1. Introduction

The accelerated development of artificial intelligence in recent years has fundamentally transformed the field of interpreting and translation. While until recently professional discussion focused primarily on computer-assisted translation and interpreting as support tools for human work, today the question of their complete replacement by automated systems is increasingly being raised. Moreover, this debate concerns not only conference or community interpreting and translation but is increasingly affecting the field of legal interpreting and translation as well.

Legal interpreters and translators represent a specific group of professionals whose activity is directly connected with the protection of fundamental rights of participants in judicial and administrative proceedings, particularly the right to a fair trial, the right to defence, and the right to be heard. The quality of interpreting and translation in this context is not merely a matter of professional standards or procedural efficiency but has a direct constitutional law dimension. Poor quality or inaccurate interpreting or translation may lead to interference with the procedural rights of participants in proceedings and, in extreme cases, to an unjust decision.

The article therefore focuses on the issue of using automatic interpreting and translation in the judicial environment, at the intersection of administrative and constitutional law, linguistics, and translation studies. The first part devotes attention to the fundamental principles of legal interpreting and translation, and the current development of technologies based on artificial intelligence. The second part then concentrates on the specific legislative proposal contained in the amendment to the Act on the Residence of Foreign Nationals, which envisages the use of automated systems instead of legal interpreters and translators, and evaluates it from the perspective of fundamental rights protection and procedural guarantees.

2. The Role of Court Interpreting and Translation

Whereas interpreting is the oral transfer of a message from one language to another occurring in real time (i.e., interpreting the spoken word)^[1], translation is, in simplified terms, the written transfer of a message from one language to another (i.e., translation of the written word)^[2]. Both interpreting and translation serve to facilitate intercultural communication and overcome language barriers between people who do not speak the same language. Court interpreting and court translation are specific disciplines. They are performed in a “legal” setting (in court, at an administrative authority, at the police station, at a notary’s office). Many countries have decided to introduce legislation regulating the activities of court interpreters and make the performance of this activity conditional upon obtaining special authorization in order to ensure a certain quality of their performance^[3]. In the Czech Republic, too, court interpreting and translation are conditional upon obtaining authorization, pursuant to Act No. 354/2019 Coll., on Court Interpreters and Translators, which forms the basis of the legal framework for the activities of court interpreters and translators.

The right to interpretation is enshrined in many international instruments and is often linked to the right to a fair trial. The International Covenant on Civil and Political Rights enshrines it in Article 14(3)(f)^[4]. The European Convention on Human Rights provides for it in Article 5(2) and Article 6(3)(e)^[5]. Directive 2010/64/EU of the European Parliament and of the Council of 20 October 2010 on the right to interpretation and translation in criminal proceedings specifies the right of accused persons and suspects to interpretation and translation in criminal proceedings in EU Member States. In the Czech legal order, the right to an interpreter is enshrined not only at the constitutional level (in Article 37(4) of the Charter of Fundamental Rights and Freedoms), but also in procedural regulations.

The right to interpretation means that a person who does not understand or speak the language of the proceedings has the right to free assistance of a court interpreter and translator in judicial proceedings or proceedings before administrative authorities. It also applies to deaf and deaf-mute persons, who are provided with a sign language interpreter. The right to interpretation, its content and scope, is further defined and specified by the international instruments themselves or by Czech procedural regulations and case law.

The right to interpretation is considered part of the right to a fair trial, as a so-called ancillary right to the principle of equality of the parties in proceedings^[6]. The right to a fair trial is one of the fundamental pillars of the rule of law and democratic society^[7]. It ensures that every person may assert their rights through established procedures before an independent and impartial court and, in specified cases, before another body^[8]. According to the Constitutional Court of the Czech Republic, it is not a one-dimensional right, but rather a complex and structured right comprising multiple independent subjective fundamental rights^[9].

The right to an interpreter is an essential component of the principle of equality of the parties. William Hewitt emphasizes that «interpreters help ensure that such persons may enjoy equal access to justice, and that court proceedings and court support services function efficiently and effectively»^[10]. Ensuring equality of the parties in all types of proceedings is a necessary prerequisite for achieving a fair trial. In order to achieve true equality of the parties in proceedings, it is essential to ensure the right to use one’s mother tongue before public authorities, from which follows the right to an interpreter, which guarantees that no party will be disadvantaged due to a language barrier^[11]. The right to an interpreter (together with the right to defence) represents a significant guarantee of properly conducted and fair proceedings before public authorities^[12].

