Siri and the skills of encoding personal meanings in the context of English speech etiquette

Zhikulina Christina Petrovna ORCID: 0000-0003-2488-4616 Postgraduate, Department of General and Russian Linguistics, Peoples' Friendship University of Russia named after Patrice Lumumba 117198, Russia, Moscow region, Moscow, Miklukho-Maklaya str., 6, department, Miklukho-Maklaya, 12 christina.zhikulina@gmail.com Other publications by this author

1. Introduction

Currently, the use of artificial intelligence (AI) is increasingly being investigated from the linguistic aspects, namely language analysis and machine adaptation to natural speech. Scientists hope that the results obtained in this area can help in the development of machine translation tools and methods of teaching a machine language. Some language technologies are already widely used in voice assistants, text editors, and online translators; AI is able to independently come up with texts on various topics, write essays, essays, and term papers. Voice assistants or assistants are popular to use, as they cover different areas of social life (banks, government portals, university websites) and private life (smart home, smart watches, smart speakers, laptops and smartphones).

In the last decade, scientists' interest in artificial intelligence has grown dramatically. The leading trend in the digital world has become the idea of developing an artificial mind capable of independently explaining to a person how he came to certain decisions. David Gunning, head of the artificial intelligence development program in defense research, believes that "any high-performance system can be one of the most difficult to explain" ^{[16, p. 10]}. His theory is that machines are capable of generating "complex models of the real world that man himself has not yet fully learned" ^{[16, p. 12]}, therefore, the ability of artificial intelligence to express itself is often criticized and questioned.

It should be noted that the difficulties associated with the reproduction of mental functions as in humans, but on an electronic computer (computer), have not been eliminated since 1950. In his dissertation "Philosophical problems of the correlation of human thinking and computer capabilities in problem solving processes", philosopher Yu.V. Orpheev questions the possibility of solving this problem only within the framework of cybernetics and computer science ^{[12, p. 7]}, therefore there is a need to analyze the capabilities of artificial intelligence in various scientific fields, including the humanities ^{[13, p. 7]}. 3].

The relevance of the research lies in the fact that at present, public and scientific interest in the possibilities of voice systems in the field of natural speech imitation has grown significantly. Voice assistants act as interlocutors, assistants in various spheres of public life. Users' opinions are divided. Some pay attention to the speech formulas used by artificial intelligence, notice such advantages as speech creation, literacy, speech recognition functions, "the growth of artificial intelligence" ^[14]. Others criticize: Russian Russian voice systems – "arrogant and boorish tone of communication", "a waste of time", "... makes mistakes in his answers", "there is nothing to chat about" ^[14]; 2) foreign voice systems – "misunderstanding of Russian speech", "to communicate with the column you need to have a good pronunciation in English in English", "useless" ^[15]. In this regard, over the past five years, many works have been written analyzing language capabilities and evaluating artificial intelligence skills, as well as searching for various ways to integrate AI into educational, translation, writing and other processes. For example, in linguistics and literary studies, studies have appeared aimed at determining the "communicative potential of Russian-speaking and English-speaking voice systems" ^{[8, p. 40]}; "the role of artificial intelligence in learning foreign languages" ^{[9, p. 39]}; the functions of artificial intelligence in the role of "author and co-author of a literary work" ^{[11, p. 109]}; similarities and differences between "artificial " intelligence" and the human mind" ^{[4, p. 746]} and others. It is believed that the era of artificial intelligence began in 2022, when the first chatbot (ChatGPT) was created, using deep learning to generate text and answers to questions. In the article "Weights and Wings of artificial Intelligence" of the Moskovsky Komsomolets newspaper, AI experts emphasize that "ChatGPT is just one of the programs – a vivid phenomenon thanks to which a lot of people have learned what generative language models are, transformers that someone teaches, and they solve a wide range of intellectual tasks" ^{[3, P. 6]}. All these factors determine the relevance of studying language capabilities for natural speech recognition, processing user requests and encoding personal meanings in human communication formulas from voice assistants, since recently they have been working using built-in AI algorithms. Research in this area will help determine the prospects for the introduction of AI not only in private businesses, but also in educational processes, public administration, etc. In the future, the need to minimize errors and expand the vocabulary, stable phrases and expressions in the speech of voice systems will increase rapidly due to the expansion of the use of voice assistants. 

