Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students

This article compares the output of three neural machine translation systems (Google Translate, DeepL, and Phrase TMS) and human translation (undergraduate level students, English into Spanish). It focuses on five formal neologisms extracted from literary texts, thus considering creativity, and technology adoption and training.


Introduction
Literary translation is a social activity that varies according to the cultures and languages involved, with conventions, norms, and expectations specific to the target system, Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students subdivided by derivation (suffixation and/or prefixation), composition, lexicalisation, acronymy, or abbreviation (Cabré Castellví, 2006).These alterations in language are the result of the dynamics of the language itself and the creativity of the speakers, who are the ones who bring about this linguistic change (Estornell Pons, 2009).And this inclusion also happens in literary texts.
Given the degree of creativity and novelty involved, translating neologisms is one of the greatest challenges faced by translators of literary texts (Guerberof-Arenas and Toral, 2022).In literary genres of any kind, including science fiction, fantasy works, thrillers or romance, onomastic creation of new words is commonplace (cf.Szymy'slik, 2018;Noriega-Santiáñez and Corpas Pastor, 2023).Depending on the type of neologism, the literary translator is faced with a wide range of problems (e.g., different compositional structures, metonymy or shift of paradigm), which can be tackled with different syntactic, semantic, or 'continuist' strategies (Postolea, 2011).Nevertheless, literary translators do not seem to have much guidance or help at their disposal to carry out such a difficult and demanding task (Noriega-Santiáñez and Rodríguez Martínez, 2020;Noriega-Santiáñez and Corpas Pastor, 2023).Due to its idiomatic nature, along with the intralinguistic and extralinguistic factors at play (Burgués Estrada and Aguilar-Amat, 2019), it is crucial to identify reliable resources and tools for documentation that help address terminological, phraseological, and contextual aspects, as it is explored below.
The interweaving of different disciplines such as Corpus Linguistics (branch that studies data obtained from corpora), Natural Language Processing (NLP) (branch of Artificial Intelligence that helps machines to understand and process spoken and written human language) or Computational Linguistics (area of NLP that studies the development of linguistic applications using computational technologies) has had a significant impact on Translation and Interpreting studies (González Fernández, 2018;Corpas Pastor et al., 2021).Against this background, several ICT technologies can be found to assist translators, such as corpora, online glossaries, repertoires, encyclopaedias or databases, spell checkers, online monolingual or bilingual dictionaries, revision tools, parallel texts, or lexicons, among others (Merlo Vega, 2005;Biau Gil and Pym, 2006;Corpas Pastor, 2013;Surià López, 2014;Bowker and Corpas Pastor, 2022;Rothwell et al., 2023).These resources can be used to uncover relevant information about a particular author or the socio-cultural context of the (literary) text, and an infinite number of nuances with which give coherence to that text.Moreover, in the translation market, computer-assisted translation (CAT) tools, that rely significantly on translation memories and termbases, have become commonplace (Carl and Braun, 2018).In addition, machine translation (MT) systems have recently been added to CAT tools (Rothwell et al., 2023).In fact, MT systems have become very useful to automate translation tasks as well as to increase the speed and efficiency of the translator under certain circumstances (O'Brien, 2012).
Their application to literary translation is a recent development.Nonetheless, some studies tentatively explore the use of MT systems for translating literary works (cf.Toral and Way, 2015;Webster et al. 2020; to name but a few).
Following the lead of pioneer studies that test MT in literary translation (Toral and Way, 2015;Moorkens et al., 2018;Matusov, 2019;Webster et al., 2020, among others), our study aims to compare the NMT versus human outputs when translating neologisms.
To this aim, a corpus-based pilot study assesses the production of three NMT systems (Google Translate, DeepL and Phrase TM) with the responses of undergraduates from the Degree in Translation and Interpreting at the University of Malaga, Spain.Therefore, it tests the scope of some of the most relevant NMT systems used in the field (cf.Webster et al. 2020, Caro Quintana andCastilho, 2022, to name but a few) in comparison with the human factor (HT) in undergraduates with translation skills and a high language proficiency.Precisely, our pilot study compares and evaluates the creativity (in terms of novelty and acceptability, cf.Guerberof-Arenas and Toral, 2022) of five formal neologisms, i.e., neologism formed by composition whose lexical basis is related to known realities of the seasons of the year (namely, "winter").The examples of neologisms are extracted from corpora of literary texts (American Google Books and British Google Books), following a rigorous selected criteria described in the Methodology section.Furthermore, an adhoc questionnaire is designed to investigate the technology adoption among undergraduates.Finally, the incorporation of these technologies in training is discussed.
The paper is organised as follows.Section 2 provides a brief discussion about the way MT systems are changing the paradigm of literary translation.Against this background, Section 3 introduces our pilot study, outlining its methodology and presenting the primary preliminary findings regarding the challenges in machine translation when dealing with neologisms.Finally, Section 4 provides the main conclusions and limitations of our study, as well as some further avenues of future research.