Failure to ensure the right to an interpreter, despite a party to the proceedings having requested it or where the need for a court interpreter is evident, results in a failure to ensure equality of the parties^[13] and a substantial procedural defect, as it significantly curtails their rights^[14]. Violation of the right to an interpreter impermissibly infringes upon human dignity under Article 1 of the Charter of Fundamental Rights and Freedoms and simultaneously violates the principles of a fair trial, thereby constituting a violation of the right to a fair trial^[15].

Without the ability to understand the ongoing proceedings and without the ability to express oneself adequately, a fair trial would remain a mere formality and would be practically unattainable for those who do not understand the language of the proceedings. This right deserves all the more attention in connection with the discussion on the possibilities of using technology and artificial intelligence in the field of interpreting and translation in judicial and administrative proceedings.

3. Machine Interpreting and Machine Translation

Automatic speech translation, also referred to as speech-to-speech translation or spoken language translation, represents an innovative field of artificial intelligence aimed at creating machines capable of translating spoken discourse from one language to another in real time and for immediate practical use. In recent years, this technology has experienced significant performance improvements due to the development of modern machine learning algorithms, particularly neural networks, and the availability of extensive language data (for certain language combinations). There are two fundamental approaches to machine interpreting: the cascade approach and the end-to-end approach^[16].

In the cascade approach, the process of oral translation is divided into several successive sub-steps: automatic speech recognition (ASR), which converts spoken discourse into text form; machine translation (MT), which translates written text from the source language into the target language; and text-to-speech synthesis (TTS), which generates the spoken form of the translated text^[17]. Cascade models outperform end-to-end models in accuracy, particularly for high-resource languages and complex tasks, due to their modularity and better error management^[18].

The end-to-end approach is based on the use of a unified component that performs direct conversion of the input audio signal into an output audio signal in the target language without generating intermediate textual representation, i.e., transcription^[19]. This approach enables the translation of spoken speech from one language directly into speech in another language without relying on intermediary text processing^[20]. End-to-end models excel in latency and deployment simplicity.

Machine interpreting systems are capable of facilitating spoken communication to a certain extent. They also achieve real-time processing of input and provide voice output^[21], but they still rely on machine translators developed primarily for text translation. These systems cannot fully analyse and interpret the content of messages, which is essential for proper mediation of their meaning. They are unable to process meanings conceptually in a manner that corresponds to human understanding, which limits their ability to accurately capture and convey the sense of communication, particularly in culturally specific and more complex situations^[22]. Machine interpreting systems are incapable of understanding typical aspects of verbal communication, such as inference from situational context, interpretation of prosodic features, correction of imperfect utterances, etc.^[23]. When processing and converting the source message, an interpreter utilizes not only linguistic skills but also their presence in the communicative situation and ability to adapt to the context of communication^[24]. This temporal and spatial immediacy of interaction, which encompasses not only the spoken word but also non-verbal expressions, makes human interpreting unique and irreplaceable by machines^[25].

MT is the automatic translation from one language to another using computer technology without human intervention^[26]. Currently, neural machine translation (NMT) is most frequently utilized, based on the use of deep neural networks and employing deep learning^[27]. The Transformer is a deep neural network composed of many encoder and decoder layers that utilize a self-attention mechanism for efficient analysis of relationships between all words in a sequence simultaneously^[28]. The most efficient type of NMT consists of models based on transformer architectures, primarily the Transformer model, which has surpassed earlier methods such as recurrent neural networks (RNN) and convolutional neural networks (CNN)^[29].

Another approach to machine translation is hybrid machine translation, combining different approaches to achieve higher accuracy and flexibility^[30]. Most commonly, they combine neural machine translation methods with older statistical or rule-based models^[31]. The functioning of a hybrid model consists in the individual methods complementing each other. While neural networks enable better understanding of context and more fluent translations, statistical methods or rule-based systems can assist with consistency of terminology and handling of specific linguistic structures.