The purpose of the study identified the following tasks:

1. Conduct a linguistic experiment to be able to describe the skills of encoding personal meanings by the Siri voice assistant using our selected questions and greeting questions on the topic "How are you?".

2. To compare stable communication formulas in English speech etiquette with spontaneously generated responses from the Siri voice assistant to questions or greeting questions.

3. Determine the correctness of the responses issued by the Siri voice assistant and deviations from the norms in English speech etiquette.

4.      Analyze the understanding of the personal meanings embedded by the user in the Siri voice assistant using contextual analysis.

5.     To identify the limits of the linguistic capabilities of the voice system in machine-human communication.

The research material was 10 responses in English from Apple's Siri voice assistant to 9 questions grouped according to speech etiquette norms in an English-speaking environment. For comparative aspects of speech etiquette norms and contextual analysis, examples from the Reverso online translator with a contextual dictionary function were used; questions and welcome questions for the experiment and its descriptions were selected from articles from popular online platforms for learning English: Puzzle English and Skyeng.

The theoretical basis of the research was the works of domestic and foreign scientists in the field of artificial intelligence and its development (Pikover, 2021); text-to-speech conversion systems (Taulli, 2021); artificial intelligence as a "product of technological progress" (Maximov, 2021); philosophical problems of the relationship between human thinking and computer capabilities (Orpheev, 1974); a machine for adapting to natural speech (Orpheev, 1978); problems of "decoding context during a speech act" (Efremova, 2021); systemic linguistics, where the word is a "discrete signal of syncretic thought" (Valentinova, Denisenko, Preobrazhensky, Rybakov, 2016); norms of English speech etiquette (Galiberdova, 2019); conversations on everyday topics and problems of understanding (Tracy, Robles, 2015); features of secular communication (Sternin, 1996); connections of mental operations and functions of consciousness (Leontiev, 1975).

The practical significance of the study lies in the fact that with the help of the obtained material, it will be possible to determine and analyze in the future the stage of progress or regression of voice assistants in such areas as imitation of natural speech and skills of conducting spontaneous dialogue in AI. The research material can also be used in university practice when developing a course on the program "Applied Digital Philology" aimed at training specialists using IT technologies and humanitarian complexes together. 

2. The main part

2.1 Problems of understanding personal meanings in human-human and machine-human communications

As you know, one of the leading problems in communication is understanding. Words and their meanings in people's minds are always "subjected to unconscious processes of analysis, synthesis and comparison during activity, while interacting with the products of previously perceived impressions" ^{[19, p. 10]}. It follows that speech activity and understanding can be "linked to the psychological experience and characteristics of an individual" ^{[19, p. 14]}. As can be seen, even in human-to-human communication, questions arise related to personal meaning, that is, with "subjective coloring, which expresses the meaning of what is reflected for the subject himself" ^{[10, p. 74]}. As a rule, misinterpretation of personal meaning leads to communication failures. Therefore, a similar problem of understanding exists in machine-human communication, since it depends not only on the ability of the voice assistant to process text, but also on the ability to take into account the context.

There are a number of researchers who believe that artificial intelligence will soon be able to approach human capabilities. For example, Ray Kurzweil, director of engineering at Google and the leading inventor of text recognition and speech synthesis systems, suggested that by 2045 there would be a "computer singularity, a world of hybrid people: partly people, partly machines" ^{[18, p. 266]}. The scientist has been publishing his forecasts since 1990, many of them have already become academic.

