing game, where two persons (A and B) of different gender try to convince the participant C that they are the woman while the participant knows that only one woman is present. C engages in a textual conversation with each person, before guessing which of the two character he or she has been talking to is indeed the woman between A and B. Turing suggested replacing A with a machine.
It is worth noting that the Turing Test was initially assumed to be testing the intelligence of machines and on this aspect received many critics. Such critics do not apply to the context of our study since the intelligence of the conversational partner is not what is of interest here. What is essential is the Test’s usefulness in assessing the humanness of this conversational partner’s behavior. Some unclear aspects of the initial Turing Test still need to be debated though. In particular, whether the participant should or should not be aware of the presence of an AI within the test, as the interpretation of non-human-like behavior could depend on this.
Saygin and Cicekli (2002) indeed showed that participants asked to elaborate on their subjective feelings toward the productions of an AI (without giving them the information that one was in fact present) have a tendency to identify it as an odd behavior, but still human: “Are some of these people mentally ill?” (Saygin and Cicekli, 2002, p. 250). In their experiment, a group of participants was told, in a first phase, to answer a questionnaire on the violation of Grice’s Maxims, then was given another questionnaire on the artificial behavior of one of the interlocutors within a conversation. The other group was given the same task in the opposite order. In the first group, the answers to the second questionnaire were much more radical than in the second group, indicating that having an understanding of the maxims helps to determine the humanness of a conversational agent’s behavior.
Yet the violation of some maxims seems to have a positive influence on the humanness of an interlocutor for the participants:
“Sometimes maxim violations can create a human-like effect. In fact, strong violations of [Manner] have invariably created favorable impressions. It can be inferred that, had the programs that used being rude or obscure as a strategy been more successfully designed to handle the syntactic components of natural language, they would have appeared quite close to human beings, albeit strange ones. If in addition to this, the semantic processing had included ways to partially handle relevance and quantity, some of these might even have passed the Loebner Test” (Saygin and Cicekli, 2002, p. 254).
If today no artificial agent has managed to pass the Turing Test, some present interesting features. These chatterbots do not necessarily require to be able to learn on their own, especially the oldest of them. They indeed often use keywords in their interlocutor’s utterances to generate an answer. The most well-known examples of such programs are ELIZA (Weizenbaum, 1966), A.L.I.C.E. (Wallace, 2009), and more recently Zo.
Because the chatterbots available are not specific on which maxims they violate, we will avoid using an AI in our experiment in order to clearly dissociate different kinds of violations, and also on account of the potential impact of different elements like vocabulary and grammar. We will still introduce one interlocutor as being an AI, as it is necessary to make participants reflect on what they expect of a human compared to what they expect of an AI.
2.3. Response Times
It is common in experimental psychology to record the delay between a stimulus and a response from participants. However, to our knowledge, studies investigating online conversations using them remain very rare (Jacquet et al., 2018).
The concept of using such measures to develop interpretations of the inner workings of the mind is not new. It is based on the general assumption that the human brain is not unlimited in its processing speed as the communication between neurons is not immediate. Indeed, almost the totality of the synapses in the central nervous system use a chemical release of neurotransmitters which can individually take half a millisecond per connection, but varies depending on the type of synapses and other factors (Katz and Miledi, 1965).
The delay between the stimulus and its response is usually called reaction time, and measured in milliseconds (Deary and Der, 2005). In our case, since we do not record the time between a stimulus and an action, but between a stimulus and a written response, we chose to be using the term response time (RT) to avoid ambiguity, and we expect delays in the order of seconds.
We do not claim that neither measuring the reaction times nor the response times is the ideal portrayal of what is really occurring within the brain while processing sentences, nor that it is an ideal measure in and on itself, and in consequence over-interpreting the absolute values of these measures must be avoided. Still, they remain a very ecological tool as they do not require any dedicated recording device and can be used for data with high variability, unlike more precise timing measures like electroencephalography. They are often sufficient to demonstrate the impact of various factors on information processing (Fitts, 1966; Lachman et al., 1974; Thorpe et al., 1996; Bowyer et al., 2009, for examples).
Another critical factor to consider is that the reaction times can vary with the age of the participants in many tasks. As a consequence, considering the age as a potential factor should not be ignored when measuring reaction times, in particular with participants under 15 years old (Hale, 1990; Deary and Der, 2005). We should take similar considerations for RT measures.
In this experiment, we consider the response times between an interlocutor and the participant’s utterances to be an indicator of the cognitive cost of processing the interlocutor’s utterance. Data will only be collected with participants above 18 years old to avoid potential RT related biases.
3. Experiment
3.1. Materials and Methods
This experiment follows a strictly similar protocol as the one introduced in a preliminary experiment, with the addition of the condition of the violations of the fourth maxim of Manner, and with additional participants in all conditions (Jacquet et al., 2018).
3.1.1. Participants
Eighty six native English-speakers familiar with textual conversations through messenger softwares (Skype, Telegram, Messenger or others) agreed to participate.
Most of them lived in North America (48), and in Europe (31). Three lived in Australia, two in Africa and two in Central America. They were recruited thanks to the help of contacts on the different continents. These contacts had to find one or more voluntary persons, of different gender whenever possible.
