ISSN: 1579-9794
Hikma 23 (Número especial I) (2024), 1 - 32
Machine Translation and Tourism Discourse: A Spanish-
English-French Study on the Localisation of the Websites
of the Top Tourist Attractions in Spain
Traducción automática en el ámbito turístico: un estudio
español-inglés-francés sobre la localización web de las
principales atracciones turísticas en España
CARMEN MORENO-ROMERO
morenoromero@ugr.es
Universidad de Granada
ANTONIO HERMÁN-CARVAJAL
antoniohermanc@gmail.com
Universidad de Granada
Fecha de recepción: 19/02/2024
Fecha de aceptación: 27/09/2024
Abstract: In the last few decades, tourism has consistently played a
significant role in the Spanish economy. Spain, one of the world's most
popular tourist destinations, received 9.6 million international tourists in
September 2024, according to the Spanish National Institute of Statistics (INE,
2024a). Due to the significance of the tourism sector, it is particularly important
that Spain’s most visited tourist attractions offer high-quality information in
multiple languages on their websites to ensure that as many people as
possible look up the information contained therein, as businesses and other
stakeholders in the tourism sector can benefit significantly from website
localisation. For this study, the linguistic adequacy of the websites of Spain’s
top 20 tourist attractions as well as their official localised versions in English
and French were analysed by taking into account a series of parameters
related to best practices in web localisation (Olvera-Lobo and Castillo-
Rodríguez, 2019; Tercedor Sánchez, 2005). Furthermore, the official localised
websites in English and French were compared with the translation proposals
of DeepL and Google Translate to assess the quality of the machine-
translated tourism-themed content. The results obtained show poor quality in
terms of localisation and linguistic adequacy for the official Spanish, English
and French versions of the analysed websites. Regarding the overall
assessment of the machine-translated content, DeepL performed better than
Google Translate and outperformed the official websites localised into English
in terms of linguistic quality.
2 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Keywords: Web localisation, Translation, Translation for tourism, Machine
translation, Machine translation quality assessment
Resumen: En las últimas décadas, el turismo ha desempeñado un papel
importante en la economía española. España, uno de los destinos turísticos
más populares del mundo, recibió 9,6 millones de turistas internacionales en
septiembre de 2024 según el Instituto Nacional de Estadística (INE, 2024a).
Debido a la importancia del sector turístico, es esencial que las atracciones
turísticas más visitadas de España ofrezcan información de alta calidad en
varios idiomas desde sus sitios web para garantizar que el mayor número
posible de personas consulte la información contenida en ellos. Tanto las
empresas como otras entidades del sector turístico pueden beneficiarse
considerablemente de la localización de sitios web. Para este estudio, se
analizó la adecuación lingüística de los sitios web en español de las 20
principales atracciones turísticas de España, así como sus versiones oficiales
localizadas al inglés y francés teniendo en cuenta una serie de parámetros
relacionados con buenas prácticas en localización web (Olvera-Lobo y
Castillo-Rodríguez, 2019; Tercedor Sánchez, 2005). Además, se compararon
los sitios web oficiales localizados en inglés y francés con las propuestas de
traducción de DeepL y Google Translate para evaluar la calidad de dichas
herramientas al traducir automáticamente contenido de temática turística. Los
resultados obtenidos muestran una calidad deficiente en lo referido a la
localización y corrección lingüística en las versiones oficiales en español,
inglés y francés de los sitios web analizados. En cuanto a la evaluación de la
calidad de las traducciones automáticas, DeepL arrojó mejores resultados
que Google Translate y superó en calidad lingüística a los sitios web oficiales
localizados al inglés.
Palabras clave: Localización web, Traducción, Traducción turística,
Traducción automática, Evaluación de la calidad de traducción automática
INTRODUCTION
Spain consistently ranks among the world's most popular destinations
as Spain’s tourist attractions are visited by millions of international tourists
every year. Among the tourists who visit Spain the most, those from the United
Kingdom, France and Germany consistently occupy the top positions (INE,
2024b). In 2022, tourism activity amounted to €155,496 million, accounting for
11.6% of Spain's gross domestic product (INE, 2023). Due to the key role
played by the tourism sector in the Spanish economy, it is essential that tourist
attractions offer up-to-date and trustworthy information in several languages
to reach as many potential visitors as possible.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 3
Hikma 23 (Número especial I) (2024), 1 - 32
According to Suau Jiménez (2012), tourism communication
encompasses two main modalities: professional communication among
experts and communication between professionals and users. The
communication between professionals and users takes place either directly
through oral interactions or indirectly through written interactions, such as
guides, brochures, specialised press, advertisements, or websites.
From researching destinations to booking accommodation and
activities, tourists increasingly rely on websites for information and assistance.
López González (2020) points out that touristic websites are aimed at
reaching foreign customers, who make up “a demanding readership that might
not travel to a destination if the website visuals and texts are not attractive
enough” (p. 63). As the tourism industry continues to thrive, businesses and
tourist attractions should aim at reinforcing their online presence. Moreover,
promotional websites are one of the most popular tools for tourism
professionals when it comes to communicating with users (Suau Jiménez,
2012, p. 145).
Agorni (2022) notes that web communication has created “web genres,”
hybrids of text, images, audio, music, and animation essential for capturing
readers’ attention (Grego, 2010; Mehler et al., 2010). Translating a website
involves considering linguistic and cultural factors to engage the specific target
audience.
Other authors have explored the use of Machine Translation (MT)
systems for translating tourism-themed content (see Fuentes-Luque and
Santamaría Urbieta, 2020; Giampieri and Harper, 2023). In their study,
Giampieri and Harper highlighted that MT performed surprisingly well with
highly descriptive and informative texts, where no issues were noted. In
addition, Fuentes-Luque and Santamaría Urbieta (2020, p. 78) also pointed
out the linguistic accuracy of MT in the tourism domain, although some cultural
issues were not correctly addressed.
Bearing all the above in mind, our research aimed to answer the
following research questions (RQs):
1) Are the websites of Spain’s top 20 tourist attractions
linguistically and culturally appropriate both in their original
Spanish version and in their localised English and French
versions?
2) How do Google Translate (GT) and DeepL perform when
translating the websites for Spain’s top 20 tourist attractions
into English and French compared to the official translations
available on these websites?
4 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
To answer our RQs, the objectives of this study were the following:
1) To assess the linguistic quality of the websites of Spain’s top
20 tourist attractions in Spanish (ES) and their official
translations into English (EN) and French (FR).
2) To analyse the quality of the machine-generated translations
using GT and DeepL for the websites of Spain’s top 20 tourist
attractions into EN and FR.
3) To compare the quality of the official EN and FR versions of the
websites of Spain’s top 20 tourist attractions and the GT and
DeepL-generated translations into EN and FR of said websites.
1. WEB LOCALISATION
According to Jiménez-Crespo (2013), web localisation is a
communicative, technological, textual, and cognitive process that adapts
interactive digital texts for audiences worldwide, beyond the original target
group. Localisation aims to enhance the comprehensibility and usability of a
product, enabling its effective use across diverse global contexts (Sin-wai,
2012). By localising a website into other languages, it is possible to expand
the horizons of the primary target audience, as well as to attract a larger
volume of tourists who will be addressed in their mother tongue.
The web genre exhibits a unique characteristic known as multimodality.
This refers to its capacity to blend various semiotic resources −such as words,
images, and sounds− within a single communicative act (Agorni, 2022). In
multimedia translation, non-linguistic elements play a key role as a support for
the text and, in many cases, as a key concept to the translation (Tercedor
Sánchez, 2005). These aspects must be considered by localisers, that is,
professional translators who possess expertise not only on the linguistic
aspects of translation but also on the cultural nuances, technical
requirements, and regional variations necessary for successful localisation.