Among the main advantages of machine translation is the elimination of typographical errors, grammatical and punctuation mistakes, which ensures formally correct output. Today, thanks to progress, NMT models are better able to adapt to linguistic variations and produce more fluent and natural translations, preserve context, more accurately interpret idiomatic expressions, and more effectively convey subtle linguistic nuances^[32].

However, despite significant progress, machine translation still faces challenges, particularly in the areas of contextual accuracy, stylistic adaptability, and specialized terminology^[33]. Problems remain with literalness, insufficient sensitivity to dialects and cultural specifics, and difficult processing of idioms and ambiguous expressions. Translation systems often inaccurately interpret specialized jargon and cannot fully preserve the stylistic nuances of different genres^[34]. Further progress therefore requires more sophisticated language models, expansion of training data, and refinement of adaptation to context and specific communicative purposes^[35].

In the case of court interpreting and translation, the demands are even higher—here, not only linguistic accuracy is required, but also deep understanding of legal concepts and their specific meaning in different legal cultures. For these reasons, current machine translation and interpreting technologies cannot be considered capable of fully replacing court interpreters or translators. Their use in proceedings before public authorities is, at this time, inappropriate from both a legal and practical perspective.

4. Machine interpretation in the Foreigners Residence Act

In 2024, the Ministry of the Interior submitted a government draft of a new Act on the Entry and Residence of Foreign Nationals in the Territory of the Czech Republic, which is to replace the currently applicable Act on the Residence of Foreign Nationals. This draft, which has not yet passed through the legislative process, contains Section 492, which was intended to enable interpretation by means of a technical device: «for the interpretation of an act in proceedings, an administrative authority may use a technical device instead of an interpreter registered in the list of court interpreters and court translators»^[36]. For the time being, this would be the only legal regulation (both substantive and procedural) in the Czech Republic that would permit machine interpreting or translation.

According to the explanatory memorandum, the main reasons for introducing the possibility of interpretation by means of a technical device are the effort to make the administrative process more efficient and faster, as automatic interpretation will enable faster communication and eliminate time delays associated with organizing interpretation with a court interpreter; cost savings, as the use of technology will lead to savings in expenditure on fees for court interpreters; elimination of errors in interpretation, as the Ministry of the Interior assumes that technology should be capable of eliminating errors arising from “human” interpretation, such as omissions or unintentional or intentional omission of information; and the development and support of digitalization of state administration.

The draft presupposes a technical device enabling automatic interpretation; however, it does not take into account the possibility of a technical device for automatic translation, despite the fact that greater progress has been achieved precisely in the field of machine translation. Section 492 provides the administrative authority with the option to use a technical device instead of a court interpreter, but without clear specification of the conditions under which this could occur. The text of the Act does not make clear whether prior informed consent of the party to the proceedings would be required^[37]. The text of the Act does not even envisage the issuance of implementing regulations that would specify the conditions for the use of this technical device.

The provision in question is conceived in such a way that the technical device cannot be regarded merely as a supportive tool intended for basic assistance in communication (e.g., in the form of a chatbot), but as an autonomous means capable of generating interpreting outputs at all stages of the proceedings. The legal regulation does not establish any limitations regarding the type of acts for which the device may be used. No distinction is thus made between situations where technology serves only for informal communication and cases where its outputs may have a direct impact on the content of decisions in the matter and the procedural rights of parties to the proceedings.

The deployment of a technical device for machine interpreting (and translation) into live operation without prior thorough testing of its reliability, accuracy, and security can be considered highly problematic. Such an approach could jeopardize the fundamental procedural rights of parties to the proceedings, particularly the right to a fair trial and the right to understand the content of the proceedings. Before any practical deployment, it is therefore essential to conduct pilot verification in a controlled environment that will enable identification of potential system failures and verification that the device meets the required standards comparable to the performance of a court interpreter or translator.

Paragraph 2 sets out the basic and only requirements for the technical device itself: «the technical device according to paragraph 1 must enable automatic interpretation of spoken foreign language into Czech language and back into the foreign language in real time. The accuracy and speed of interpretation must be comparable to an interpreting act performed by an interpreter registered in the list of court interpreters and court translators»^[38].