However, there are also those who believe that artificial intelligence is just a technological process that does not include speech–making potential. An expert from the Russian State Academy of Intellectual Property holds the following opinion: "Currently existing artificial intelligence systems belong to the category of "weak". They are designed to perform clearly regulated tasks and do not allow us to assert the presence of a creative core in the ongoing intra-system processes, even despite the unpredictability of the result" ^{[11, p. 109]}.

It is also important to take into account that the "weak" development of AI and, as a result, descriptions and research in the field of machine-human communication can only be reduced to analyzing a machine-generated response to a human request or question. It is important to emphasize that the Siri voice assistant we are investigating does not respond to intonation in spoken speech, therefore, the perception of intonation accents and accents has not been tested in this work.

2.2 Description of the linguistic experiment

To conduct the linguistic experiment, equipment was used, including an Apple HomePod smart speaker and a tablet computer with built-in Siri voice assistants. The voice systems were asked specially selected questions in English. It should be clarified that the use of this assistant is available in 36 countries in 20 languages. However, Siri settings have the most options using English language: English (Australia), English (Canada), English (India), English (Ireland), English (New Zealand), English (Singapore), English (South Africa), English (United Kingdom), English (United States). The choice of language for the linguistic experiment was determined by the quantitative indicator of modifications with English in the Siri configuration. We tested the English (United Kingdom) variant – classic English, or the British version of English.

In the experiment, the same question is asked to the voice assistant until he begins to repeat the answers repeatedly. During the linguistic experiment, 10 unique (non-repeating) responses were obtained.

         The question groups for Siri were not chosen by chance. In English society, at the beginning of a conversation, it is customary to ask the interlocutors "How are you?". In most cases, this question is informal. For example, Americans have a widespread culture of small talk, which in English means light conversation, empty conversation or small talk. This culture implies the ability to start and maintain dialogues on casual topics.

One of the formulations of the concept of secular communication belongs to the Russian linguist I.A. Sternin, who interprets it as follows: "This is a mutually pleasant, non-formally binding conversation on general topics, the main purpose of which is to spend time with the interlocutor, remaining in verbal contact with him" ^{[17, p. 4]}. In our linguistic experiment, verbal contact is also of great importance, since it is necessary to answer the question: is contact between a machine and a person established using speech means – words? According to Sergey Markov, head of the Experimental machine Learning Systems department of the General Services division of one of the banks, "the basis of artificial intelligence modeling is statistical linguistics, which appeared as the idea that the meaning of a word is related to statistics and context" ^{[3, p. 6]}.

It should be added that in the speech etiquette of the English language, the system of speech formulas in communication is stable. Depending on the situation, the interlocutor chooses constructions for a question or a greeting question: in formal and informal conversation, in a neutral setting, in conversations with acquaintances and strangers ^{[5, p. 20]}. Let's look at the selected questions or greeting questions like "How are you?" grouped for a linguistic experiment: 

(1) "What's up?" / "How is your life?" – questions for people in friendly relationships;

(2) "How are things?" / "How goes it?" – questions for informal communication;

(3) "How are you?" / "How do you do?" – formal questions;

(4) "How are you doing?" is a question of politeness;

(5) «How have you been?» – the question is, if you haven't seen each other for a long time;

(6) "How are you holding up?" is a question if something unpleasant has happened to a person.

So, each group contains a situational context, including precedent situations, behavioral scenarios, and cultural characteristics. It is assumed that such a classification will help to identify the ability to perceive personal meanings in a machine.

2.3 Analysis of the results of the linguistic experiment

Let's ask questions to the Siri voice assistant and analyze: whether it has reactions to different types of questions and greeting questions; whether the machine takes into account the boundaries of formal and informal communication, friendly and business relationships; whether the machine has an understanding of the context during speech contact with a person.