In the localisation process, it is essential to consider the target culture.
Therefore, full communicative competence −including linguistic,
sociolinguistic, and pragmatic subcompetences− is necessary (Gutiérrez-
Artacho et al., 2019)1. Localisers are skilled in adapting content to ensure that
it resonates with the target audience and maintains its intended meaning while
accounting for language, cultural references, idiomatic expressions, date and
time formats, currency symbols, and other region-specific elements.
1 For more information on translation competence, see Kelly (2002).
Carmen Moreno-Romero y Antonio Hermán-Carvajal 5
Hikma 23 (Número especial I) (2024), 1 - 32
Occasionally, businesses and organisations overlook the global
customer base they could reach by creating localised content. However,
localisation is an extremely valuable tool for companies pursuing
internationalisation strategies (Olvera-Lobo and Castillo-Rodríguez, 2019).
Several studies indicate that users spend up to twice as long on websites
localised into their first language have a more positive attitude towards them,
and are nearly three times more likely to purchase a product (Baack and
Singh, 2007; Singh and Pereira, 2005). Lastly, Agorni (2022) points out that
“as more and more companies consider the whole world to be their market,
consumers expect every company to have at least a bilingual website” (p. 33).
Due to the pivotal role of web localisation in today’s interconnected
world, Spanish tourist attractions should offer high-quality, localised
multilingual content on their websites. Considering the culture and audience
of the target audience is crucial for effective localisation.
2. MACHINE TRANSLATION FOR WEB LOCALISATION
Among the many possibilities there are to localise web content,
businesses and other institutions can resort to MT for their localisation
process.
Over the past few years, the improvement of MT systems has resulted
from transitioning from phrase-based statistical MT systems to neural machine
translation (NMT) systems (Daems and Macken, 2019). In 2014 and 2015, the
first research publications on NMT emerged, and by 2016, academic and
industrial research teams demonstrated that NMT systems outperformed
previous approaches (Rothwell et al., 2023). Kumar et al. (2022) highlight that,
compared to the earlier statistical models, NMT models have proved their
ability to capture contextual dependency of long sentences and they have
become the “de facto standard of MT” (p. 46).
Using MT offers several advantages, such as the ability to cover a wide
range of languages. MT tools can be developed ad hoc for a specific purpose
or be accessed online free of charge. When it comes to free online MT tools,
GT and DeepL are two of the most popular options. GT is a MT tool developed
by Google and was initially launched in 2006. It primarily relied on Statistical
Machine Translation techniques (Och, 2006). In 2016, Google incorporated
neural network technology into its MT engine (Le and Schuster, 2016). As for
DeepL, this NMT tool was launched in 2017 (DeepL, n.d.). These tools allow
users to translate plain text, full documents or websites within seconds.
6 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Nevertheless, several aspects need to be considered when making use
of MT. While MT has improved significantly in the last decade, these tools still
have limitations and can produce inaccurate translations, which could
potentially lead to miscommunication issues. According to the International
Organisation for Standardisation (ISO: 18587:2017, p. V):
There is no MT system with an output which can be qualified as
equal to the output of human translation and, therefore, the final
quality of the translation output still depends on human translators
and, for this purpose, their competence in post-editing.
In this regard, it is advisable to rely on human translators whenever
possible.
3. HUMAN OUTPUT AND MACHINE TRANSLATION QUALITY ASSESSMENT
MT, MT Post-Editing (MTPE), and MT quality assessment are areas of
great interest for the industry as well as for academia (see Briva-Iglesias,
2021; Kenny, 2022; O’Brien, 2022; Rico Pérez, 2024, inter alia).
One of the techniques used to assess the quality of MT output is human
evaluation. Human evaluation of MT involves ranking exercises to compare
system preferences, fluency and adequacy measures using evaluators’
scores, and MTPE evaluation to assess temporal, technical, and cognitive
effort (Rothwell et al., 2023).
In addition, there are well-established metrics for carrying out human
quality translation assessment, such as the DQF-MQM Error Typology. This
framework serves as a resource for assessing both human translation quality
as well as MT output quality (TAUS, n.d.). It provides a systematic approach
for categorising and assessing translation quality and was developed by the
Translation Automation User Society (TAUS), a translation industry think tank.
The DQF-MQM framework is the result of combining the Dynamic Quality
Framework (DQF) and the Multidimensional Quality Metrics (MQM) error
typology.
Lommel (2018) states that after the Localisation Industry Standards
Association (LISA) ceased operations, two groups took on active roles in
translation quality assessment: i) TAUS, and ii) the EU-funded QT LaunchPad
project, led by the German Research Centre for Artificial Intelligence (DFKI).
DQF was the error typology proposed by TAUS. According to Castilho et al.
(2018), one of the noteworthy points of the DQF is that rather than dealing
with problems after the translation process, quality issues should be
considered before the actual translation process begins.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 7
Hikma 23 (Número especial I) (2024), 1 - 32
As for the MQM, this error typology was developed within the frame of
QTLaunchPad project to address the shortcomings of previous quality
evaluation (Lommel et al., 2014). Castilho et al. (2018) point out that MQM
can apply to both professional translations and MT output. In other words, the
metric is designed to assess the quality of the translation product, regardless
of the method used to generate the target text.
In 2014, TAUS and DFKI began the harmonisation of DQF and MQM
to bridge the gap between the definitions and specifications of the two models
(Görög, 2014). The DQF and MQM are the latest large-scale initiatives aimed
at standardising translation quality assessment, and they are especially
relevant as they bring together approaches that initially developed
independently in both research and industry (Castilho et al., 2018).
The resulting harmonised framework consists of seven high-level error
types, each further subdivided into more specific error categories as shown in
Table 1:
High-level error type
Granular error type
1) Accuracy
Addition
Omission
Mistranslation
Over-translation
Under-translation
Untranslated text
Improper exact TM match
2) Fluency
Punctuation
Spelling
Grammar
Grammatical register
Inconsistency
Link/cross-reference
Character encoding
3) Terminology
Inconsistent with term base
Inconsistent use of terminology
4) Style
Awkward
Company style
Inconsistent style
Third-party style
Unidiomatic
5) Design
Length
Local formatting
Markup
Missing text
Truncation/text expansion
8 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
6) Locale convention
Address format
Date format
Currency format
Measurement format
Shortcut key
Telephone format
7) Verity Culture-specific reference
Table 1. DQF-MQM error typology
Source. Summarised from TAUS DQF-MQM error typology by the authors.
4. METHODOLOGY
In this study, a methodology consisting of several phases was
designed: 1) selection of Spain’s top 20 tourist attractions, 2) content and
sample selection, 3) analysis of the original ES version, 4) analysis of the pre-
existing localised EN and FR versions, and 5) analysis of the GT and DeepL-
generated translations. Each of the phases is detailed below:
4.1. Selection of Spain’s top 20 tourist attractions
The tourist attractions chosen for this study were selected based on the
article written by Cynthia M.R. and published in the Spanish economic journal
Expansión in March 2022. At the time this study was conducted, the article
provided the latest details on Spain’s most popular tourist attractions.
