The standardization of soundscape protocols in the ISO/TS 12913 series of standards (ISO2014ISOFramework; ISO2018ISO/TSRequirements; ISO2019ISO/TSAnalysis) have greatly unified the reporting standards and data collection methodology in soundscape studies. However, the protocols for data collection were only standardized in English, leaving the use of soundscape terminologies largely unstandardized in regions of the world where English is not the main language. Since the soundscape studies, by definition, are concerned with the relationships between an acoustic environment and the human experience in context (ISO2014ISOFramework), soundscape studies are highly intertwined with the socio-cultural context not only of the acoustic environment itself, but also of the language medium the studies are conducted in. The lack of standardization in soundscape has posed a serious challenge in the transferability and comparability of soundscape knowledge along the linguistic borders.
1.1 Related work in soundscape translation
Prior to the publication of ISO 12913-2:2018 and 12913-3:2019, much of the work on soundscape assessment has been reliant on the Swedish Soundscape-Quality Protocol (SSQP) by Axelsson2009AQuality; Axelsson2010APerception; Axelsson2012TheProtocol, which, together with the work of Cain2013TheSoundscape, would later become the basis for the two-dimensional model in the ISO 12913-3:2019. The SSQP itself has previously been translated into 10-15 languages, albeit without experimental validation to ensure interlingual compatibility. Unsurprisingly, a cross-national study by Jeon2018AExperiments found statistically significant differences in soundscape assessment across France, Korea, and Sweden, highlighting important comparability issues of the soundscape assessment scales across languages. These linguistic issues are also present in the findings of studies conducted in Japan (Nagahata2018LinguisticResearch; Nagahata2019ExaminationJapanese), and a French-speaking region of Canada (Tarlao2016ComparingMontreal).
After the publication of the ISO 12913-2:2018 and 12913-3:2019, an international collaboration was initiated amongst soundscape researchers in order to develop experimentally validated and cross-linguistically compatible sets of translations of the soundscape attributes. The collaboration, named Soundscape Attribute Translation Project (SATP), is currently working on 19 languages, with several whose experimental validations have been completed. Generally, most working groups followed a two-stage structure where the first is concerned with developing a set of provisional translations while the second is concerned with validating the translations via a listening test (Aletta2020SoundscapeLanguages).
Due to the lack of ‘gold standard’ procedures for translation in the field of soundscapes, provisional translation stage for Mandarin Chinese, Yue Chinese, Croatian, Dutch, French, Italian, Korean, Spanish, Swedish, Turkish, Vietnamese (Aletta2020SoundscapeLanguages), and Bahasa Indonesia (Sudarsono2022TheStudy), have largely relied on expert panels and/or focus group discussions, and, if any, previously soundscape works involving translations in their respective languages. However, solely relying on an expert panel for the translation process may not always produce a set of translations that is compatible with the English standards. As seen in the Indonesian study (Sudarsono2022TheStudy), significant deviations from the English results were found for the translations of eventful, uneventful, and chaotic during the experimental validation.
Admittedly, it is possible to approach the translation process and validation process in an iterative manner. However, this is usually neither desirable nor practical, as conducting listening tests is a time-consuming and labor-intensive process. Moreover, testing multiple sets of candidate translations in one sitting is also often not advisable, as the number of candidate translations is limited by experimental fatigue of the participants (Schatz2012TheRatings; Schwarz2016EffectsTests).
Interestingly, the Portuguese provisional translation process (Antunes2021ValidatedAssessment) took a more structured and slightly more quantitative approach with Stage 1 divided into two substages. Although the first substage concerning the initial selection process of candidate translations remains inevitably dependent on an expert panel, in the second substage, an online questionnaire administered cross-nationally in Portugal and Brazil was employed to finalize their set of provisional translations before proceeding to the listening test stage. For each soundscape attribute, the questionnaire item starts with a sentence describing an acoustic environment with a perceptual quality of the attribute, followed by asking which of the candidate translations is considered most “suitable”. It is important to note that the participants are also given a free response box for each item to suggest another translation. The results of the questionnaire provided the Portuguese working group with quantitative insights on the suitability of each candidate translation, as well as cross-cultural differences in the use of the Portuguese language between Portugal and Brazil.