Related work on MT systems in literary translation
Technology has created a new vision in the market and the translator's praxis, and literary translators are not left out of this evolving panorama (Noriega-Santiáñez and Corpas Pastor, 2023).In fact, an increasing number of studies focuses on the possibilities offered by MT in the translation of literary works.Our study follows this trend, as it focusses on the performance of NMT systems for the translation of literary texts, in comparison with human translation.
NMT systems emerged in the 1980s and 1990s and its popular uptake from the mid-2010s onwards "started with the integration of neural language models into traditional statistical machine translation (SMT) system" (Koehn, 2017: 5).NMT systems such as Google Translate (formerly a SMT engine) and DeepL became popular worldwide in open access free online platforms, as their output was more refined and much more like a HT (Large, 2018).According to Toral and Way (2018: 2), these systems "can attain better translation quality than the dominant approach to date, namely phrase-based statistical MT".While MT is commonly linked to technical or scientific texts, its application has expanded in recent years to include literary translation (Toral et al., 2018;Webster et al., 2020;Guerberof Arenas and Toral, 2022).Given "the maturity of post-editing in

237
industry", as well as the new paradigm of NMT (Toral and Way, 2018: 2;Moorkens et al. 2018), the extent to which these tools help literary translators is increasingly being evaluated (Toral et al., 2018).This shift has prompted an exploration of various advantages, constrains, and ethical issues, some are mentioned below in relation to our pilot study.
In what follows, we provide a brief overview of relevant related work on MT comparing to HT, with special emphasis on creativity and educational dimension.First, some studies that primarily assess the performance of MT systems for literary texts will be mentioned.
Voigt and Jurafsky (2012) used Chinese literary and non-literary texts to test the degree of referential cohesion between HT and Google Translate, concluding that the former SMT system was not able to fully emulate cohesion, probably due to the lack of literary data in its corpus.Toral and Way (2015)  HT demonstrated richer style.
Precisely, the assessment of creativity with NMT is one of the cornerstones on our study.The main authors that explored this feature were Guerberof Arenas and Toral (2020,2022), who measured levels of creativity based on the textual elements in literary works.In both studies, they analysed creativity in literary translation using MT, machine translation post-editing (MTPE) and HT.The results of both studies showed that the intervention of the literary translator is essential in the transfer of creativity.However, while the first study proved that there were creativity-related aspects in which HT and post-edited MT performed similarly, the second study suggested that post-edited MT might hampers creativity.In contrast, the MT output appeared currently unable to convey a creative solution.What emerges from this body of work so far is that the comparison of HT against MT remains a popular subject of study, given the great complexity of literary texts.
Moreover, the incorporation of technologies into the students' training is also observed.
Our study will explore whether the use of technologies, especially MT, is a conditioning factor for human translators and whether the NMT provides creative responses.To this end, a pilot study with student translators has been set up, as described in section 3.

Methodology
In what follows, we will deal with the methodological aspects of our study, namely data collection and selection of neologisms, NMT systems chosen for the experiments and subjects.