It is precisely in the requirement set out in paragraph 2 for accuracy and speed comparable to an interpreting act performed by a court interpreter that one of the fundamental deficiencies of the legislative proposal can be seen. It is in no way clear who and how should assess this comparability and what the tolerated deviation is. It can be assumed that in real time this is not even possible when the administrative authority and the party to the proceedings each command only one of the languages used. If this is a condition that must be met before the deployment of the technical device into “live operation”, it is not clear through what criteria the performance of court interpreters will be assessed and who will evaluate their performance in order to subsequently establish parameters for evaluating the performance of the technical device.

It is also necessary to emphasize that machine interpreting technology is dependent on the availability of large amounts of quality language data from which it learns. However, for less widespread languages (so-called minor languages), there is a shortage of such data, which makes adequate training of systems impossible. The outputs of machine interpreting could therefore be not only inaccurate but also completely misleading, thereby infringing upon the right of parties to the proceedings to a fair trial.

In order to ensure effective control of machine interpreting outputs, it would be essential to systematically make recordings of interpreted acts and simultaneously establish clear rules for their retention and subsequent review. In cases of doubt, it would be necessary for the recording to be evaluated by a court interpreter, which would, however, bring not only additional costs but also possible prolongation of the proceedings.

Furthermore, it remains unclear who and in what manner would be authorized to subsequently review the outputs of machine interpreting, or whether the burden of proof would rest on the party to the proceedings, who would have to prove the discrepancy between the interpreted content and the actual meaning of the statement, as well as the impact of such an error on the course or outcome of the proceedings and on their procedural rights. It is also not clear whether the party would have access to a recording of the machine interpreting, its transcript, or whether they would not be granted access to any materials.

Another question that needs to be resolved before deploying such a device is protection against cyberattacks and data security. Interpreting, like translation, often involves working with highly sensitive information, which may include personal data, information relating to ongoing proceedings, etc.^[39]. Data processed by the technical device may be stored or transmitted through cloud services, which increases the risk of their theft or unauthorized access.^[40] A leak of such information could have serious consequences, including jeopardizing ongoing proceedings, harming the parties, or undermining trust in public administration.

A fundamentally crucial legal question is liability for the functioning of the technical device, particularly in connection with possible errors in automatic interpreting or translation, leakage of sensitive data, or manipulation of the algorithm. Unlike a court interpreter or translator, who is legally liable under Czech law for their errors (cf. Section 21, Section 37(1)(c) or (2)(c) of Act No. 354/2019 Coll., on Court Interpreters and Translators) and for breach of the duty of confidentiality (cf. Section 37(1)(h) or (2)(h) of Act No. 354/2019 Coll., on Court Interpreters and Translators), the technical device bears no legal liability.

The Czech legal order does not address the liability of a technical device, thus creating a legal vacuum, as it is not clear who should bear liability for potential errors—whether the administrative authority that used the device, the software manufacturer, or the technology operator^[41]. The state would likely bear liability for any damage caused by this device. This approach is based in the Czech legal order on strict liability without the need for fault on the part of the obligated person (the so-called no-fault liability model). The state, which implemented the possibility of using machine interpreting or translation into the legal order, would also be liable for any damage, cf. Section 1(1) of Act No. 82/1998 Coll., on Liability for Damage Caused in the Exercise of Public Authority by a Decision or by Incorrect Official Procedure. The purpose of Act No. 82/1998 Coll. is to enable compensation for all cases in which damage was caused by an unlawful decision or incorrect official procedure^[42].

Act No. 82/1998 Coll. allows, in certain cases, recourse claims to be made against official persons, but does not allow them to be made against other entities, such as software manufacturers. The state is thus presented with the option of using the private law route, most likely liability for damage caused by a defective product under Section 2939 et seq. of Act No. 89/2012 Coll., the Civil Code^[43]. Application of this provision is conceivable in situations where during the production process, typically when programming artificial intelligence algorithms, a bug occurs that later causes deviation from the standard and required behavior of artificial intelligence^[44]. For fully autonomous devices, however, the detectability and provability of model failure appears problematic, as the manner in which it arrives at individual outputs remains non-transparent and difficult or even inaccessible for humans to understand^[45]. This insufficient explainability of algorithms leads to the so-called black box phenomenon^[46], where even a technical expert is unable to understand its internal functioning (so-called internal opacity)^[47], retrospectively identify the specific cause of the error, and consequently is unable to determine the liable entity^[48]. In such a case, there would be no entity liable for the damage that occurred^[49] and it would be an example referred to in the literature as a responsibility gap^[50].