In this work, we build on the insights from the Portuguese translation process and propose an extended quantitative evaluation framework for the translation of soundscape attributes. Crucially, instead of a choice-based evaluation of overall suitability used in Antunes2021ValidatedAssessment, our framework utilizes a score-based approach to assess the quality of a translation across multiple criteria concerning the appropriateness of translation, understandability, clarity, and its linguistic relationships to other soundscape attributes. In particular, the use of multiple criteria in a score-based framework allows for the use of statistical analysis in the selection process, as well as the identification of particular strengths or weaknesses each candidate translation may possess. Overall, the framework seeks to identify a set of provisional translations which preserves not only the original meanings, but also the inter-attribute relationship between the eight soundscape descriptors.
The quantitative framework proposed in this paper has been used for the translation process for the Thai language (ISO 639-3: tha) and Bahasa Melayu (ISO 639-3: zsm). In this work, we will focus on the design of the quantitative framework, and its application to the Thai language. Due to the cross-national nature of Bahasa Melayu, a full discussion of its translation process is beyond the scope of this paper and will be addressed in detail in a separate work.
1.3 Terminology and Organization
This paper largely follows the soundscape attribute terminology used in the ISO/TS 12913 standards (ISO2014ISOFramework; ISO2018ISO/TSRequirements; ISO2019ISO/TSAnalysis). Additionally, we introduce terms to describe the relationship between soundscape attributes in the circumplex model as follows. Attributes located from each other are considered adjacent. Attributes located from each other are considered orthogonal, and attributes located from each other are considered antipodal. Additionally, clockwise and counterclockwise qualifiers are used to specify the relative location of an attribute with respect to another based on their locations on the circumplex model. Figure 1 illustrates the circumplex model of soundscape attributes and the relation terminology used in this paper.
This paper is organized as follows. Section 2 discusses the design of the proposed quantitative evaluation framework. Section 3 and Section 4 discuss the application of the proposed framework into the Thai language, with the former discussing the initial expert translation phase, and the latter discussing the validation questionnaire. Finally, Section 5 concludes the paper and lays out future work. For readability, Thai words used in this paper are followed by its International Phonetic Alphabet (IPA) transcription, and, where appropriate, also by its literal meaning following the lit. indicator.
2 Design of the Quantitative Evaluation
In most translation processes, several candidate translations are often proposed via either an expert panel discussion or a parallel translation method. Traditionally, the consensus stage to obtain the final set of translations is also expert-based. However, such a consensus stage is highly qualitative and subjective in nature and provides no quantitative means of validating the experts’ opinion before proceeding to subsequent validation stages. As such, the proposed quantitative evaluation framework is intended to serve as a more robust consensus-cum-bias evaluation stage of the translation process (Gudmundsson2009GuidelinesInstruments; Wild2005PrinciplesAdaptation), before proceeding to listening experiments. The design of the quantitative evaluation is based loosely on the guidelines by Gudmundsson2009GuidelinesInstruments and the Test Development and Confirmation Guidelines in the ITC Guidelines for Translating and Adapting Tests (ITC2017). Due to the nature of the soundscape attribute translation, where individual attributes are translated to allow for standardized usage across various research methodologies, method bias cannot be investigated. As such, the main goal of the quantitative evaluation process is to assess the psychometric equivalence between the translated attributes and the standardized English attributes.
Since the soundscape attributes together form a circumplex model, there is also a need to preserve the inter-attribute psychometric properties in the translations. However, testing all interactions between all local translations would be intractable. Instead, we rely on bilingual speakers of the local language and English to perform the validation. By anchoring all other attributes in English and only testing each candidate translation one at a time, it is still possible to select the most suitable candidate translation that fits in the circumplex model without using impractically many test items. This is done by ensuring that for each attribute, the interaction between the other seven English attributes and the best candidate translation is as close to that of its English counterpart as possible. By extension, it follows that the interaction between the eight final translations will approximate that of the original circumplex model in English.