Corpora of literary texts
Given the complex nature of terminological variation and their degree of occurrence in literature, the neological challenge proposed in this pilot study was limited to formal neologisms (cf.Szymy'slik, 2018;Rodríguez Martínez, 2020;Ridruejo, 2020, to name but a few).In the first stage, we searched for formal neologisms in contexts within corpora of literary texts, i.e., American Google Books (155 billion words from American English "winter" as a lexical basis was employed, using the substring function (i.e., "*winter" and "winter*").
Therefore, the following neologism selection criteria was designed: 1) one of its constituents must be "winter" (as it is a very common unit and reality in many languages; hence it is prone to modification and widely appear in literature); 2) the second constituent must be a recognised word; that is, the type of formal neologism chosen must be formed by composition of two radicals; and 3) it must be a neologism according to the literary context of the work, i.e., the author consciously used it as a means to simulate the novelty of a neologism (following the lexical creation parameters by Díaz Hormigo, 2012), evoking a distinct atmosphere (medieval, children's, fantasy, etc.).
Among all the results obtained, we checked in some of the most prestigious online dictionaries (Cambridge Dictionary, Merriam-Webster, and Collins Dictionary) if the examples were indeed neologisms.Even if any appear, they do so with a different meaning or were neologisms in the process of becoming established.Thus, the selected neologisms were manually checked within the context of the literary work to ensure their literariness, i.e., that they are created or used for the specific purpose within the novel's setting.For this reason, the neologisms have been used in their context in both human and machine translation tests.Therefore, the particularity of doing this practice in literary translation is the need to contextualize the phraseological challenge within the setting of the work.
Once the neologisms that met the criteria mentioned above were chosen, a fact sheet with these five examples organised alphabetically was made, including the neologism in context, the novel's author, title, and year of publication (see Table 1, below).

Neologism Neverwinter
Context Against all odds, Drizzt and Dahlia join forces in the aftermath of battle, united in their desire for vengeance against the sorceress who destroyed Neverwinter.

Novel and author
Neverwinter

Novel and author
Winterland: A Novel by Alan Glynn (1960)

Novel and author
Wintertide by Linnea Sinclair ( 2016) Table 1.Formal neologisms The five literary examples chosen are found in three fantasy novels (neologisms 1, 3, and 5); one children's novel (4) and one mystery novel (2), respectively.In particular, neologisms 1, 2, and 4 are toponyms, neologism 2 is the name of a business company, and neologism 5 refers to a seasonal time.

Choice of NMT systems data
The second stage is the use of NMT systems (DeepL, Google Translate and Phrase TMS), to translate into Spanish these neologisms.Both the neologisms with and without context were tested.That is, first we assessed the translation of the neologism in isolation ("no context, NC"), and then the neologism in context ("in context, IC"), i.e., the complete sentence in which the neologism appeared has been introduced in the MT systems and used as context; however, the assessment focuses solely on the neologism.Then we evaluated the performance of these three NMT systems and we identified the most used techniques in the translation of these five formal neologisms, following the classification proposed by Hurtado Albir (2011).It encompasses techniques such as borrowing, literal translation, explanation, generalization, among others.

Human translation setup
In the third stage, the collection of HT data is explained, including the participants, the modules, the questionnaires, and the sessions involved.
A total of 54 students participated in our pilot study, who are enrolled in two different modules of the Degree in Translation and Interpreting at the University of Malaga, taught by the same teacher who kindly agreed to let us set up this study: • Module 1: Traducción General "BA-AB" (II) Inglés-Español/Español-Inglés.This is a second-year subject and the second translation module in the degree (45 hours per semester).A total of 38 students participated, primary aged from 19 to 22, having English as a second language and Spanish as a first or third language.
Most students are in their second year, with two in their third year and one in their first year.Only two of the students have professional translation experience (in a family-owned translation company).• Module 2: Traducción General "CA-AC" (II) Inglés-Español/Español-Inglés.This is a third-year subject and the third translation module, but two of them were in the French-Spanish language pairs (45 hours per semester).A total of 16 students participated, ranging from 20 to 22 years old.They have English as a second or third language (the language used in this subject) and Spanish as a first or third language.All participants are third-year students.Only two of the students have professional translation experience (an internship in a translation agency and volunteering).
Each module has been divided into two groups: • Group A: the students were asked to translate the neologisms specifically without using MT systems.• Group B: the students were free to use any tool or resource of their choice.