Table 2 presents the chosen tourist attractions, the code assigned to
each of them and a shortened link redirecting to the official website of each
attraction:
Code
Tourist attraction2
Website link
1SF
Sagrada Familia (Barcelona)
https://t.ly/vB-Y
2ALH
Alhambra (Granada)
https://t.ly/n32V
3MEZ
Mosque-Cathedral Monumental Site of Cordoba
https://t.ly/AISF
4CST
Cathedral of Santiago (Santiago de Compostela)
https://t.ly/CYdo-
5CBU
Cathedral of Burgos
https://t.ly/azHbU
2 The English names of the monuments are those which appear on each of their English-language
websites. In cases where there is no English version of the monument's website, a literal
translation is provided.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 9
Hikma 23 (Número especial I) (2024), 1 - 32
6ASG
Alcazar of Segovia
https://t.ly/A3A92
7CPM
Cathedral of Mallorca
https://t.ly/STJW
8ZGZ
Cathedral-Basilica of Nuestra Señora del Pilar
(Zaragoza)
https://t.ly/2iD_
9MER
Roman Theatre (Mérida)
https://t.ly/ls6V
10GIR
Giralda (Seville)
https://t.ly/TLU7
11CMI
La Pedrera-Casa Milà (Barcelona)
https://t.ly/HgeGO
12CAC
La Ciutat de les Arts i les Ciències (Valencia)
https://t.ly/3GUu
13MRS
Museo Reina Sofía (Madrid)
https://t.ly/KaF_
14RAS
Real Alcazar (Seville)
https://t.ly/Io1a
15GUG
Guggenheim Museum (Bilbao)
https://t.ly/pWsf
16MLG
Alcazaba of Malaga
https://t.ly/VOQO
17MP
Monasterio de Piedra (Nuévalos, Zaragoza)
https://t.ly/MBgN
18MES
Royal Site of San Lorenzo de El Escorial (Madrid)
https://t.ly/Qemt
19MPR
Museo del Prado (Madrid)
https://t.ly/SMA5O
20PR
Royal Palace of Madrid
https://t.ly/AZNN
Table 2. The 20 tourist attractions selected, accompanied by their assigned
code and the link to their website
Source. Elaborated by the authors
4.2. Content and sample selection
Web pages providing a general overview of each tourist attraction were
selected. A sample of approximately 350 words was extracted from the ES
website of each tourist attraction. In addition, the equivalent official
translations into EN and FR were selected, so as to compare the same
sections in all of the three official versions of the content.
When the selected web pages did not meet the 350-word threshold for
the sample, other sections, such as “How to get there” or “Schedule hours”,
were also included. In exceptional cases in which these sections still did not
meet the minimum word threshold, some historical background information
available on the website was also included.
After selecting the 20 tourist attractions, the HTML files of the websites
to be analysed were downloaded to ensure access without relying on online
availability. To this end, the “Save As” option in Microsoft Edge browser was
used. Rather than copying and pasting plain text, downloading the web pages
allowed the authors to consider parameters such as colour choice and the use
10 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
of icons when analysing the degree of localisation of each website. However,
in order to assess linguistic adequacy, a single file containing plain text was
generated for each tourist attraction website. The resulting document
comprised the sections containing the segments to be analysed at a later
stage.
4.3. Analysis of the original ES version
With the aim of assessing the linguistic quality of the content, 350 words
from the ES version of each website were selected. Thus, approximately
7,000 words were analysed in total.
Errors in the original ES version were annotated by means of the DQF-
MQM error typology. Even though the typology is meant to be used for
assessing translations, several error types described within such framework
can be used to assess monolingual content, i.e. punctuation, spelling,
grammar, grammatical register, inconsistency, link/cross-reference character
encoding, awkward, inconsistent style, unidiomatic, address format, date
format, currency format, measurement format, shortcut key, telephone format,
and culture-specific reference.
Firstly, each author carried out the analysis of the ES versions
individually. Afterwards, both authors compared the errors which had been
identified individually to eliminate any errors mistakenly categorised or which
could have been identified based on a personal preference.
4.4. Analysis of the pre-existing localised EN and FR versions
In the first phase of the analysis and prior to the linguistic and cultural
analysis, the degree of localisation of each website was examined by means
of the following questions:
1) How many languages have the websites been localised into?
2) Is the language menu presented correctly, i.e. avoiding the use
of flags to refer to a language?
3) Are the errors identified in the localised version a consequence
of an error in the ES version?
4) Are the websites fully localised or do they contain any
untranslated section?
5) Are there any errors in the localised websites? (i.e.
untranslated content in Spanish present in the EN or FR
versions)
6) Are there any errors in the cultural references present in the
localised versions?
Carmen Moreno-Romero y Antonio Hermán-Carvajal 11
Hikma 23 (Número especial I) (2024), 1 - 32
To answer these questions, the authors drew on the ideas proposed by
Olvera-Lobo and Castillo-Rodríguez (2019) and Tercedor Sánchez (2005),
who established various aspects to consider when assessing website
localisation. According to Olvera-Lobo and Castillo-Rodríguez (2019), these
aspects could be classified as linguistic, cultural, and technical issues.
Tercedor Sánchez (2005) organised them into categories related to text,
images, icons, graphical elements, technical features, and cognitive factors.
The answers to these six questions allowed for a first overall analysis showing
the amount of 1) partial localisations, 2) unlocalised websites, 3) spelling and
punctuation errors, and 4) localisation errors. After this first overall analysis,
an in-depth analysis of 5) linguistic and translation errors was conducted.
Regarding the analysis of linguistic and translations errors, the 350-
word samples of the pre-existing EN and FR versions of the websites were
considered reference translations. Such reference translations in EN and FR
were segmented and aligned at sentence-level with the original segments in
Spanish. Each author then compared the EN and FR reference translations
with the original ES version of the websites with the aim of assessing linguistic
quality as well as to examine cultural adequacy to potential readers (i.e.
converting units into the imperial system for US readers for the English locale,
adding Spain’s telephone prefix in the EN and FR localised telephone
numbers, text in multimedia content such as images and video not left
untranslated, etc.). Lastly, the identified errors were annotated using the DQF-
MQM error typology.
4.5. Analysis of the GT and DeepL-generated content
After analysing the ES original versions as well as the EN and FR pre-
existing localised websites, an analysis of the GT and DeepL-generated
translations was carried out. In order to compare the quality of the pre-existing
localised websites with that of the MT-generated translations, GT and DeepL
were asked to translate the same excerpts that were already available in EN
and FR. As previously done, errors were annotated in accordance with the
DQF-MQM error typology. The results from this annotation were used to
compare the quality of the official EN and FR versions. Approximately 15,000
machine-translated words were analysed.
5. RESULTS
After concluding all the stages of the analysis, the following data were
obtained. Firstly, an overview of the degree of localisation of the different
websites is presented:
12 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
- 15% of the analysed websites did not offer a localised version
in any other language apart from ES.
- The most translated website was that of the Guggenheim
Museum, offering its content in 12 languages including ES.
- 50% of the websites were translated into languages other than
EN or FR.