5. Conclusion

A robotic interpreter or translator could in the future represent an efficient alternative to a human court interpreter and translator, particularly in terms of economic efficiency, speed, and objectivity of interpreting and translation processes.

From an economic perspective, these technologies could bring significant savings in operational costs, as after the initial investment in acquiring the system and training personnel, the costs of their long-term operation would be lower than the fees paid to court interpreters and translators.

Furthermore, the use of a robotic interpreter or translator would lead to considerable acceleration of administrative processes, as delays associated with organizing interpretation or waiting for the production of a translation by a court translator would be eliminated.

Another significant advantage would be the objectivity of interpretation. It would be free from subjective elements, as automated systems do not work with individual interpretations or subjective views of the content of the message, which could reduce the risk of distortion of the meaning of interpretation, thereby contributing to greater neutrality.

At present, however there is no such system that could equal the performance of a “human” interpreter^[51], and contemporary technologies of machine interpreting or translation cannot be considered suitable for use in proceedings before public authorities. The reasons are primarily their insufficient quality and reliability, which can lead to significant errors in outputs, distortion of statements by parties to the proceedings, or incomprehensibility of translated text. Such deficiencies would directly threaten the right of parties to the proceedings to interpretation, and thereby their right to a fair trial.

In the field of translation, however, it can be reasonably anticipated that technologies enabling automated text processing will achieve a high level of accuracy in the near future. Particularly for repetitive acts, such as the translation of standardized documents like birth certificates, the effective use of machine translation can be considered given their template-like structure. In these cases, automated systems could achieve a high degree of reliability, as their outputs would be based on predefined patterns and established terminological databases.

Alongside technical limitations, however, there also exists a legal vacuum. Current legislation does not contain any framework regulating liability for incorrect outputs of machine translation or interpreting, nor does it address mechanisms by which it would be possible to ensure defense against potential deficiencies in machine-generated outputs. It is not yet clear how one would proceed, for example, in the case of objections by a party to the proceedings against the incorrectness of translation or interpreting—whether the output would be reprocessed by machine, possibly by different software, or whether it would be necessary to call in a court interpreter or translator for verification. The absence of these legal solutions currently represents one of the obstacles to the deployment of these technologies in practice.

On the other hand, it is undeniable that legislation will sooner or later have to adapt to rapidly developing technologies, particularly in the field of machine translation. Section 492 of the draft Act on the Residence of Foreign Nationals can be perceived in this regard as a first step toward integrating these innovations into the area of public administration. Given the current challenges and unresolved questions, however, it would be advisable to approach this possibility with prudence.

One possible way to integrate machine outputs into proceedings before public authorities could be a pilot project^[52], within which a machine interpreting or translation system would operate alongside a human interpreter or translator, who would simultaneously perform a control role or post-editing of outputs. It would be necessary to test the device for each language under consideration separately, as the device learns from different datasets for each language pair. Such experimental deployment would enable not only evaluation of the actual quality and benefit of the technologies in real conditions, but also identification of weaknesses and preparation of appropriate legislative measures.

The introduction of machine interpreting or translation into administrative proceedings has the potential to significantly affect the rights of parties to the proceedings, particularly their right to interpretation, and thereby the right to a fair trial. In this context, a legitimate question arises as to whether the primary motive for introducing machine interpreting is not primarily economic savings, at the expense of the quality of interpretation and protection of the rights of parties to the proceedings. Particularly concerning is also the risk of discrimination against persons who speak less widespread languages, for which sufficiently high-quality language models do not exist. It is evident that the draft completely disregards the fundamental role of court interpreting as a procedural safeguard, which serves not only to overcome language barriers, but primarily to protect the rights of parties and to ensure fairness of proceedings.