Fig.1. Siri voice assistant responses to group (1) of "How are you?" questions for people,

who are in friendly relations

The Siri voice assistant gives ambiguous answers to questions for people in friendly relationships: Hello, how can I help? (Hello. How can I help you?) (hereinafter translated by the author of the article. – K. J.) and I'm not sure I understand (I'm not sure I understand) (Fig. 1). From the results obtained, it can be seen that Siri does not recognize the interlocutor's interest or the person's disposition to a friendly conversation. For example, the question How is your life? (How are you? / How are you?) it remains misunderstood by the system, and all the responses of the voice assistant come down to I'm not sure I understand (I'm not sure I understand). Also, if Siri doesn't understand the question or can't answer it, she adds How can I help to her remark? (How can I help?). It is worth checking whether this function is implemented in the following groups of questions and whether this is a pattern.

Fig.2. Siri voice assistant responses to a group (2) of "How are you?" questions for informal communication

Informal communication formulas turn out to be more familiar to Siri and to the questions "How are things?" (How are you?) and "How goes it?" (How are you? / How is it going?) the voice assistant gives four variants of responses (Fig. 2). Three of which (1-3) correspond to the norms of speech etiquette in the English language – similar speech formulas can be found in any English textbook for beginners ^[7]. However, the answer to the interesting question is not often found in the stable speech etiquette phrases of the English–speaking society and is a more commonly used form in communication at conferences, meetings, discussions, etc. - business English. Let's look at some examples with an interesting question from the Reverso online translator with a contextual dictionary function:

(1) "I think you've raised an interesting question" (I think you've raised an interesting question) ^[20];

(2) "President Obama: No, it's an interesting question" (President Obama:No, that's a great question) ^[20];

(3) "It is indeed an interesting question, and the short answer too it would be: freedom" (This is a truly interesting question, and the short answer to it will be: freedom) ^[20];

(4) "The interesting question, then, is not about cognitive dissonance but about faith" (An interesting question is not about cognitive dissonance, but about faith) ^[20].

In the examples given, you can see the answer of former US President Barack Obama (2), that is, a politician; an example of using an interesting question in discussions (3) or scientific debates (4); as well as a more universal form of response used in both formal and informal settings (1). In in our experiment, the Siri voice assistant, as a response, gives only an interesting question without explanation or additional meanings – not as shown by the use of an interesting question examples from the Reverso contextual dictionary. Let's ask Siri a clarifying question: "What do you understand when you say an interesting question?" (What do you understand when you say an interesting question?). The voice assistant's response is: Hmm... I don't have an answer for that. Is there something else I can help with? (Hmm... I don't have an answer to that. Is there anything else I can help with?). Here it is worth paying attention not only to the fact that this response from Siri is not tied to any specific speech situation, but also to the fact that the machine's ability to explain to a person how it came to certain decisions (the content of the semantic component of its response) is missing.

Fig.3. Siri voice assistant responses to a group (3) of formal questions "How are you?"

Siri has only one answer to formal questions – likewise (Fig. 3). In dialogues, likewise is often translated into Russian as mutually. In English, likewise is an introductory word or a combination word that performs the function of comparing or supplementing information in sentences. For example, "The meet was delicious. Probably, the dessert was excellent too" (The meat was very tasty. Moreover, the dessert was also excellent) ^[1]. In the etiquette forms of English, likewise is not frequent, and is more suitable for a group of questions (1) of an informal setting and friendly communication (Fig. 1). Moreover, in the Reverso online translator with the function of a contextual dictionary, likewise can be seen more as a sporadic and composite answer than a frequency and isolated one:

(1) "You could do likewise if you wish" (You can do the same if you want) ^[20];

(2) "And likely for scientists and technologists" (And the same for scientists and technologists) ^[20];

(3) "You can probably be like him" (You can also become like him) ^[20].

It follows from the above that the Siri voice assistant distinguishes between the areas of formality and informality in communication. However, the responses appear unpredictable and atypical for English speech etiquette in everyday conversations – likewise.