Table 3 shows the percentage of monuments offering their websites in
languages other than ES, EN, and FR:
Language
Percentage
Available in
Catalan 30%
La Sagrada Familia, Palma de Mallorca
Cathedral, La Pedrera-
Casa Milà, Ciutat de les
Arts i les Ciències, Museo Reina Sofía,
Guggenheim Museum
Galician 15%
Santiago de Compostela Cathedral, Museo Reina
Sofía, Guggenheim Museum
Italian 15%
La Pedrera-Casa Milà, Alcazaba of Málaga,
Guggenheim Museum
Basque
10%
Museo Reina Sofía, Guggenheim Museum
Chinese 10%
La Pedrera-Casa Milà, Ciutat de les Arts i les
Ciències
German
10%
Alcazaba of Málaga, Guggenheim Museum
Korean
10%
La Pedrera-Casa Milà, Guggenheim Museum
Portuguese
10%
Roman Theatre of Mérida, Guggenheim Museum
Arabic
5%
Alhambra
Japanese
5%
Guggenheim Museum
Russian
5%
Guggenheim Museum
Table 3. Percentage of monuments offering their websites in languages other
than Spanish, English, and French
Source. Elaborated by the authors
Carmen Moreno-Romero y Antonio Hermán-Carvajal 13
Hikma 23 (Número especial I) (2024), 1 - 32
The data reveal significant variations in the extent to which the websites
have been adapted for the EN and FR locales. Approximately 90% of the
websites were translated into EN, indicating a significant investment in
catering for the English-speaking audience. This may suggest that the website
content managers recognised the importance of reaching a broader English-
speaking user base.
However, the data show a less extensive effort when localising the
websites for the French-speaking audience. Only 30% of the websites were
translated into FR, which might indicate a lower level of priority placed on
catering to this language. This discrepancy between the EN and FR versions
could be attributed to various factors, such as the perceived importance of
each language market or resource limitations faced by the website content
managers.
The following paragraphs provide a classification of the identified errors
and examples of each error category after the first overall analysis:
1) Partial localisations. Six websites did not include the same
information as the reference website in ES in their localised versions into EN
and/or FR. This error was identified in the websites of Santiago de
Compostela Cathedral, the Basílica del Pilar in Zaragoza, Seville Cathedral,
La Pedrera-Casa Milà in Barcelona, La Ciutat de les Arts i les Ciències, and
the Real Alcázar in Seville, which account for 30% of the analysed websites.
2) Unlocalised websites. 15% of the websites had not been localised
into other languages, thus simply providing a version in ES. Although this is
not an error, unlocalised websites make it difficult for non-Spanish-speaking
tourists to access information. The following three monuments did not include
an EN version of their websites: the Alcázar of Segovia, the Giralda3 and
Burgos Cathedral. Lastly, the following fourteen attractions did not provide a
FR version of their website: Sagrada Familia, the Alhambra, Santiago de
Compostela Cathedral, Burgos Cathedral, the Alcázar of Segovia, Palma de
Mallorca Cathedral, the Basílica del Pilar in Zaragoza, the Roman Theatre of
Mérida, the Giralda, La Ciutat de les Arts i les Ciències, Museo Reina Sofía,
the Royal Site of San Lorenzo de El Escorial, Museo del Prado and the Royal
Palace (Madrid).
3 In the early stages of this research, the Giralda website offered an English version. However,
during the final stages of this work, the website discontinued its localised English version. At the
time of this article's submission, the content was available only in Spanish.
14 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
3) Spelling and punctuation errors. Among the websites analysed, 30%
exhibited punctuation errors, including instances where the symbol “€” was
incorrectly placed after the numerical value in localised content in English.
4) Localisation errors. Localisation errors encompass instances where
content, such as weights and measures, is not appropriately adapted to the
target language and culture. Additionally, leaving text untranslated within
images or videos is also considered a localisation error.
Six websites used flags as icons for the language switching menu. The
use of flags on websites for switching between languages has long been
considered inadvisable (Tercedor Sánchez, 2005). Associating a flag of a
country with a language sets aside other countries in which that language is
also spoken.
Instead, using codes for the representation of names of languages is
preferred, such as those suggested in ISO 639:2023, which comprises
language code elements of one to three language identifiers. In total, 75% of
the analysed websites had localisation errors, and 30% of them had resorted
to flags for the language switching menu.
5) Linguistic and translation errors. Linguistic and translation errors
were annotated through TAUS DQF-MQM error typology. The results are
shown in the following sections.
5.1. Results obtained from the analysis of the ES version of the websites
The samples of the original ES versions of the websites had 37 errors.
The analysis of the content of the websites in ES revealed that all the
examined websites had at least one spelling and/or punctuation error. This
suggests that there may have been some oversight or lack of attention to detail
during the creation or maintenance of these websites.
Table 4 includes an example for each identified error, along with the
corresponding text code:
Carmen Moreno-Romero y Antonio Hermán-Carvajal 15
Hikma 23 (Número especial I) (2024), 1 - 32
Identified errors
Justification
18 punctuation errors
la Catedral,
recursos de accesibilidad,
In Spanish, it is incorrect
to place a comma
between the subject and
the predicate.
14 spelling errors
arruinadas desde antiguo
accesos, sólo quedaban
en pié los siete cuerpos
Neither the adverb “solo”
nor the noun “pie” in
Spanish carry accent
marks.
3 inconsistencies
asistentes que adopten
una conducta respetuosa
y vistan decorosamente.
Si venís en grupo de más
de 25 personas [...].
The sentence begins
addressing the reader in a
formal tone by using the
form “(ustedes) adopten”
but then shifts to the
informal “(vosotros)
venís”.
2 errors classified as
“others”
entrada a la Catedral a y
The preposition “a” before
the conjunction “y” is
superfluous and could be
a typo.
Table 4. Number and type of errors present in the ES version sample extracted
from the 20 websites
Source. Elaborated by the authors
As can be seen, the errors are mostly due to a presumed lack of
attention or unawareness of the orthographic rules and conventions of the
Spanish language. The inconsistencies and errors that fall into the “others”
category suggest that the content might not have been proofread before
publishing.
16 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
5.2. Results obtained from the analysis of the EN version of the websites
Figure 1 presents the errors identified in the localised EN version as
well as in the EN machine-translated content:
Figure 1. Errors in the localised EN versions vs. errors in the MT-generated
versions into EN by Google Translate and DeepL
Source: Elaborated by the authors
In general, the pre-existing EN versions of the websites had a great
variety of errors, but the overall number of errors was significantly lower than
in the EN GT-translated content and only slightly higher than in the EN DeepL-
translated version. Both MT systems exhibited notably superior performance
in punctuation, as GT and DeepL improved the punctuation in the translations
they produced, with DeepL producing texts with almost no punctuation errors.
It is especially noteworthy that the main errors detected in the MT-generated
translations fell into the categories of “mistranslation,” “unidiomatic,” “culture-
specific reference,” and notably, “awkward.”
Carmen Moreno-Romero y Antonio Hermán-Carvajal 17
Hikma 23 (Número especial I) (2024), 1 - 32
The following sections present the results obtained from the analysis of
the EN versions. Firstly, the results of the analysis of the pre-existing localised
EN versions are provided, followed by the results of the analysis of the MT-
generated contents.
5.2.1. Results obtained from the official localised version of the websites
The sample of the localised EN versions of the websites contained 39
errors. An example for each error category is provided in Table 5:
Identified errors
Justification
6 mistranslations
hasta el siglo XIV
14RAS, EN: The Alcazar
The translation does not
accurately represent the
century mentioned in the
ES source text.
6 punctuation errors 8ZGZ, EN: 9.00 €
The euro sign (€) should
precede the numerical
value.
5 grammar errors
DAYS
The noun is in plural form
when it should be
singular.
3 terminology
inconsistencies
Salvador. [...] Cathedral
The name of the
monument was not
translated consistently.
3 spelling errors
“Muslim” was spelled
incorrectly.
2 inconsistencies
different venues: the Main
Sabatini Building and the
Nouvel Building, and the
Palacio de Velázquez and
Unlike the other sites
listed, “Palacio de
Velázquez” does not
include the article “the”.
2 unidiomatic phrases
LLEGAR.
The intended meaning is
not accurately conveyed
by the literal translation.