Demography
Module  The neologism translation practice consisted of translating the five neologisms (see Table 1) into Spanish.The post-translation questions focused on translation time, encountered The HT in this study has been compiled in a 90-minutes session for each module, following the structure and the scheduled time detailed below: 1.The students filled in the pre-translation questions (5 minutes).
2. The participants were given a training seminar entitled "Formal neologism: a whole new wor(l)d".The seminar focused on some theoretical and practical definitions of neology: the difference between "neology" and "neologism", the classification of neologisms both in English and Spanish, with special emphasis in formal neologisms, and some translation strategies (i.e., adaptation, borrowing, literal translation, etc.) into Spanish through examples in literary contexts (30 minutes).
3. The students filled in the neologism translation exercise and the post-translation questions (35-45 minutes).
4. They discussed their translations and shared some thoughts about the use of technologies and MT systems in literature (5-15 minutes).
In the fourth stage of this study, the data obtained from the four groups of students is compiled, compared, and evaluated.

Evaluation and results
In this section, the results obtained by the NMT systems and the undergraduates are evaluated in terms of creativity and linguistic and contextual adequacy.Additionally, a literal translation into English is included in square brackets to enhance the understanding of the Spanish MT and HT output. 1 .
Through a quantitative approach, the performance of MT and HT is measured, as well as the adoption of technologies in the translation of formal neologisms.

Machine translation
The following tables show the first outputs of neologisms produced in-context (IC) and no-context (NC) by the NMT systems (DeepL, Google Translate and Phrase TMS).In some cases, there are more than one term produced, but only up to the second has been included. 1 In those examples where it is unfeasible to provide a literal translation, the neologism is specified in the following ways: 1) [word1 + word2] when it is a neologism formed by the composition of two constituents; 2) [word'] when a neologism has been formed by the derivation or suffixation of a word; and 3) ["word"], when an English word has been used in the Spanish translation.In Phrase TMS, there is little difference between using the neologism in context and without context, since only one of the neologisms has been translated into Spanish.From these data, it can be deduced that students are not entirely satisfied with the tools at their disposal to tackle creative phraseological challenges (i.e., neologisms).
The detailed study and specific factors for each module are discussed below.

Module 1
This section entails the performance of both groups in Module 1 regarding to the translation of neologisms.
• Group A A total of 21 students participated in group A, who were not allowed to use MT systems.In their pre-translation questions, 16 students have a B2 English level, 3 a C1 and 2 a C2.Concerning the translation practice, most of them spent between 10 and 30 minutes translating all the neologisms.Table 8 shows their results.) and thus providing more creative formal neologisms.However, some had not taken the context into account (as in the case of Winterland, which should be kept as a loanword as it is the name of a company).Others adapted the neologism phonetically to Spanish (Nevergüinter [Never + "winter"] or Güinterland ["Winter" + land]), which ended up losing the meaning in the target language.Finally, some participants failed to address the neologism's meaning and focused on its denotative aspect, leading to a false meaning (Pequeños árboles de China [Small trees from China] or avalanche [avalanche]).

Neverwinter
Concerning the tools and resources used in this practice, almost 40% of respondents specifically employed Wordreference, compared to almost 48% who said they worked with online monolingual dictionaries (most notably Cambridge Dictionary and Collins Dictionary) and thesaurus.Very few mentioned the Google's search engine (mostly using Wikipedia).As for the difficulties encountered in this practice, some students primarily highlighted the challenge of conveying the meaning of the neologism into Spanish and the complexity of neological creation.Others highlighted the lack of useful information from the available sources.
• Group B   ), as well as some terms that came up by using the meaning of the semantic field from the lexical base (Sinvierno [No + winter], Dulcerable [Sweet']).Some HT output was created by suffixing the word "inver" (invernalia [winter + land/ winterfell], inverna [winter']), which conveys the concept of "winter" and "land" in the target language.
In addition, the meaning of some neologisms did not correspond to that English Finally, there were some literal translations (El invierno del nunca jamás [The winter of never ever], probably mistaken by the word Neverland).
Regarding the tools and resources used, 52.9% mentioned Wordreference and MT systems (in which DeepL and Google Translate stand out, although Reverso is also mentioned), compared to 41.2% who have used the Google's search engine (i.e., Google Photos, etymology pages or wiki fandoms).As for the difficulties encountered, some 251 students had problems to transfer the creative and contextual meaning of the neologism, as well as others regarding the limited time available to do this practice.Finally, other participants did not know the target language well enough.