The quantitative evaluation process is questionnaire-based, with a fixed set of questionnaire items for each attribute-translation pair. To minimize any further translation, the questionnaire is conducted in English, with only the translation candidate being in the local language. Responses to each questionnaire item are rated on an 11-point Likert scale between 0 to 10. For ease of score computation, all responses are normalized to the range . We define the normalized rating of a participant by where [qn] represent the question, the contribution of a rating to a score by where [cr] represent the criterion, and the overall score by . Note that all scores are purposely designed to be in the range for ease of comparison.
For brevity, we indicate a ‘field’ in a questionnaire item in square brackets […]. [loc] indicates the candidate translation in the local language that is being evaluated. [eng] indicates a soundscape attribute in English. Without additional qualifier, [eng] refers to the source attribute that [loc] is being translated from. [eng] with additional qualifier, e.g., [adjacent eng], refers to the related attribute in English of the source attribute.
2.1 Appropriateness (appr)
The first evaluation criterion for the translation is the appropriateness of the translation with respect to the English attribute. Simply put, the translation candidate has to first be a commonly accepted translation of the English attribute with sufficiently similar meanings. This criterion applies to all attributes on both the main and derived axes.
The prompt for the appropriateness evaluation reads:
To what extent do you agree/disagree that [loc] is an appropriate translation of [eng]?
with full disagreement represented by the rating and the full agreement represented by the rating . For example, a prompt for evaluating the appropriateness of “angenehm” as a German translation for pleasant would read
To what extent do you agree/disagree that “angenehm” is an appropriate translation of pleasant?
For appropriateness score (appr), we use a simple contribution system .
2.2 Understandability (undr)
The next evaluation criterion is concerned with the general understandability of the translation candidate in the target population. That is, the translation candidate has to be a sufficiently commonplace term amongst the speakers of the local language, and not an expert or academic jargon that may not be easily understood by the general public. This criterion applies to all attributes on both the main and derived axes.
The prompt for the understandability evaluation reads:
To what extent do you agree/disagree that [loc] is easily understood by a typical general [local language] speaker?
with full disagreement represented by the rating and the full agreement represented by the rating . For example, a prompt for evaluating the understandability of “dynamique” as a French translation for vibrant would read
To what extent do you agree/disagree that “dynamique” is easily understood by a typical general French speaker?
For understandability score (undr), we also use a simple contribution .
2.3 Clarity (clar)
Depending on linguistic peculiarities, certain local translation candidates can be easily confused or more often associated as a translation for an adjacent attribute, instead of the target attribute. As such, it is important to quantify the degree in which the local translation candidate will be unambiguously perceived as the target translation, instead of adjacent attributes.
The prompt for this questionnaire item reads,
To what extent do you agree/disagree that [loc] is more often associated as a translation of [adjacent eng]?
with full disagreement represented by the rating and the full agreement represented by the rating . For each attribute, clarity is evaluated twice, once against the clockwise adjacent attribute (), and once against the counterclockwise adjacent attribute ().
For example, the two prompts for evaluating the clarity of “membosankan” as a Bahasa Melayu translation of monotonous would read
To what extent do you agree/disagree that “membosankan” is more often associated as a translation of uneventful?
To what extent do you agree/disagree that “membosankan” is more often associated as a translation of annoying?
with the rating of the former being as uneventful is the counterclockwise adjacent attribute of monotonous, and the rating of the latter being as annoying is the clockwise adjacent attribute of monotonous.
The ratings from these questionnaire items are then used to compute the clarity score (clar), by penalizing the total extent in which the candidate translation may be confused as a translation of an adjacent attribute, such that
2.4 Antipodal Antonymity (anto)
For attributes on the main axes, the translation candidates have to reflect the antonymous relationship between the pair of attributes on each end. For example, the translation of pleasant should have an antonymous relationship to both the translation of annoying. However, since the VQ itself is used to evaluate the suitability of the translation candidates, we use the English term as a proxy for the evaluation of antonymity. The prompt for antonymity evaluation reads:
To what extent do you agree/disagree that [loc] is a direct antonym of [antipodal eng]?
with full disagreement represented by the rating and the full agreement represented by the rating .