18 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
2 errors concerning local
formatting
al Gótico.
The source text does not
capitalise the segment,
thus capitalisation is not
required in the English
version.
2 omissions
fue ciudad
1870.
2ALH, EN: The Alhambra
until its declaration as a
The English translation
only conveyed the
meaning of “Monument”,
without further
explanation.
2 locale convention
errors
15GUG, EN: With 24,000
space [...].
In English grammar, the
common practice for
separating thousands is
to use a comma as a
thousands separator.
Therefore, the format
“9.000” is incorrect and
directly transposed from
the Spanish language.
1 error related to missing
text
partir de las 15:00 (cierre
taquillas a las 14:00).
20PR, EN: December 24:
The Spanish source text
provides information
about the closing time of
the ticket offices, which
was not included in the
English translation.
1 addition
PASEO DE GRACIA era
la avenida más
[…].
11CMI, EN: In the year
1900, Passeig de Gràcia
The English version
contains “the year,” which
is potentially unnecessary
since the context already
implies that 1900 refers to
a year.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 19
Hikma 23 (Número especial I) (2024), 1 - 32
1 date-and-time-format
error
7CPM, EN: The visits are
unguided and the opening
Friday 10:00 to 16:30 and
Saturday 10:00 to 13:30.
In informal settings, the
12-
hour format is more
commonly used in
English. In American
English, the 12-
hour is
also much more used
than the 24-hour format.
1 culture-specific
reference error
[...].
2ALH, EN: With the
revolutions of 1868
disconnected
“La Revolución de 1868”
is known in English as
“The Glorious
Revolution.”
1 under-translation
una altura de 20 metros y
km/h.
Botafumeiro
from a height of 20 metres
and can pick up great
The Spanish text
conveyed the speed the
censer reaches.
1 terminology error
[...].
disconnected
The verb “disconnected”
does not accurately
convey the intended
meaning of the source
text.
Table 5. Number and type of errors present in the localised English version of
the websites
Source: Elaborated by the authors
The number of errors (39) is practically identical to those identified in
the original ES version (37). However, not all the websites had an EN version,
resulting in a proportionately higher number of errors in the EN versions
compared to the original ES websites. As for the identified errors, they are
similar in nature to those identified in the original ES version.
20 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Errors such as mistranslations, under-translations, omissions, and
those related to culture might suggest a lack of certain translation skills among
the teams that translated the content. In contrast, errors related to
punctuation, grammar, spelling, and locale conventions could be explained by
an insufficient EN level or a lack of thorough proofreading.
5.2.2. Results obtained from the MT-generated content using GT and DeepL
The sample of the EN websites translated using GT contained 93
errors, while the sample translated using DeepL had 36. Table 6 provides
some translation proposals by GT and DeepL into English:
Original segment in
Spanish
Translated using
Google Translate
Translated using DeepL
1) 16MLG: Ofrecer una
visión amplia y global de
la historia de España de
una forma amena y
entretenida, es el
ambicioso objetivo de
esta entrada.
Offering a broad and
global vision of the history
of Spain in a pleasant and
entertaining way, is the
ambitious objective of this
entry.
To offer a broad and
global vision of the history
of Spain in a pleasant and
entertaining way is the
ambitious aim of this
entry.
2) 15GUG: El Museo está
rodeado de atractivos
paseos y plazas en una
zona de reciente
urbanización, superado
su pasado industrial.
The Museum is
surrounded by attractive
promenades and squares
in a recently urbanized
area,
its industrial past
gone.
The Museum is
surrounded by attractive
promenades and squares
in an area of recent
urbanisation, having
overcome its industrial
past.
3) 20PR: 1 de enero:
cerrado en jornada
completa
6 de enero: cerrado en
jornada completa
1 de mayo: cerrado en
jornada completa
January 1: closed full time
January 6: closed for the
whole day
May 1: closed full time
1 January: closed for the
full day
6 January: closed for the
full day
1 May: closed for the full
day
Table 6. Examples of some translation proposals using Google Translate and
DeepL into English
Source. Elaborated by the authors
As can be seen, DeepL outperforms GT in translation quality. GT
frequently produces literal translations, as seen in its replication of elements
from the original ES sentence, such as the comma between subject and
predicate in the 16MLG fragment. In contrast, DeepL takes a more natural
approach to translation.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 21
Hikma 23 (Número especial I) (2024), 1 - 32
Many of the identified errors fell under the “awkward” category, where
numerous instances of ES structures were translated literally into EN. One
example can be seen in the 15GUG fragment presented in Table 6. GT
translates superado su pasado industrial as “its industrial past gone”, which
appears awkward, especially when compared to DeepL’s “having overcome
its industrial”, a potentially more natural way to express the original ES
message in EN. Another example (not shown in Table 6) is the sentence el
horario de entrada será de lunes a viernes de 10:00h a 16:30h. This sentence
is built using the future tense of the verb in ES. GT produced the following
literal translation: “the entrance hours will be from Monday to Friday from
10:00 a.m. to 4:30 p.m.”, which resulted in awkward phrasing. In contrast,
DeepL produced a more natural translation in EN by using the present tense
and suggested the following sentence: “the entrance hours are from Monday
to Friday from 10:00h to 16:30h.” Lastly, GT also produced an inconsistency
when translating the opening hours of the Royal Palace in Madrid, as shown
in the third example of Table 6.
5.3 Results obtained from the analysis of the FR version of the websites
Figure 2 presents the number of errors identified in the localised FR
versions as well as in the MT-generated content into FR. The results show the
error categories that had at least one instance:
Figure 2. Errors in the localised FR versions vs. errors in the FR MT-generated
versions by GT and DeepL
Source. Elaborated by the authors
22 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
As can be seen, the MT-generated texts into FR featured more errors
than the pre-existing FR versions. MT performed significantly worse in issues
related to terminology, grammar and punctuation, while only outperformed the
pre-existing FR version in spelling and adequacy to locale conventions.
The following sections provide more detailed results from the analysis
of the FR versions. Firstly, the results of the analysis of the pre-existing
localised FR versions are provided, followed by the results of the analysis of
the MT-generated contents.
5.3.1 Results obtained from the official localised FR version of the websites
The localised FR content from the analysed sample of all websites
contained six errors. It is important to note that only 30% of the websites
provided a localised FR version, which explains the lower absolute number of
errors compared to the ES and EN versions. An example for each error
category is provided in Table 7:
Identified errors
Example of erroneous
segment
Justification
1 omission error
14RAS, ES: Este Palacio
de al-Mubarak, el
Bendito, fue ya el centro
de la vida oficial y literaria
de la ciudad.
14RAS, FR: Ce palais
d´al-Mubarak fut le centre
de la vie officielle et
littéraire de la ville.
The omission of “El
Bendito” is not deemed
justified.
1 punctuation error
14RAS, FR: [...] la
construction d´une
nouvelle enceinte.
Incorrect punctuation
mark used instead of the
apostrophe in the
contraction “d’une”.
3 spelling errors
15GUG, FR: Une fois sur
la place, le promeneur
accède au Vestibule en
descendant un large
escalier, un recours
architectural peu fréquent
qui résout ici avec
bonheur la différence
entre la cote de la ria du
Nervión.
The absence of a
circumflex accent in the
word “côte” changes the
intended meaning. The
error is not due to a
drafting mistake in the
original Spanish version.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 23
Hikma 23 (Número especial I) (2024), 1 - 32
1 locale convention error
15GUG, FR: Avec ses
24 000 m2 de superficie,
dont 9.000
destinés aux
expositions.