Module 2
This section entails the performance of both groups in Module 2 regarding to the translation of neologisms.
• Group A Group A encompassed 10 students, 5 have a B2 English level, 3 a B1, 1 a C1 and 1 a C2.The participants took an average of 15-30 minutes to complete the practice.some neologisms made by composition were not understood in the target language (Macasar, Marglas [Tide']).There were also several false meanings, such as the case of using the lexical base "tide" of Wintertide as "marea [tide]" (Mar Helado [Icy Sea], Marea de invierno [Winter tide]), when in this context tide is an archaic word referring to a specified time or season.Finally, some students focused on the denotative meaning (Chimonanthus praecox, Cierre congelado [Freezed closing]), so that the author's intention is lost.

Neverwinter
The participants could use any resource or tool except MT systems during this practice.Most of them highlighted the use of the Google's search engine (60%), especially for contextual searches of the novels; also, electronic dictionaries such as the Cambridge Dictionary (40%) or Wordreference (30%); finally, parallel texts (10%) and other sources (20%) such as blogs or Pinterest are mentioned.The greatest difficulties highlighted by the participants were the creative skills and the ability to adapt the neologism and all its nuances to the target language.
• Group B There were 6 students in this group, 5 of whom have a B2 level of English and 1 a B1.The practice of translating these neologisms took them an average of 20 to 40 minutes, the results of which are as follows:  or Amazon (16.7% each) were mentioned.As for the difficulties encountered during this practice, the students highlighted the creative component needed to transfer the neologism, as well as the context or the search for synonyms.