All participants were adults between 18 and 45 years old (M = 25, SD = 5.7). 46 of them were males, while 40 were females.
Participants had different backgrounds to avoid generating too much of a bias that could come from specialized professions or academic backgrounds. Results of our questionnaire on this question revealed very varied backgrounds in the general fields of Arts, Sciences, and Services. 24 did not answer this question.
3.1.2. Variables
3.1.2.1. Factor—maxim violations
The main discriminating factor between our conditions was the type of the Gricean maxim that was violated (the first maxim of Quantity, the maxim of Relation and the fourth maxim of Manner) by the experimenter (referred to as the actor) during the conversations with each participant. The conversation order (AI actor first or Human actor first), and the gender of the participants were also considered to be potential factors and were controlled.
3.1.2.2. Main variable—response times
Our main dependent variable was the delay in seconds (the response time) between the moment the experimenter sent an utterance and the moment the participants sent their reply.
Since the length of the messages could potentially influence the response time, we designed a mathematical correction that was applied to all the recorded response times to remove the amount of delay that was likely caused by the number of characters in the sentences. We used a multiple linear regression model (with interaction) between the length of the experimenter’s utterance and the length of the participant’s reply on the observed delays. The model was inferred from the discussions between the participant and the human actor (which had no intentional violations of Grice’s maxims) to create predictions of the delay as it should be without violations.
This model allowed us to calculate a theoretical delay (D) for each of the participants’ responses.
where Ce is the length of the experimenter’s utterance, and Cp the length of the participant’s utterance (both in number of characters). w, x, y and z are the coefficients of the model.
This theoretical delay was then removed from the observed delay. The resulting difference was then used to test our hypotheses and represented the deviation from normal (human-like) response times.
where d is the observed delay.
3.1.2.3. Secondary variable—identification percentage
The percentage of correct identification of the AI actor was also recorded to be compared to 50% (Random Chance). A value above random chance meant that the actor was perceived to be machine-like. A value around a random chance indicated that the actor was recognized to be human-like (Participants were unable to distinguish the two actors correctly).
3.1.2.4. Control variables
Some other variables were recorded to control any potential bias. These control variables included the participants’ gender, their age, the duration of each conversations (in seconds), the self-evaluated knowledge about AI (a Likert scale from 1 to 7) and self-evaluated knowledge about computer science (Likert scale from 1 to 7), and finally the confidence in their guess in the Turing Test (Likert scale from 1 to 7).
3.1.3. Procedure
Since the experiment was entirely online, participants did not need more than a computer with an internet connection to be able to participate and could do so at home. To avoid adding bias to the recording of the response times, mobile devices could not be used to participate in this study, since typing speed can be influenced by the kind of device used. The chat where the experiment happened was hosted on a private French server and had been created for this study. Participants were required to communicate on Likert scales their knowledge in AI and Computer Science, their field of study or career, their gender and their age. This information was only sent to the server if consent was explicitly given for participating in the experiment.
Once in the ChatBox, the experimenter (displayed as “Moderator”) explained in detail the rules participants had to respect during the conversations as well as their task.
All conversations could only last up to 15 min, during which the participants could decide to stop the conversation if they had guessed which actor (either AI or human) they were talking to.
Both interlocutors (AI and human) were displayed with the same name (Andrew), and both tried to portray the same fictional character. This was done to invite the participants to ask questions to the fictional character instead of questioning the actors themselves. This meant that both actors would provide the same semantic information to the participant (since the character they’re portraying is the same), but in different ways.
As we have mentioned before, there was actually no AI in our study. The two actors were, in fact, the same person (a male student in experimental psychology). The two roles only differed in their behavior regarding violations of the maxims. The experimenter in the human role had to behave like a “normal” human, without voluntarily adding violations. In the AI role, the experimenter was constrained in his behavior, and was not able to answer normally, but instead had to follow guidelines designed to produce as many violations of the required kind for the condition as possible, and in consequence to change the feeling of humanness given to the participant, since we expected this constrained behavior to feel closer to that of an AI. For each sentence, the actor indicated if a voluntary violation had been introduced because of these constraints.
The choice not to use an actual artificial intelligence during this test was motivated by the desire to restrict the differences between our conditions to the ones we could control. Vocabulary differences, syntax issues, or a defective understanding of the participant’s utterances could have added further violations during the conversations that would have been difficult to predict in our protocol. It is also for the same reason we kept the same experimenter for the two roles.
As the Turing Test is about free-flowing conversations, it was impossible to keep them strictly identical between conditions, and even between participants within the same condition. To avoid strong biases, any new information given about the character of Andrew that had not been anticipated was written down so that the same information could be used again in different conversations. Participants sometimes reported being a bit surprised to see the same information between the two conversations.
Participants were able to discuss with both interlocutors (in random order), and each participant was assigned to one of the three conditions randomly (see the conditions below). The use of smileys, hypertexts, and double-posts was not allowed. Utterances (both from participants and actors) could not contain more than 255 characters.
3.1.4. Conditions
A summary of the conditions and their expected effects is available in Table 1.