Inconsistency when
separating thousands.
Table 7. Number and type of errors present in the localised FR version of the
websites
Source. Elaborated by the authors
For the first example in Table 7, the translator might have overlooked
the significance of “El Bendito” or assumed it was unnecessary. The second
example, while a minor error that a tourist might leave unnoticed, indicates a
lapse in attention to detail. Regarding the third error, it is a spelling error where
cote, meaning ‘quotation’, ‘rating’, or ‘dimension,’ is used instead of côte,
meaning ‘shoreline,’ significantly altering the intended meaning. Again, while
a tourist might infer the intended meaning, such errors can slightly affect their
perception of the information. For the fourth example, the translator might
have followed a different locale convention or overlooked the standard French
practice when separating thousands. Therefore, certain errors identified could
result from the website localiser lacking adequate FR language proficiency,
including issues related to spelling, punctuation, or locale conventions.
5.3.2 Results obtained from the MT-generated content using GT and DeepL
When it comes to the MT-generated versions of the websites translated
into French, DeepL had 10 errors, while GT had 19. DeepL produced better
texts than GT, as it did when translating the sample into EN. Many of the errors
identified repeated patterns shown in the previous tables. Some examples of
the FR sample are included in Table 8:
Original segment in
Spanish
Translated using
Google Translate
Translated using DeepL
1) 17MP: Disfruta de la
bella arquitectura sobria y
luminosa que caracteriza
la transición del
ROMÁNICO AL GÓTICO
CISTERCIENSE.
Appréciez la belle
architecture sobre et
lumineuse qui caractérise
la transition du ROMAN
AU GOTHIQUE
CISTERCIEN.
Profitez de la belle
architecture sobre et
lumineuse qui caractérise
la transition entre le gotic
ROMAIN et le gotic
CISTERTIEN.
2) 3MEZ: Mezquita
fundacional de
Abderramán I
Mosquée fondatrice
d'Abderramán I
Mosquée fondatrice
d'Abderraman Ier
3) 15GUG: La plaza y la
entrada principal del
Museo se encuentran
enfilando la calle
Iparragirre [...].
La place et l'entrée
principale du Musée sont
situées le long de la Calle
Iparragirre [...].
La place et l'entrée
principale du musée sont
situées le long de la rue
Iparragirre [...].
24 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
(4) 15GUG: [...] el edificio
representa un hito
arquitectónico por su
audaz configuración y su
diseño innovador,
conformando un seductor
telón de fondo para el arte
que en él se exhibe.
[...] le bâtiment
représente un repère
architectural par sa
configuration audacieuse
et son design innovant,
formant un écrin
séduisant pour l'art qui y
est exposé.
[...] le bâtiment
représente un point de
repère architectural pour
sa configuration
audacieuse et son design
innovant, formant une
toile de fond séduisante
pour l'art exposé.
Table 8. Translation proposals using Google Translate and DeepL into French
Source. Elaborated by the authors
Terminology-related errors were among the most frequent ones. In the
first example shown in Table 8, the original ROMÁNICO AL GÓTICO
CISTERCIENSE was translated by GT as “ROMAN AU GOTHIQUE
CISTERCIEN” and by DeepL as “entre le GOTIC ROMAIN et le GOTIC
CISTERTIEN”. DeepL provides a completely wrong translation that includes
a spelling error (cistertien instead of cistercien) and two wrong terminology
choices (“gotic”, which is not a word in French, instead of gothique, and
romain, instead of the notably more appropriate roman).
There were also errors identified in the FR MT-generated content that
fell into the category of “untranslated text”. This was the case for the second
example in Table 8, as Abderramán I was left untranslated in GT’s version,
while DeepL omitted the accent mark and translated I for “1er”). DeepL seems
to have attempted an adaptation of the proper noun to the French language.
Another instance of an untranslated word appears in the third example, where
GT keeps the original la calle, while DeepL correctly translates it to the French
equivalent, “rue.” Since calle is not part of the name itself, DeepL’s translation
is more accurate and avoids the error made by GT.
Lastly, for the fourth example, which is not a translation error, GT
provided a more detailed expression, emphasising the location (l’art qui y est
exposé) where the art is exhibited, whereas DeepL provided a more literal
translation (pour l'art exposé).
6. DISCUSSION
The answers to the RQs will be discussed in this section. As a reminder,
this paper’s RQs were the following:
RQ1: Are the websites of Spain’s top 20 tourist attractions linguistically
and culturally appropriate both in their original Spanish version and in
their localised English and French versions?
Carmen Moreno-Romero y Antonio Hermán-Carvajal 25
Hikma 23 (Número especial I) (2024), 1 - 32
RQ2: How do GT and DeepL perform when translating the websites for
Spain’s top 20 tourist attractions into English and French compared to
the official translations available on these websites?
To address the first RQ, an overall analysis of the degree of localisation
of the websites was conducted. It was observed that the content of the 20
websites was unevenly presented; some websites stood out for their
comprehensive information, while others provided outdated or minimal
information. The number of websites offering localised versions in EN and FR
was also uneven. While 90% of the websites offered EN versions, only 30%
had FR versions. As previously mentioned, the discrepancy between the
English and French versions could be due to several factors, such as the
perceived importance of each language market or resource limitations faced
by the website content managers. Given that EN is the world’s lingua franca,
it seems evident that priority was given to translating websites into EN. This
decision might come from an assumption that tourists from non-English-
speaking countries possess some knowledge of English.
Regarding the linguistic and cultural appropriateness of the localised
versions, one of the main issues was the misuse of the language selection
menu through the use of flags, which is nowadays discouraged and should be
replaced by standardised codes for each language (Olvera-Lobo and Castillo-
Rodríguez, 2019; Tercedor Sánchez, 2005).
Regarding the linguistic quality of the original websites in ES, many
errors were identified, such as outdated spelling (accent mark for the adverb
sólo) and misspellings (the adjective éstas puertas, the noun pié). As for the
official localised versions into EN and FR, several typos, punctuation and
spelling errors, as well as literal translations and terminological
inconsistencies were identified (8ZGZ, EN: Cathedral of Salvador, Cathedral
of El Salvador).
The repeated occurrence of such errors revealed the need to improve
the quality of the analysed websites, as localisation and spelling errors could
have been avoided by experts such as localisers, proofreaders, or content
creators. Nevertheless, translation quality can still vary depending on the
specific context, content, and translator’s expertise. Overall, these findings
highlight the need for meticulous attention to detail in the original ES versions
of websites to avoid spelling and punctuation errors. Additionally, they
emphasise the importance of considering the linguistic diversity of the target
audience when localising websites. In summary, addressing RQ1, while the
original versions in ES and the official localised EN and FR versions are not
inappropriate, they do require further refinement.
26 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Regarding our second RQ, DeepL consistently outperformed GT when
it came to translation accuracy, as DeepL’s results produced more fluent
translations both in EN and FR. This was also the conclusion reached by
Hidalgo-Ternero (2021) in his comparative study of machine-translated
phraseology from ES into EN using GT and DeepL. In addition, Peña Aguilar
(2023) examined specific linguistic challenges in translating between ES and
EN. In her study, she demonstrated that DeepL outperformed other popular
MT systems, such as Bing and GT, by correcting some problems present in
the source text. Our study confirms this finding, as shown in the first example
of Table 6, where DeepL did not replicate the error present in the original
Spanish version (a comma between the subject and predicate, an error that
GT did replicate). The tendency toward literal translation by MT tools (in our
study, especially by GT) was also pointed out in Fuentes-Luque and
Santamaría Urbieta’s (2020) study on the performance of MT when translating
tourism texts in the English-Spanish combination.