Discussion
The quality of the results of the neological creation made by a human is not comparable to that of the MT systems, as pointed out Awadh and Khan (2020) and Sahin and Gürses (2021).Indeed, they concluded that students provided better but not entirely accurate translations.In our study, the NMT output show lower lexical richness than students' translations, which is in line with some findings from Webster et al. (2020).Regarding creativity, the students produced translations that exhibit higher levels of originality, as some studies pointed out (Guerberof Arenal and Toral, 2020;2022).
However, some students relied on MT, indicating a growing use of these systems even in creativity-related challenges.Our findings show that DeepL is also the NMT system preferred by most students.Furthermore, DeepL stands as the most useful system in terms of level of lexical accuracy, far ahead of Phrase TMS.This contradicts the outcomes of Webster et al. (2020), who proved that Google Translate makes less accuracy, but more fluency errors in literary text compared to DeepL.
In terms of productivity, there does not seen to be a significant difference between those students who have used MT and those who have not.In fact, those groups who could use MT have taken a little longer to do the practice, even though some studies pointed to a higher productivity with MT systems (cf.Moorkens et al., 2018;Toral et al., 2018).The time it took them to complete the exercise also varies between modules, which can be due to different levels of language proficiency and practice: Module A students have more experience translating from English into Spanish than those of Module B, and they also have a better English proficiency.
Concerning technology adoption, the most used tools and resources in both modules were the Google's search engine, online dictionaries (monolingual and bilingual dictionaries, and thesaurus), MT systems, and, particularly, Wordreference (as it is a multifunctional tool that provides the definition of the term, its synonyms, and/or its translation into different languages).No student used paper-based tools or resources during the practice, and only very few stated that they did not use any technology.It can be argued that translation students have deeply integrated technological tools into their workflow, even to address literary challenges, which is in line with the findings of Sahin and Gürses (2021).However, there is a general feeling that there are not many specific sources of documentation on translating (formal) neologisms, as some studies highlighted (Noriega-Santiáñez and Rodríguez Martínez, 2020;Noriega-Santiáñez and Corpas Pastor, 2023).This need explains the relevance of teaching undergraduates MT and other e-tools (such as corpora, online platforms, CAT tools or softwares) to translate literature, which follows the outcomes of several studies (cf.Zanettin, 2017;Dimitroulia and Goutsos, 2017;Tian and Zhu, 2020;Youdale and Rothwell, 2022; to name but a few).Thus, this practice can be used to reflect on the incorporation of these technologies into the student's curricula, so they can learn about their limitations in practice, for instance, when dealing with complex phraseological challenges (e.g., formal neologisms).This is in consistency with the results of previous studies (cf. Hidalgo Ternero and Corpas Pastor, 2020;Abdulaal, 2022).
In addition, there are some ethical issues to consider that other studies have also brought up.For example, the extent to which it is good for literary translation students to use MT that may constrain their creative voice in the text (cf.Kenny and Winters, 2020;Matusov, 2019;Guerberof Arenas and Toral, 2022).In fact, it might have had a negative impact on the student's translation in some of our study's groups.As Taivalkoski-Shilov (2019) pointed out, the MT is often not able to render all the stylistic features 255 and meaning of the source text.Perhaps this would make translations less original and of poorer quality, as the translator's voice is further limited in post-edited works, as highlighted by Kenny and Winters (2020).Therefore, undergraduates would not make the effort to come up with new terms.
Finally, the limitations of our pilot study relate to the low number of participants (although students who have English as their second and third language of study have been also included), and the number of formal neologisms, which is also rather low (due to the intricate nature of conducting research of translating new words in a time-limited seminar).