To illustrate, the prompt for evaluating antipodal antonymity of “sinh động” as a Vietnamese translation of eventful would read
To what extent do you agree/disagree that “sinh động” is a direct antonym of uneventful?
Antipodal antonymity score (anto) also uses a simple contribution system .
2.5 Orthogonal Unbiasedness (orth)
For attributes on the main axes, it is also important that the linguistic orthogonality between the pleasant-annoying and the eventful-uneventful axes are preserved after the translation. In other words, an attribute on a main axis should be as neutral as possible with respect to the two orthogonal attributes on the other main axis. As with antonymity, we use the English terms as a proxy for the evaluation.
The prompt for bias evaluation reads:
To what extent is [loc] (as a description of an acoustic environment) biased with respect to the [ccw orthogonal eng]–[cw orthogonal eng] axis?
where full bias towards the clockwise orthogonal attribute is represented by the rating and full bias towards the counterclockwise orthogonal attribute is represented by the rating . For instance, the prompt for evaluating the orthogonal bias of “keyifsiz” as a Turkish translation of annoying would read
To what extent is “keyifsiz” (as a description of an acoustic environment) biased with respect to the uneventful–eventful axis?
with full bias towards uneventful represented by the rating and full bias towards eventful represented by the rating .
The orthogonality score (orth) is based on the extent which the rating deviates from the neutral point (), and is thus given by
2.6 Connotativeness (conn), Nonconnotativeness (ncon), and Implicative Balance (ibal)
Lastly, the implicative meanings of the translated candidates are evaluated. For this particular category, the desired behavior of the candidate translation differs between attributes on the main axes and those on the derived axes. For attributes on the main axes, their translation should also preserve their ‘basis’ nature, in the sense that they should not imply adjacent attributes on the derived axes. For attributes on the derived axes, their translations should imply both adjacent attributes on the main axes. To illustrate, describing a soundscape as pleasant does not imply that the soundscape is necessarily calm nor vibrant. On the other hand, describing a soundscape as vibrant, should imply that the soundscape is both pleasant and eventful. Additionally, since the attributes on the derived axes also serve as the approximate anchor of the angular midpoint between their respective adjacent attributes on the main axes, translations of the attributes on the derived axes should serve the same function in the target language. Note the distinction between the concept of implication in this item and the concept of confusion in the clar criterion.
The questionnaire prompt for this item reads,
To what extent do you agree/disagree that [loc] (as a description of an acoustic environment) implies that the environment is also [adjacent eng]?
with full disagreement represented by the rating and the full agreement represented by the rating .
For example, the two questionnaire prompts to obtain the implicative ratings “kaotiskt” as a Swedish translation of chaotic would read
To what extent do you agree/disagree that “kaotiskt” (as a description of an acoustic environment) implies that the environment is also annoying?
To what extent do you agree/disagree that “kaotiskt” (as a description of an acoustic environment) implies that the environment is also eventful?
with the former represented by as annoying is the counterclockwise adjacent attribute of chaotic, and the latter by as eventful is the clockwise adjacent attribute of chaotic.
For attribute on the main axes, where implying adjacent attributes is undesirable, the non-connotativeness score (ncon) is computed similarly to the clarity score, such that
For attributes on the derived axes, where implying adjacent attributes is desirable, the connotativeness score (conn) is given by
Lastly, the implicative balance score (ibal) is computed by penalizing the difference between the two scores, such that
Although the ibal score is mainly designed to evaluate attributes on the derived axes, it is important to note that in practice, no translation of the attributes on the main axes would be perfectly non-connotative of their respective adjacent attributes. With this in mind, we also compute the ibal score for attributes on the main axes to ensure that, even if they are not completely non-connotative of adjacent attributes, the extent of connotativeness remains similar between the clockwise adjacent and the counterclockwise adjacent.
3 Phase I: Initial Translation by Experts
Due to the lack of a translation protocol specific to psychoacoustics, we relied on several guidelines used for translation of psychological instruments (Gudmundsson2009GuidelinesInstruments; Borsa2012Cross-culturalConsiderations; ITC2017) and made adjustments specific to the requirements of soundscape attribute translation.