While GT has made significant strides over the years, it still occasionally
fell short in terms of accuracy. Although GT can handle straightforward and
commonly used phrases well, the tool struggled when confronted with more
complex or context-dependent content, another conclusion similar to the
findings of Fuentes-Luque and Santamaría Urbieta (2020). This was the case
of the error regarding the opening hours in Table 6, in which the future tense
of the verb to be in ES to express a reiterative pattern of opening hours was
translated literally into EN. DeepL did better, as it did not mimic the ES
structure.
After using MT to render the ES content into EN and FR, we concluded
that the MT-generated content had to be revised to ensure first-class quality
standards. Human evaluation is still a valuable resource for assessing the
quality of MT-generated output, as it provides the expertise necessary to
identify and address linguistic nuances, ensuring that the final translations
meet the desired standards of accuracy and fluency. Consider example 3 in
Table 6, where GT translated a previously translated segment differently,
without any apparent justification, or example 1 in Table 8, where DeepL’s
translation included words which did not exist in FR. Overall, our results align
with those of the study conducted by Leiva Rojo (2020), where translations of
English museum texts into Spanish were assessed. Although Leiva Rojo’s
study revealed that many of the official translations analysed were “very poor”
and, in contrast, our study found acceptable official translations, both studies
show that MT did not significantly improve their quality.
Thus, answering RQ2, when the goal is to provide high-quality, verified
content, GT and DeepL can serve as complementary tools in the translation
process. However, when a translation from scratch by a professional
Carmen Moreno-Romero y Antonio Hermán-Carvajal 27
Hikma 23 (Número especial I) (2024), 1 - 32
translator is not feasible and machine translation is used instead, the
involvement of professionals with high linguistic competence remains
essential. Post-editing machine-translated content is necessary to ensure
optimal content quality that is adapted to the readers’ needs. Finally, it is
important to note that the need for human intervention is not unique to the
translation of tourist attraction websites. Other tourism-related texts translated
using MT, such as those involving culturemes in gastronomic texts, also
require human intervention to achieve high quality (Cuadrado Rey and
Navarro Brotons, 2024).
CONCLUSIONS
Our study had three objectives: 1) to assess the linguistic quality of the
websites of Spain’s top 20 tourist attractions in Spanish and their official
translations into English and French, 2) to analyse the quality of the machine-
generated translations using GT and DeepL for these websites into EN and
FR, and 3) to compare the quality of the official EN and FR versions with the
GT and DeepL-generated translations into EN and FR.
Regarding the first objective, the first conclusion is that there is a clear
need to improve both the original ES versions as well as the official localised
EN and FR versions of the websites analysed. The analysis revealed a
significant number of punctuation errors, typos and misspellings in the original
ES versions of several websites. Such errors could potentially hinder the
quality of translations produced by MT. A total of 37 errors were detected in
the original ES versions. The official localised versions of the websites had 39
errors (EN) and 6 (FR), with mistranslations, spelling errors and spelling being
some of the most usual errors. It is also important to note that 90% of the
websites offered EN versions, but only 30% had FR versions. We believe,
however, that many of the websites did not employ professional translators to
proofread the original ES versions and translate their content into EN and FR.
Regarding the second objective, there was a noticeable difference in
translation quality between GT and DeepL. DeepL (36 errors in EN, 10 in FR)
outperformed GT (93 errors in EN, 19 in FR). GT often copied existing
structures and errors from the Spanish versions of the websites, resulting in
many errors categorised as “mistranslation” and “awkward” in the EN
translations. In contrast, DeepL managed to correct some of these issues. For
FR, a common problem for both MT tools was terminology, with GT frequently
leaving text untranslated.
Thus, regarding the third objective, it was observed that the linguistic
quality of the machine-generated translations was generally not better than
the official localised versions. In fact, GT produced significantly worse
translations both in EN and FR than those already published in the official
28 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
websites. The only exception was the EN translations by DeepL, which slightly
surpassed the quality of the official localised websites for the EN language. In
any case, the intervention of highly trained human translators is essential to
provide tourists with thoroughly refined texts. This could potentially enhance
their overall experience, given the significant role websites play in their
tourism-related decision-making (López González, 2020).
Although this study has shed light on the linguistic quality of the
websites of Spain’s top tourist attractions and on the performance of GT and
DeepL when translating them into EN and FR, it is also necessary to
acknowledge its main limitation: the sample size. It should be expanded in
future studies for a more comprehensive understanding of the quality of
tourism-themed websites in Spanish and their localised versions, especially
when comparing it to the output of MT tools such as GT or DeepL.
Future directions could focus on web accessibility, assessing whether
the websites of Spain’s top 20 tourist attractions meet basic accessibility
criteria. Regarding the quality of the machine-translated content, the identified
and categorised errors could be further classified into minor or major
translation errors, depending on whether they hinder the understanding of
what was conveyed in the original content in Spanish.
ACKNOWLEDGMENTS
This study was supported by the Spanish Ministry of Science,
Innovation and Universities through the FPU University Staff Training
Programme (FPU20/05312 and FPU20/00039). We would also like to thank
the anonymous reviewers for their enriching feedback.
REFERENCES
Agorni, M. (2022). Corporate Website Translation and Effective
Communication Strategies. The Costs of Language Deficiencies.
Lingue Linguaggi, 52, 31-44.
https://doi.org/10.1285/I22390359V52P31
Baack, D. W., & Singh, N. (2007). Culture and Web Communications. Journal
of Business Research, 60(3), 181–188.
https://doi.org/10.1016/j.jbusres.2006.11.002
Briva-Iglesias, V. (2021). Traducción humana vs. Traducción automática:
Análisis contrastivo e Implicaciones para la Aplicación de la Traducción
automática en Traducción jurídica. Mutatis Mutandis, 14(2), 571-600.
https://doi.org/10.17533/udea.mut.v14n2a14
Carmen Moreno-Romero y Antonio Hermán-Carvajal 29
Hikma 23 (Número especial I) (2024), 1 - 32
Castilho, S., Doherty, S., Gaspari, F., & Moorkens, J. (2018). Approaches to
Human and Machine Translation Quality Assessment. In J. Moorkens,
S. Castilho, F. Gaspari & S. Doherty (Eds), Machine Translation:
Technologies and Applications (pp. 9-38). Springer International
Publishing. https://doi.org/10.1007/978-3-319-91241-7_2
Cuadrado Rey, A., & Navarro Brotons, L. (2024). Aproximación a la traducción
automática de culturemas gastronómicos en el ámbito turístico: estudio
de caso (español, alemán, francés). Hikma, 23(1), 111-139.
https://doi.org/10.21071/hikma.v23i1.15694
Daems, J., & Macken, L. (2019). Interactive Adaptive SMT versus Interactive
Adaptive NMT: a User Experience Evaluation. Machine Translation, 33,
117-134. https://doi.org/10.1007/s10590-019-09230-z
DeepL. (n.d.). Press Information: Media Kit. https://www.deepl.com/en/press
Fuentes-Luque, A., & Santamaría Urbieta, A. (2020). Machine Translation
Systems and Guidebooks: an Approach to the Importance of the Role
of the Human Translator. Onomázein, Núm. Especial 7, 63-82.
https://doi.org/10.7764/onomazein.ne7.04
Giampieri, P., & Harper, M. (2023). Tourism Translation: from Corpus to
Machine Translation (and back). Umanistica Digitale, 6(14), 119-135.
https://doi.org/10.6092/ISSN.2532-8816/15109
Görög, A. (2014). Dynamic Quality Framework: Quantifying and
Benchmarking Quality. Tradumàtica - Tecnologies de la Traducció, 12,
443-454. https://doi.org/10.5565/rev/tradumatica.66
Grego, K. S. (2010). Specialized Translation: Theoretical Issues, Operational
Perspectives. Polimetrica International Scientific Publisher.