Conclusions
To the best of our knowledge, this is one of the few studies that compares the quality between MT and HT of formal neologisms extracted from literary works in the English>Spanish language pair.The outcomes reinforce the idea that MT performance is not yet up to solve all complexities presented by formal neologisms in literary works (from their connotative and denotative meaning in context to their adaptation to the target language).When the role of creativity comes into play, MT systems are still unable to convey all these features, precisely because of all the intricate aspects involved in the creation of new terms (e.g., linguistic, denotative and/or referential meaning, to name but a few).However, students continue to use them as a source of reference or inspiration, but traditional technological resources (as electronic thesaurus or dictionaries) are still the most employed.The emergence of technologies in the classroom is an inherent reality for students and it significantly affects their translations.Thus, many students rely heavily on electronic resources, even for phraseological challenges that require creativity skills.However, some of our main contributions point to the advantages (effective terminology and contextual queries) and limitations (lack of results, literal, or inadequate translations) of technologies to overcome the challenges posed by formal neologisms in literary translation.Hence this contribution reinforces the need to effectively teach the use of these technologies to literary translation students.
Following our promising preliminary results, we plan to expand our pilot study in various ways: by adding another pair of languages, increasing the number of students, and incorporating other groups of subjects, such semi-professionals (i.e., 4 th year or MA students) and professional translators.We also intend to refine our methodology and triangulate results by using evaluation metrics and human annotators to assess aspects such as readability, naturalness, and creativity.Thus, our future research will follow the path of studies that address emerging topics such as MT performance in pedagogical context within different literary genres.Particularly, this research might delve into the potential uses of these tools when dealing with complex phraseological challenges.In addition, this study might explore the integration of MT into the literary translator's workflow as an aid, but never to replace human translators.
investigated the MT output of a French novel into English and Italian (revealing that closer languages are easier to post-edit) and proposed the idea of training a customised SMT system for a Spanish>Catalan literary translation.In another paper,Besacier and Schwartz (2015)  studied the translation of an American essay from English into French by an MT system and then post-edited and proofread by non-professional translators.Both the postedited and proofreading version were evaluated by BLEU and then the MT+PE text by readers' feedback.Although readers generally found the MT+PE acceptable, several ethical reasons were raised (e.g., whether the author is willing to sacrifice the quality of the translations of their work for wider dissemination).In addition,Moorkens et al. (2018) called in the help of professional literary translators, the translation of a novel chapter (English>Catalan); one version was translated from scratch, the other two translations were post-edited using NMT and SMT engines.They concluded that participants preferred to translate from scratch because it was considered more creative than PE, even if it took longer.Furthermore, Matusov (2019) concluded that customizing NMT systems (in this case, with literary texts translated from English into Russian and German into English) significantly improved the quality of translation compared to general domain NMT systems.Finally,Webster et al. (2020)    assessed the use of Google Translate and DeepL in translating classic English novels into Dutch.The study revealed that NMT output contained errors and tacked creativity.
in a hotel bar in the city center with an associate of his own, Paddy Norton, the Chairman of Winterland Properties.
Data collection was carried out by means of two model questionnaires (Model 1, filled by Group A and Model 2, filled by Group B) designed in Google Forms, which are divided into three stages: 1) pre-translation questions, 2) neologism translation practice, and 3) post-translation questions.All the questionnaires have in common the pre-translation demographics-type questions, related to their academic year, level of proficiency in English, translation experience and the translation tools or resources they normally use.
Neologisms by Group A, Module 2 In group A, there is a variety of translations, as most of the neologisms have been provided by 10% of the class, i.e., one per student.Only one proposal was made by 20% of the class (Wintertide), two by 30% (Neverwinter and Winterland), and finally one by 40% (Winterlock).Thus, the participants tent to keep the neologism as a loanword, sometimes more successfully (Winterland) than others (Wintersweet).The main technique they used was composition (Dulcinver [Sweet + winter], Cerrainvierno [Lock + frozen]).In other examples, the students came up with some very original formal neologisms, as they relied on the connotative meaning of the lexical basis of the neologism itself (Sininvierno [No + winter], Candelado [Lock + frozen], Hiberlandia [Winter + land]).
Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21 Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21 239 works) and British Google Books (34 billion words from British English works), spanning both from the 16 th to the 21 st century.From both corpora, we looked for examples only in contemporary works from 1951 to 2021, since we wanted to include from the most current indexed works (2021) up to 70 years ago.A specific search strings of the word Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21 on two sides by the Wild and by the frozen wastes of Winterlock to the west, is all but a forsaken land.
ContextThen, at Wintertide, the Hill Raiders attack the village of Cirrus Cove.

Table 2 .
Demography of students per module and group Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21 difficulties, feasibility of neologism translation tools, and the tools and resources used, highlighting the adoption of technologies.The students from Group 2 are specifically inquired whether they had used any MT system.

Table 4 .
Neologisms in Google Translate

Table 5 .
Neologisms in Phrase TMSTables 1-5 show that the performance of the three NMT systems is noticeably different.In general, these NMT systems provided more translation options when the neologism was isolated (without context), as is the case of DeepL and Google Translate.In addition, Moreover, false senses have been identified in other examples such as (Wintersweet < Agridulce de invierno [mistaken by bittersweet] or Neverwinter < Nunca jamás [mistaken by Neverland]).However, among the three NMT systems, DeepL is the one that has best succeeded in translating neologisms.For instance, this is the case of Wintersweet < Dulce invierno [Sweet winter], which fits the children's context of the novel, or Wintertide < invernada[wintering], as it is a seasonal period.By contrast, Google Translate did not translate any neologism in context.Instead, it seems to be more effective when the neologism was used without context.Nevertheless, its first output sometimes reflected a literal translation (Neverwinter < Nunca invierno [Never winter]) or a translation that mixed both English and Spanish (Wintersweet < Sweet de invierno ["Sweet" of winter] or Neverwinter < Nunca winter [Never "winter"]), resulting in a poor output.In other cases, the MT system simply translated the base meaning of the neologism as in Winterlock < Bloqueo[Blocking]or Wintertide < invernal these NMT systems tent not to translate the neologism, since they did not interpret it as a new word requiring translation, either with or without context.DeepL and Google Translate produced a more varied output than Phrase TMS, although DeepL provided the most diverse translations by far.DeepL tent to keep the neologism as a loanword.However, this NMT system's output also provided either a literal translation of both the base words that form the neologism (Wintersweet < Dulce invierno [Sweet winter] or Winterlock < Bloqueo Invernal [Winter Blockade]); or explanatory when the neologism appeared in isolation (Winterland < Winterland (País de invierno) ["Winterland" (Winter country)] or Winterlock < Winterlock (candado de invierno) ["Winterlock" (winter lock)].[winter/wintry], the latter more successfully.