3.1.1 Translation team
The initial expert translation phase involved five linguistic experts who are all native Thai speakers and bilingual in English. As the linguists are not soundscape experts, summaries of the soundscape methodologies and standards, particularly on the circumplex model, were provided prior to the start of the translation process. The translation process is additionally facilitated by a soundscape researcher who is also a native Thai speaker.
3.1.2 Method of translation
The experts were first asked to independently produce a set of potential translations for each of the eight English soundscape attributes without consulting any other expert. As remarked in Gudmundsson2009GuidelinesInstruments, a process based on parallel independent translations was adopted as the translation methodology, instead of validation via back translation, due to the need to prioritize the psychometric equivalence of the translated soundscape attributes. Moreover, the use of independent translations has the additional benefit of helping to identify as much potential ambiguity in translation as possible.
Due to the constraints associated with the translations of the soundscape attributes, the experts are also provided with guidelines specific to the English-to-Thai translation of soundscape attributes, which was adapted from the generic provisional translation guidelines used in the SATP project. The guidelines provided are summarized in the next section (Section 3.1.3).
Following the initial translation, the experts were convened to discuss the initially proposed translations. New candidate translations may also be proposed at this point. Instead of the usual aim of reaching a consensus for a final translation, the goal of this discussion is to shortlist a few candidate translations per soundscape attribute to proceed to the quantitative evaluation phase. By doing so, the typically subjective and unquantifiable consensus process for arriving at a final translation set is eliminated and replaced by a quantitative analysis.
The shortlisted translation candidates at the end of this phase are shown in Table 1.
|Eng. Attr.||Thai Translation Candidates|
|Monotonous||จืด ๆ||/cẀ:t.cẀ:t/||น่าเบื่อ||/nâ:.bẀa/||เนือย ๆ||/nW̄aj.nW̄aj/||เอื่อย ๆ||/PẀaj.PẀaj/|
3.1.3 Translation guidelines
First, the translations should strive to use common words that are easily understood by laypeople. This also means academic terms and jargon which may not be easily understood by most Thai-speaking general population should be avoided.
Next, each English attribute should be translated in relation to the perception of sounds. It is more desirable to retain the meaning in the acoustic-perceptual sense rather than pursuing a literal translation. If a single Thai word does not sufficiently capture the original meaning of the English attribute, a set of two to three Thai words can be proposed instead. The use of วลี /wàP.lī:/, lit. short phrases, are also allowed since the distinction between adjectives and an adjectival phrases can be ambiguous in Thai (Post2008AdjectivesClasses).
For the attributes on the main axes, the attributes should be translated as neutrally as possible with respect to the orthogonal axis. For example, pleasant and annoying should be translated as neutrally with respect to the eventful-uneventful axis as possible. Additionally, it is desirable that translations of antipodal attributes are antonyms of each other. However, as far as possible, avoid the use of the negative particle ‘ไม่’ /mâj/, lit. not, to avoid ambiguity. Similar to the English word ‘not’, ‘ไม่’ /mâj/ may be interpreted either as a truly negating operator — e.g. not cold being interpreted as warm — or as a neutralizing operator — e.g. not cold being interpreted as neither cold nor warm (Atlas1977NegationPresupposition; Takahashi1997NegationStudy).
3.2.1 The use of auditory indicator morpheme in candidate translations
Several candidate translations were proposed by the expert to contain the morpheme ‘หู’ /hǔ:/, lit. ear, to indicate that the adjective is acoustical in nature. The morphemes preceding ‘หู’ /hǔ:/ in most translation candidates containing it do not inherently have acoustical connotations, but the addition of ‘หู’ /hǔ:/ provides a clear indication that the preceding morpheme is describing an acoustic environment. For example, ‘รื่น’ /rW̌:n/, lit. enjoyable or comfortable, in รื่นหู /rŴ:n.hǔ:/ and ‘ระคาย’ /ráP.khā:j/, lit. to irritate, in ระคายหู /ráP.khā:j.hǔ:/. In a similar manner, ‘ฟัง’ /fāN/, lit. to listen, serves a similar purpose in น่าฟัง /nâ:.fāN/.