Gutiérrez-Artacho, J., Olvera-Lobo, M. D., & Rivera-Trigueros, I. (2019).
Competencia comunicativa y nuevas tecnologías en el proceso de
localización web: modelo MDPT para la formación de profesionales en
localización. Fuentes, 21(1), 73-84.
https://doi.org/10.12795/revistafuentes.2018.v21.i1.05
Hidalgo-Ternero, C. M. (2021). Google Translate vs. DeepL. Analysing Neural
Machine Translation Performance under the Challenge of
Phraseological Variation. MonTI. Monografías de Traducción e
Interpretación, Núm. Especial 6, 154-177.
https://doi.org/10.6035/monti.2020.ne6.5
30 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Instituto Nacional de Estadística. (2023). Cuenta satélite del turismo de
España. Año 2022. [Press release]
https://www.ine.es/prensa/cst_2022.pdf
Instituto Nacional de Estadística. (2024a). Estadística de Movimientos
Turísticos en Fronteras (FRONTUR) Septiembre 2024. Datos
provisionales. [Press release]. https://t.ly/27ooT
Instituto Nacional de Estadística. (2024b). Movimientos Turísticos en
Fronteras. Número de turistas según país de residencia.
https://www.ine.es/jaxiT3/Datos.htm?t=10822
International Organisation for Standardisation (ISO). (2017). ISO
18587:2017(en) Translation services — Post-editing of Machine
Translation Output — Requirements.
https://www.iso.org/obp/ui/en/#iso:std:iso:18587:ed-1:v1:en
International Organisation for Standardisation (ISO). (2023). ISO
639:2023(en) Code for Individual Languages and Language Groups.
https://www.iso.org/obp/ui/en/#iso:std:iso:639:ed-2:v1:en
Jiménez-Crespo, M. A. (2013). Translation and web Localization. Routledge.
Kelly, D. (2002). Un modelo de competencia traductora: bases para el diseño
curricular. Puentes, 1, 9-20.
Kenny, D. (Ed.). (2022). Machine Translation for Everyone: Empowering
Users in the Age of Artificial Intelligence. Language Science Press.
https://doi.org/10.5281/zenodo.6653406
Kumar, J., Kumar Henge, S., & Dubey, P. (2022). An Intensive Literature
Review on Machine Translation Using Neural Approach. In P. K. Singh,
Y. Singh, M. H. Kolekar, A. K. Kar. & P. J. S. Gonçalves (Eds.), Recent
Innovations in Computing. Lecture Notes in Electrical Engineering, Vol.
832 (pp. 45-59). Springer. https://doi.org/10.1007/978-981-16-8248-
3_4
Le, Q. V., & Schuster, M. (2016). A Neural Network for Machine Translation,
at Production Scale. Google Research Blog. https://t.ly/6F7ud
Leiva Rojo, J. (2020). ¿Es la traducción automática una opción factible?
Reflexiones acerca de un estudio comparado de textos museísticos
traducidos (inglés-español). In S. Rodríguez Tapia & A. González
Fernández (Eds.), Lenguas y turismo: estudios en torno al discurso, la
didáctica y la divulgación (pp. 179-200). Peter Lang.
Carmen Moreno-Romero y Antonio Hermán-Carvajal 31
Hikma 23 (Número especial I) (2024), 1 - 32
Lommel, A. (2018). Metrics for Translation Quality Assessment: A Case for
Standardising Error Typologies. In J. Moorkens, S. Castilho, F. Gaspari
& S. Doherty (Eds.), Translation Quality Assessment: From Principles
to Practice (pp. 109-127). Springer International Publishing.
https://doi.org/10.1007/978-3-319-91241-7_6
Lommel, A., Uszkoreit, H., & Burchardt, A. (2014). Multidimensional Quality
Metrics (MQM): A Framework for Declaring and Describing Translation
Quality Metrics. Tradumàtica - Tecnologies de la Traducció, 12, 455-
463. https://doi.org/10.5565/rev/tradumatica.77
López González, R. C. (2020). Putting Touristic Destinations On The Map
Through Translation: How Public Institutions Provide Multilingual Visual
Guides. In C. Sarmento & M. Quincé (Eds.), Intercultural Routes Across
Multicultural Spaces. From Theory to Dialogue (pp. 57-77). Centro de
Estudos Interculturais.
Martín R., C. (2022). Los 20 monumentos más visitados de España.
Expansión. https://t.ly/85Rta
Mehler, A., Sharoff, S., & Santini, M. (Eds.). (2010). Genres on the Web:
Computational Models and Empirical Studies. Springer Science &
Business Media.
O’Brien, S. (2022). How to Deal with Errors in Machine Translation: Post
editing. In D. Kenny (Ed.), Machine translation for everyone:
Empowering users in the age of artificial intelligence (pp. 105-120).
Language Science Press.
Och, F. (2006). Statistical Machine Translation Live. Google Research Blog.
https://research.google/blog/statistical-machine-translation-live/
Olvera-Lobo, M. D., & Castillo-Rodríguez, C. (2019). Website Localisation in
the Corporate Context: A Spanish Perspective. Journal of Digital
Information Management, 17(1), 34-43.
https://doi.org/10.6025/jdim/2019/17/1/34-43
Peña Aguilar, A. (2023). Challenging Machine Translation Engines: Some
Spanish-English Linguistic Problems Put to the Test. Cadernos de
Tradução, 43(1), 1-26. https://doi.org/10.5007/2175-7968.2023.e85397
Rico Pérez, C. (2024). Re-thinking Machine Translation Post-Editing
Guidelines. The Journal of Specialised Translation, 7, 26-47.
https://doi.org/10.26034/cm.jostrans.2024.4696
Rothwell, A., Moorkens, J., Fernández-Parra, M., Drugan, J., &
Austermuehl, F. (2023). Software, Web, and Game Localisation. In A.
32 Machine Translation and Tourism Discourse: A Spanish-English-French […]
Hikma 23 (Número especial I) (2024), 1 - 32
Rothwell, J. Moorkens, M. Fernández-Parra, J. Drugan &
F. Austermuehl (Eds.), Translation Tools and Technologies (pp. 174-
189). Routledge. https://doi.org/10.4324/9781003160793-10
Singh, N., & Pereira, A. (2005). The Culturally Customized Web Site.
Routledge.
Sin-wai, C. (2012). Approaching Localization. In C. Millán & F. Bartrina (Eds.),
The Routledge Handbook of Translation Studies (pp. 347-362).
Routledge. https://doi.org/10.4324/9780203102893.ch25
Suau Jiménez, F. (2012). El turista 2.0 como receptor de la promoción
turística: estrategias lingüísticas e importancia de su estudio. PASOS.
Revista de Turismo y Patrimonio Cultural, 10(4), 143-153.
https://doi.org/10.25145/j.pasos.2012.10.060
TAUS. (n.d.). Start Tracking Errors with DQF-MQM. https://info.taus.net/dqf-
mqf-error-typology-template-download
Tercedor Sánchez, M. (2005). Aspectos culturales en la localización de
productos multimedia. Quaderns: Revista de Traducció, 12, 151-160.
https://ddd.uab.cat/pub/quaderns/11385790n12/11385790n12p151.pd
f