Table 6
This section entails the HT results, divided by Module (1 and 2) and by Group (A and B).The results on the adoption of technologies in the pre-translation (Table7) and posttranslation questions are addressed, as well as the neological translation performance in the sections below.Computer-assisted translation tools(Trados Studio, MemoQ, Omega T, etc.)

Table 7 .
Adoption of technologies by Module 1 and Module 2

Table 8 .
Neologisms by Group A, Module 1 In the practice of neologisms, most of the students have produced unique solutions and there were very few repetitions of terms.4.8% of the class, i.e., each student, provided a different proposal.Five proposals were given by 9.6% of the class, namely Eternoestío [Forever + winter], Eternoestío [Forever + winter], Neverwinter, Invernalia [Winter + land/ Winterfell], and Winterlock.Finally, three proposal were given by 14.4% of the class, specifically Dulcinvierno [Sweet + winter], Marea invernal [Winter tide] and Villainvierno [Village + winter].
As for the translation techniques used, most of the participants in this group chose to create some correspondence with another formal neologism by composition, which conveys the same meaning as the original one (Dulcinvierno [Sweet + winter], Marhelada [Sea + freezing] or cierregelido [Lock + freezing]).In other examples, they kept the neologism as a loanword or used a more descriptive technique (La ciudad que no conoce el invierno [The city that knows no winter] or Periodo invernal [Winter period]).Other students came up with neologisms by suffixation (Invierne [Winter + current]), while others went a step further by using their connotative meaning (Glacidulce [Glacial + sweet], Eternoestío[Forever + winter]or Invernapolis[Winter + polis]

Table 9 .
Neologisms by Group B, Module 1 In group B, most neologisms have been proposed by 6.3% of the class, i.e., one per student.3 proposals have been provided by 9.6% of the class (i.e., Flor de invierno [Winter flower], Madreselva [Honeysuckle], and Marea invernal [Winter tide]).Finally, a proposal was made by 18.8% of the class (Nuncainvierno [Never + winter] and another by 31.3% (Neverwinter).It has been observed creative output.For instance, translation by composition on the lexical was frequent (Invernalado [Winter + frozen], Cierrinvierno [Lock + winter] Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21

Table 11 .
Neologisms by Group B, Module 2 In group B, most of the proposals have been made by 16.7% of the class, i.e., one per student.Only two proposals were made by 33.4% of the class (Winterland and Iverlandia [Winter + land]) and one by 50% (Noyvern [No + winter].Therefore, a contrary tendency to keep the neologism as a loanword is observed.The students preferred to use techniques such as composition by combining the two lexical bases of the original neologism (Dulcierno [Sweet + winter]), but they also have some rather elaborated creations which have focused on both the denotative and connotative meaning of the source neologism (Hivernunca [Winter + never], Nuncanieva [Never + snow] or Parafrío [Stop + cold]), leading to a successful translation.Other neologisms were more descriptive (A la caída del invierno [At the fall of winter] or Flores de invierno [Winter flowers]).There were also false meanings, for instance, Wintertide (Marea Gélida [Freezing Tide] or Mareainvernal [Tide + winter]), or certain neologisms whose spelling differs from the lexical combinations of the target language (Skade or Noyvern [No + winter]).
Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21 Laura Noriega-Santiáñez / Gloria Corpas Pastor Machine vs Human Translation of Formal Neologisms in Literature: Exploring E-tools and Creativity in Students Revista Tradumàtica 2023, Núm.21