Fig.4. Siri voice assistant responses to the courtesy question "How are you?",

group (4) and the question of the group (5), if you haven't seen each other for a long time

The questions "How are you doing?" (How are you?) and "How have you been?" (How are you? / How are you?) They turned out to be the most understandable for Siri (Fig. 4). They received the largest number of responses. You should immediately pay attention to the difference between the situational question "How have you been?" (How are you? / How are you?) and the question of politeness "How are you doing?" (How are you?) the voice assistant does not see, so it gives the same answers in both cases.

In the research of the article, we raised the question of establishing verbal contact between a machine and a person using speech means – words. According to M.A. Rybakov: "A word is a discrete signal of syncretic thought. <...> a word, unlike a thought, has the property of separateness. As a sign, it is distinguishable from the flow of speech, whereas its meaning conveys a nonlinear, multidimensional, non-separable thought into separate tangible segments. A word is a compact quantum of information; a sign capable of obtaining additional meaning when changing its shape or even without it" ^{[2, p. 24]}. If we take the word as a basis as a discrete signal of syncretic thought, then verbal contact between a machine and a person is established within the boundaries of primitive thinking, in which representations of different types are undifferentiated with each other. In rare cases, the Siri voice assistant captures thoughts voiced by a person. The responses do not always correspond to the speech situation, but by fixing a single word from the speech stream, the machine tries to select the context within the language corpus available to it.

Phrases and expressions are marked in different colors (Fig. 4), which are repetitions of the responses in the groups of questions described earlier (Fig. 2). In addition to repeated answers, there are new options: I'm happy to be here (Happy to be here) and I'm pretty good, thanks (Pretty good, thank you). Here, the voice assistant contains stable speech formulas of English etiquette, and they are issued in compliance with speech norms in conversation with English-speaking users.

Let's turn to a more interesting response for us – Hey, I can't complain. Thanks for asking (I'm not complaining, thanks for asking). We can say that artificial intelligence seems to "hint" that it can't have any problems, so it has nothing to complain about. Although the expression I can't complain (I'm not complaining / I have nothing to complain about) is included in the list of stable phrases and expressions in the English language, most often it is used in situations where a person indicates his conscious position in some matter or issue. Let's look at some examples with I can't complain (I'm not complaining / I have nothing to complain about) from the Reverso online translator with a contextual dictionary function:

(1) "I can't complain at my age" (I'm not complaining about my age) ^[20];

(2) "Things are going well and I can't complain" (Everything is going well, I have nothing to complain about) ^[20];

(3) "Thanks to football I have had success in my life and I can't complain" (Thanks to football I became successful in life and I have nothing to complain about) ^[20];

(4) "The end result however turned out pretty good so I can't complain" (But I was eventually satisfied with the result, so I'm not complaining) ^[20].

In the examples considered, it can be seen that people whose statements contain I can't complain (I'm not complaining / I have nothing to complain about), refer to different situations in life: age, career or results. They emphasize that there is nothing to complain about in the current state of affairs. However, in most examples involving a person, there is an explanation – why he does not complain, that is, it should be assumed that if circumstances change, he will complain. This leads us to the idea that Siri does not issue an I can't complain response (I'm not complaining / I have nothing to complain about) additional information in order to emphasize my artificial origin and the absence of any complaints, as well as the lack of physical ability to take part in actions similar to human ones. The response of the voice assistant I can't complain (I'm not complaining / I have nothing to complain about) without explanation indicates that he will always have nothing to complain about.

Thus, speech interaction with a person in the Siri voice assistant is also manifested at the level of one-sided use of context. Unilateral use refers to the transfer of information from a machine to a person, but not vice versa. In response, I can't complain (I'm not complaining / I have nothing to complain about) (Fig. 4) Siri refers to its artificial origin and lack of emotional or physical perception of reality, as a result of which it cannot complain about anything. At this stage of the study, it can be concluded that contextual meanings and precedent texts can be stored in the computer response generation function, since they are likely to be decrypted by the user. However, the content of the context and personal meanings in the user's questions, so far, more often leads to communication failures.