3.2.2 Translation of eventful and uneventful
Amongst the perceptual attributes, the so-called arousal-axis attributes eventful and uneventful prove particularly difficult to translate, especially with respect to the connotation of valency. Translations of eventful and uneventful to Thai usually depend on the context and typically take the form of one of the adjacent attributes on the derived axes. For this pair of attributes, the guideline on the avoidance of ‘ไม่’ /mâj/, lit. not, was relaxed, partly due to the presence of the ‘un-’ morpheme in uneventful.
A particularly interesting pair of translation candidates are มีอะไร /mī:.PàP.rāj/ and ไม่มีอะไร /mâj.mī:.PàP.rāj/, as the pair are morphologically very close to the English terms. The Thai morpheme ‘มี’ /mī:/, lit. to have, can be considered to correspond to the English morpheme ‘-ful’, while ‘อะไร’ /PàP.rāj/, lit. something, can be considered to correspond to the free morpheme ‘-event-’. Although both terms were remarked by the experts as not commonly used in formal Thai, it was nonetheless agreed by them as perhaps the most inherently neutral with respect to perceptual valency out of all the eventful-uneventful candidates. Other candidate translations of eventful and uneventful were all considered by at least one expert to be inherently leaning towards one of the terms on the derived axes during the discussion – an opinion which will be later supported by the evaluation questionnaire.
4 Phase II: Quantitative Evaluation on Thai Translation Candidates
The quantitative evaluation phase was conducted between November 2021 and February 2022, during which a total of 31 participants who are bilingual in Thai and English were recruited across Thailand, Singapore, and the United Kingdom. The participants were also asked to report their length of stay outside Thailand, of which 4 participants () reported less than one year, 13 participants () reported between 1 to 5 years, 6 participants () reported between 6 to 10 years, and 8 participants () reported more than 10 years.
The participants were asked to rate their proficiency in Thai and English on the Interagency Language Roundtable (ILR) scale. For the Thai language, 30 participants () reported Native Proficiency and 1 participant () reported Professional Working Proficiency. For English, 5 participants () reported Native Proficiency (ILR 5), 12 participants () reported Full Professional Proficiency (ILR 4), 9 participants () reported Professional Working Proficiency (ILR 3), 4 participants () reported Limited Working Proficiency (ILR 2), and 1 participants () reported Elementary Proficiency (ILR 1).
The survey was administered online via Google Forms. Google Apps Script was used to programmatically generate the questionnaire. In total, the questionnaire contains 178 items, excluding the demographic information section.
The evaluation scores introduced in Section 2
were calculated using the results. For each evaluation criterion of each English attribute, without assuming normality, a Kruskal–-Wallis test(KWT; Kruskal1952UseAnalysis) was performed on the score contributions with respect to the candidate translations. If a statistical significance was found at significance level using the KWT, then a posthoc Conover–Iman test111The Conover–Iman test is similar to the more well-known Dunn’s test (Dunn1964MultipleSums) but uses the Student’s t
-distribution instead of the normal distribution. We use the more statistically powerful Conover–Iman test in this paper.(CIT; Conover1979OnProcedures) with Bonferroni correction (Dunn1961MultipleMeans) was performed to identify pairwise differences. No posthoc test is performed if the KWT is not statistically significant.
Table 2 and Table 3 show the mean evaluation scores222We have also experimented with weighting the contributions to the scores by the sum of English and Thai proficiency ILR rating, but the weighting has negligible effects on the scores.. The results of the pairwise tests are shown in Table A.2 in the Appendix.
The raw results of the questionnaire are shown in Figure 2 and Figure 3 for the attributes on the main and derived axes, respectively. The following sections will discuss the results of the validation questionnaire and the final translation candidates, which are also summarized in Table 4.
|Eng. Attr.||Thai Translation Candidate||APPR||UNDR||CLAR||ANTO||ORTH||NCON||IBAL|
With the exception of undr, all translation candidates for pleasant performed similarly across all other criteria, showing no significant difference using the KWT at . Since น่าฟัง /nâ:.fāN/ performed significantly better than all other translation candidates in terms of understandability, with CIT against all other candidates, น่าฟัง /nâ:.fāN/ was selected as the final translation for pleasant.