At this moment, the most ideal approach appears to be a hybrid model that combines the use of artificial intelligence with the human factor. Such models promise to accelerate and simplify the work of court interpreters and translators while maintaining the required level of accuracy. In the field of translation, these are so-called CAT tools (computer-assisted translation)^[53] and in the field of interpreting, computer-assisted interpreting^[54], which involves the use of digital technologies to increase the efficiency and accuracy of human interpreter work^[55].

From the perspective of the systematics of legal regulation, it can be recommended de lege ferenda that in the event of achieving such technological progress that would allow machine interpreting and translation to be considered equivalent to the performance of a court interpreter and translator, the legal regulation of their use should be enshrined directly in procedural regulations. This would ensure uniform and coherent application of these technologies in all types of proceedings, instead of their partial and unsystematic regulation in individual sectoral acts.

1. I. Čeňková, Úvod do teorie tlumočení, Česká komora tlumočníků znakového jazyka, Prague, 2008, p. 11. ↑
2. D. Knittlová et al., Překlad a překládání, Univerzita Palackého v Olomouci, Filozofická fakulta, Olomouc, 2010, pp. 221-225. ↑
3. C. Wadensjö, Interpreting as Interaction, Addison Wesley Longman, New York, 1998, p. 74. ↑
4. «In the determination of any criminal charge against him, everyone shall be entitled to the following minimum guarantees, in full equality: […] To have the free assistance of an interpreter if he cannot understand or speak the language used in court». ↑
5. «Everyone who is arrested shall be informed promptly, in a language which he understands, of the reasons for his arrest and of any charge against him.» a «Everyone charged with a criminal offence has the following minimum rights: […] to have the free assistance of an interpreter if he cannot understand or speak the language used in court.». ↑
6. J. Stavinohová, Zásada rovnosti účastníků, in D. Hendrych et al., Právnický slovník, III Ed., C. H. Beck, Prague, 2009. ↑
7. J. Blahož, K. Klíma, J. Skála, Ústavní právo Evropské unie, Aleš Čeněk, Dobrá Voda, 2003, p. 590. ↑
8. Art. 36(1) Charter of Fundamental Rights and Freedoms of the Czech Republic. ↑
9. Constitutional court, judgement, 22 January 2008, n. Pl. ÚS 54/05. ↑
10. W. Hewitt, in H. Mikkelson, Evolving Views of the Court Interpreter’s Role: Between Scylla and Charybdis, in C. Valero-Garcés, A. Martin (eds.), Crossing Borders in Community Interpreting: Definition and Dilemmas, John Benjamins, Amsterdam, 2008, p. 81. ↑
11. A. Winterová, Civilní právo procesní, III Ed., Linde, Prague, 2004, pp. 66-67. ↑
12. Constitutional court, judgement, 12 March 2019, n. III. ÚS 3464/17. ↑
13. Constitutional court, judgement, 8 August 2005, n. II. ÚS 186/05. ↑
14. Supreme Court, resolution, 12 July 2001, n. 6 To 550/2001. ↑
15. Constitutional court, judgement, 7 July 2021, n. II. ÚS 482/21. ↑
16. T. Etchegoyhen et al., Cascade or Direct Speech Translation? A Case Study, in Applied Sciences, 22, 3, 2022, pp. 1019-1121. ↑
17. J. Iranzo-Sanchéz et al., Streaming cascade-based speech translation leveraged by a direct segmentation model, in Neural Networks, 142, 2021, pp. 303-315. ↑
18. M. Labied et al., A Multicriteria Comparison of End-to-End and Cascade Speech-to-Text Translation Models, in Bulletin of Electrical Engineering and Informatics, 14, 4, pp. 2837-2848. ↑
19. H. Wang et al., Progress in Machine Translation, in Engineering, 18, 11, 2022, p. 148. ↑
20. C. Fantinuoli, The Emergence of Machine Interpreting, in EST Newsletter, 62, 2023, p. 1. ↑
21. Ibid, p. 12. ↑
22. W. Xiaoman, C. Fantinuoli, Exploring the Correlation between Human and Machine Evaluation of Simultaneous Speech Translation, in Proceedings of the 25th Annual Conference of the European Association for Machine Translation (volume 1), Sheffield: European Association for Machine Translation, 2024, pp. 329-330. ↑