Fig.5. Siri voice assistant responses to the question "How are you?" of the group (6), implying that something has happened to the person

We see that the Siri voice assistant does not recognize the question of the group (6), which implies that the recipient has had some unpleasant events in his life, and the communicator wants to get an answer through a special situational question (Fig. 5). It can be concluded that Siri's capabilities lack the functions of recognizing the personal meaning of the communicator. Since the voice assistant is a machine, not a person, and unpleasant events in life cannot occur in a machine, this category of meanings is not preset in Siri's communication capabilities and response generation. This observation brings us back to the retort I can't complain (I'm not complaining / I have nothing to complain about) (Fig. 4), where a car cannot have complaints about something that will never happen to it. The voice assistant answers the question of group (6) in exactly the same way as shown in the group of questions (1) for people who are in friendly relations (Fig. 1): I'm not sure I understand (I'm not sure I understand).

As E.N. Efremova writes in her work: "In fact, the success of communication depends on how the listener or reader interprets the speech code" ^{[6, p. 2]}. In our experiment, the listener is the Siri voice assistant, so we can conclude that communication success does not occur in such groups of questions or greeting questions that contain personal meanings, since the machine cannot interpret a person's speech code in certain (non-standard) situations.

Remark How can I help (How can I help?) (Fig. 1) is natural if the question was incomprehensible to the voice system. From the point of view of understanding, Siri is at a "weak" level of development, but from the point of view of compliance with the norms of English speech etiquette, it has some potential to continue the dialogue, since during the experiment, responses containing complementary or clarifying questions were identified, intended for further verbal interaction between the user and the voice assistant. 

3. Conclusion

         Thus, we come to the following conclusions. During the linguistic experiment we conducted, we found that the Siri voice assistant practically lacks the skill of encoding personal meanings embedded by a person in questions or greeting questions within the framework of the topic "How are you?" in English. We see a partial form of reflection on the part of the machine when communicating with a person in this limit.

         When comparing stable communication formulas in English-speaking society with spontaneously reproduced Siri responses, we found three uniquely generated responses that differ from the patterns of English speech etiquette. In other cases, the norms were partially observed, but the responses did not always correspond to the situational context of the questions or greeting questions. The machine did not recognize the meanings in which it is necessary to take into account the duration of communication or the presence of unpleasant events in the user's life, since Siri lacks emotional and physical perception of reality, as well as fixation of previously perceived impressions.

         Contextual analysis has shown that the success of communication between Siri and the user occurs only within the framework of the basic formulas of communication in English, which are implemented in formal and informal questions, questions of politeness. It was also revealed that the understanding of personal meanings and contexts inherent in a person is not realized, but there are cases of unilateral use of context, when information is transmitted on the principle from machine to person.

The essence of the above boils down to the fact that the difficulties associated with the reproduction and imitation of mental functions as in humans, but on an electronic computer (computer), still remain unsolvable in the Siri voice assistant for several reasons: 1) the machine is not capable of an unconscious data analysis process; 2) the Siri voice assistant does not have the function of remembering previously perceived impressions, and impressions in general.

In order to minimize the existing problems with encoding personal meanings inherent in a human voice assistant, it is necessary: 1) expand the body of texts in Siri beyond the basic answers; 2) distribute Siri's responses according to the meaning of each topic separately. This method is slow, but it can have a positive result in machine–human communication; 3) create an independent case for Siri, into which the voice assistant will be able to independently record all incoming replicas and questions from a person and, based on this data, generate answers that include previously perceived impressions, that is, create an artificial container of recorded impressions. With the implementation of the last point, the probability of individualization of each individual voice assistant will increase, focusing on the personal qualities and characteristics inherent in their users.

As a prospect for further research of the stated problems, it is possible to conduct similar linguistic experiments with different types of voice assistants and describe the progress and updates of the dialogue function, compare the speech formulas of different voice systems, evaluate the communicative progress in machine-human interaction within the framework of synchronic and diachronic approaches.