Both candidates for annoying show no statistically significant difference across the clar, anto, orth, ncon, and ibal criteria. For the appr and undr criteria, น่ารำคาญ /nâ:.rām.khā:n/ scored higher than ระคายหู /ráP.khā:j.hǔ:/ with CIT on both. As such, น่ารำคาญ /nâ:.rām.khā:n/ was selected as the final translation for annoying.
With the exception of anto, KWT showed statistically significant differences in all other criteria for eventful.
In terms of appr, อึกทึก /PẀk.kàP.thẂk/ performed better than มีอะไร /mī:.PàP.rāj/ with . Other pairs show no significant difference at . In terms of undr, วุ่นวาย /wûn.wā:j/ performed better than มีอะไร /mī:.PàP.rāj/ and อึกทึก /PẀk.kàP.thẂk/, with on both, while มีอะไร /mī:.PàP.rāj/ performed better than อึกทึก /PẀk.kàP.thẂk/ with . For the rest of the criteria, มีอะไร /mī:.PàP.rāj/ performed significantly better than วุ่นวาย /wûn.wā:j/ with in all. Against อึกทึก /PẀk.kàP.thẂk/, มีอะไร /mī:.PàP.rāj/ performed better in clar and ncon, but no significant difference was found in orth and ibal.
As earlier remarked by the experts during the initial translation phase, neither is มีอะไร /mī:.PàP.rāj/ a word commonly used for the translation of eventful, nor is it a word commonly found in formal usage of the Thai language. As such, the lower appr and undr scores are somewhat expected. In fact, มีอะไร /mī:.PàP.rāj/ being an uncommon word may have contributed to the high clar score, as it would not be easily confused as a translation of another soundscape descriptor. วุ่นวาย /wûn.wā:j/, however, despite having a very high undr score, is very easily confused as a translation of chaotic as seen in Figure 2 — and it, in fact, will be chosen as the final translation for chaotic in Section 4.1.6. Considering the evaluation criteria as a whole, มีอะไร /mī:.PàP.rāj/ was selected as the final translation for eventful.
In terms of appr, ไม่มีอะไร /mâj.mī:.PàP.rāj/ performed significantly better than เรียบ /rîap/ with , but no significant difference was found against สงบ /sàP.Nòp/. In terms of undr, สงบ /sàP.Nòp/ and ไม่มีอะไร /mâj.mī:.PàP.rāj/ both performed better than เรียบ /rîap/, with and , respectively. No significant difference was found between สงบ /sàP.Nòp/ and ไม่มีอะไร /mâj.mī:.PàP.rāj/. Interestingly, despite being just as uncommonly used as the eventful candidate มีอะไร /mī:.PàP.rāj/, ไม่มีอะไร /mâj.mī:.PàP.rāj/ was rated with relatively good appr () and undr () scores, whereas the former was rated much lower with appr () and undr ().
In terms of anto, สงบ /sàP.Nòp/ performed better than เรียบ /rîap/ with , but no significant differences were found with other pairs. With orth, ไม่มีอะไร /mâj.mī:.PàP.rāj/ performed better than both other candidates with against สงบ /sàP.Nòp/ and against เรียบ /rîap/. At the same time, เรียบ /rîap/ performed better than สงบ /sàP.Nòp/ with . As shown in Figure 2, สงบ /sàP.Nòp/ is generally rated as strongly biased towards pleasant, and a similar bias is also seen with เรียบ /rîap/ to a lesser degree.
In terms of clar, ไม่มีอะไร /mâj.mī:.PàP.rāj/ performs better than สงบ /sàP.Nòp/ with , but significant difference was not found against เรียบ /rîap/ with at significance level. เรียบ /rîap/ and สงบ /sàP.Nòp/ performed very similarly in this criterion, with , although the nature of the association is somewhat different. สงบ /sàP.Nòp/ has a very strong association as a translation of calm rather than uneventful but much less so with monotonous. On the other hand, เรียบ /rîap/ has moderately high associations as a translation of both calm and monotonous. A similar result was seen with ncon, where ไม่มีอะไร /mâj.mī:.PàP.rāj/ performed better than สงบ /sàP.Nòp/ with , but no significant difference was found against เรียบ /rîap/. Expectedly, analysis of the ibal scores also shows that both ไม่มีอะไร /mâj.mī:.PàP.rāj/ and เรียบ /rîap/ performed significantly better than สงบ /sàP.Nòp/ with .