23. N. P. Russo, Faut-il craindre l’interprétation automatique ? De la fiction à la réalité, le point sur les «traducteurs électroniques», in Traduire, 246, 2022, p. 102. ↑
24. Y. Cao, Comparative Analysis of Machine Interpreting and Human Interpreting: Insights into Consecutive Interpreting Teaching, in Advances in Social Science, Education and Humanities Research, 914, 851, 2024, pp. 730-741. ↑
25. F. Pöchhacker, Is Machine Interpreting Interpreting?, in Translation Spaces, 2024, p. 13. ↑
26. W. J. Hutchins, H. Somers, An Introduction to Machine Translation, Academic Press, London, 1992, p. 150; cf. P. Naveen, P. Trojovský, Overview and Challenges of Machine Translation for Contextually Appropriate Translations, in iScience, 27, 10, 2024, p. 2. ↑
27. Deep learning is a specific method of machine learning that uses deep neural networks. Each layer analyzes the inputs it receives from the previous layer, and its output is further processed by the next layer. ↑
28. Y. Zhao, J. Zhang, C. Zong, Transformer: A General Framework from Machine Translation to Others, in Machine Intelligence Research, 20, 4, 2023, pp. 515-519. ↑
29. S. Lankford, H. Afli, A. Way, Human Evaluation of English–Irish Transformer-Based NMT, in Information, 13, 7, 2022, pp. 1-19. ↑
30. A. Anugu, G. Ramesh, A Survey on Hybrid Machine Translation, in E3S Web of Conferences, 2020, p. 3. ↑
31. R. K. Dwivedi, P. Nand, O. Pal, Hybrid NMT Model and Comparison with Existing Machine Translation Approaches, in Multidisciplinary Science Journal, 7, 4, 2024, p. 1. ↑
32. A. Sugiyama, N. Yoshinaga, Data Augmentation Using Back-Translation for Context-Aware Neural Machine Translation, in Proceedings of the Fourth Workshop on Discourse in Machine Translation (DiscoMT 2019), Association for Computational Linguistics, Hong Kong, 2019, pp. 35-44. ↑
33. Chamber of Court Interpreters and Translators of the Czech Republic, AI – umělá inteligence v překladatelské praxi?, in Soudní tlumočník, 2, 2023, p. 8. ↑
34. P. Naveen, P. Trojovský, Overview and Challenges of Machine Translation for Contextually Appropriate Translations, in iScience, 27, 10, 2024, pp. 8-9. ↑
35. For less widely spoken languages, an additional challenge arises in the form of limited training data. To improve the performance of neural machine translation models, back-translation may be employed; however, its main drawback lies in the generation of so-called pseudo-sentences, i.e. low-quality translations that may negatively affect model performance (S. Ranathunga et al., Neural Machine Translation for Low-Resource Languages: A Survey, in ACM Computing Surveys, 55, 11, 2023, pp. 1-37). ↑
36. Translation from Czech; original wording in Czech. ↑
37. According to the explanatory memorandum to the draft law on the residence of foreigners, Section 492, it is up to the administrative authority to decide whether technical equipment will be used. ↑
38. Translation from Czech; original wording in Czech. ↑
39. I. Horváth, AI in Interpreting: Ethical Considerations, in Across Languages and Cultures, 23, 1, 2022, pp. 8-9. Cf. J. Moorkens, D. Kenny, F. De Carmo, Fair MT: Towards Ethical, Sustainable Machine Translation, in Translation Spaces, 9, 1 (special issue), 2020, pp. 1-11. ↑
40. C. Canfora, A. Ottmann, Risks in Neural Machine Translation, in Translation Spaces, 9, 1, 2020, pp. 58-77. Cf. K. Siau, W. Wang, Artificial Intelligence (AI) Ethics: Ethics of AI and Ethical AI, in Journal of Database Management, 31, 2, 2020, pp. 74-87. P. Yang et al., Data Security and Privacy Protection for Cloud Storage: A Survey, in IEEE Access, 8, 2020, p. 2. ↑
41. Uncertainty in this area could negatively affect the exercise of the rights of parties to proceedings, particularly with regard to claims for damages, which could undermine public confidence in procedural fairness. ↑
42. The situation described above can be compared to liability for damage caused by particularly dangerous operations under Section 2925(1) of the Civil Code, where the operator is always the person responsible for a particularly dangerous plant. According to this provision, anyone who operates a particularly dangerous plant or other facility shall compensate for damage caused by the source of increased danger; operation is particularly dangerous if the possibility of serious damage cannot be reasonably excluded in advance, even with the exercise of due care. Case law has concluded that things whose use, under certain circumstances, has secondary harmful effects that are not fully controllable by humans and which are associated with a high degree of probability of causing harm (such as a frame saw located in unsuitable geological conditions) should also be considered a source of increased danger (Supreme Court of the Czechoslovak Socialist Republic, judgment 31 May 1983, n. 1 Cz 13/83 [R 24/1986]). Artificial intelligence, which is capable of reaching its own conclusions based on collected and analyzed data and subsequently behaving in a manner that may be unpredictable, can be considered a source of increased danger. ↑
43. The responsible person pursuant to Sections 2939 and 2940 of the Civil Code may be (i) the manufacturer of the product, (ii) the quasi-manufacturer (a person who labels the product with their name, trademark, or other symbol that leads the customer to believe that such person is the manufacturer), (iii) the importer who imported the product as part of their business for the purpose of placing it on the market, or (iv) if the manufacturer or importer cannot be identified, the supplier is the responsible person (M. Hulmák et al., Commentary on Sections 2055–3014, in M. Hulmák (ed.), Civil Code: Commentary, vol. VI, Obligations Law – Special Part, C. H. Beck, Prague, 2014, p. 1652.). ↑
44. In the development of artificial intelligence systems, it is also relevant to consider the concept of shared responsibility among entities involved in design, programming and production (D. C. Vladeck, Machines without Principles: Liability Rules and Artificial Intelligence, in Washington Law Review, 89, 2014, p. 149.). ↑
45. T. Svoboda, Chatboty ve veřejné správě – stručný nástin problematiky, in Správní právo, 6-7, 2024, p. 513. ↑
46. J. Nešpor, Automated Administrative Decision-Making: What Is the Black Box Hiding?, in Acta Universitatis Carolinae – Iuridica, 2, 2024, p. 69 ff. ↑
47. J. Strakoš, Právní aspekty automatizace ve správním trestání v kontextu algoritmů strojového učení, in Správní právo, 6-7, 2024, p. 486. ↑
48. M. Koubová, Umělá inteligence ve zdravotnictví: Kdo ponese odpovědnost v případě škody?, Ekonomický deník, online, 16 August 2019. ↑
49. L. Kolaříková, Odpovědnost (za) robota aneb právo umělé inteligence, in Bulletin advokacie, 3, 2018, p. 15. ↑
50. D. G. Johnson, Technology with No Human Responsibility?, in Journal of Business Ethics, 127, 4, 2015, pp. 707-715 or A. Matthias, The Responsibility Gap: Ascribing Responsibility for the Actions of Learning Automata, in Ethics and Information Technology, 6, 2004, pp. 175-183. ↑
51. Explanatory Memorandum to the Draft Act on the Entry and Residence of Foreign Nationals, Section 492, p. 434. ↑
52. A pilot project can be likened to a regulatory sandbox within the meaning of the Artificial Intelligence Regulation (AI Act), which provides businesses with the opportunity to experiment with new products and services with limited application of certain regulatory requirements, thereby facilitating innovation and investment in the field of artificial intelligence. ↑
53. CAT tools are software programs that help translators streamline their work by storing previously translated segments, managing terminology, and ensuring consistency and accuracy in translation (T. Svoboda, Kapitoly z překladatelské praxe: odborný překlad mezi němčinou a češtinou, Charles University, Faculty of Arts, Prague, 2012, p. 80.). ↑
54. CAI tools include, for example, speech-to-text tools (automatic transcription of spoken words on screen) or remote interpreting tools that connect interpreters and participants remotely. ↑
55. B. Prandi, Computer-Assisted Simultaneous Interpreting: A Cognitive-Experimental Study on Terminology, Language Science Press, Berlin, 2023, pp. 35-38. ↑