Considering all criteria as a whole, ไม่มีอะไร /mâj.mī:.PàP.rāj/ was chosen as the final translation, particularly due to the poor evaluation of สงบ /sàP.Nòp/ in clar, orth, and ncon.
|Eng. Attr.||Thai Translation Candidate||APPR||UNDR||CLAR||CONN||IBAL|
Whereas สงบ /sàP.Nòp/ did not perform well as a translation of uneventful, it was rated better as a translation of calm. In terms of appr, สงบ /sàP.Nòp/ performed better than both other candidates with . สงบ /sàP.Nòp/ also performed better than สบายหู /sàP.bā:j.hǔ:/ in undr and ibal with . For clar and conn, no significant difference in distribution were found using the omnibus test. As such, สงบ /sàP.Nòp/ was selected as the final translation for calm.
With appr, all other candidates performed better than จอแจ /cŌ:.cĒ:/, with for วุ่นวาย /wûn.wā:j/ and ยุ่งเหยิง /jûN.j7̌:N/ and for ปั่นป่วน /pàn.pùan/. Excluding จอแจ /cŌ:.cĒ:/, no other significant pairwise differences were found for appr.
For undr, วุ่นวาย /wûn.wā:j/ performed better than all other candidates with against จอแจ /cŌ:.cĒ:/ and ปั่นป่วน /pàn.pùan/, and against ยุ่งเหยิง /jûN.j7̌:N/. For clar, the only significant pairwise difference with ยุ่งเหยิง /jûN.j7̌:N/ against วุ่นวาย /wûn.wā:j/ with . No significant differences in distribution were found for conn and ibal.
Considering that วุ่นวาย /wûn.wā:j/ outperformed all other candidates in undr and performed similarly to ยุ่งเหยิง /jûN.j7̌:N/ and ปั่นป่วน /pàn.pùan/ in appr, วุ่นวาย /wûn.wā:j/ was selected as the final translation for chaotic.
The omnibus tests indicate statistically significant differences in distributions for all evaluation criteria. For appr, มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/ performed better than สดใส /sòt.sǎj/ with and หวือหวา /wW̌:.wǎ:/ with . สดใส /sòt.sǎj/ and คึกคัก /khẂk.khák/ also performed better than หวือหวา /wW̌:.wǎ:/ with . มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/ and คึกคัก /khẂk.khák/ did not have statistically significant difference (), and neither did สดใส /sòt.sǎj/ and คึกคัก /khẂk.khák/ ().
In terms of undr, มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/, สดใส /sòt.sǎj/, and คึกคัก /khẂk.khák/ all performed similarly with . Against หวือหวา /wW̌:.wǎ:/, the former three all performed better with . Interestingly, despite the relatively lower appropriateness and understandability ratings, หวือหวา /wW̌:.wǎ:/ performed the best in terms of clar with against the other three, while other pairs are not statistically significant. The better performance of หวือหวา /wW̌:.wǎ:/ is likely due to the other three having a stronger association towards pleasantness than they do annoying, as seen in Figure 3.
However, หวือหวา /wW̌:.wǎ:/ perform poorly in terms of conn, as it indicates eventfulness much more strongly than it does pleasantness. The posthoc tests give with respect to คึกคัก /khẂk.khák/ and มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/, and with respect to สดใส /sòt.sǎj/. In terms of ibal, สดใส /sòt.sǎj/, คึกคัก /khẂk.khák/, and หวือหวา /wW̌:.wǎ:/ performed very similarly with . มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/ performed better than คึกคัก /khẂk.khák/ with . All other pairwise tests are not statistically significant at .
Overall, มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/ performed the strongest in undr and ibal, with no statistically significant differences against คึกคัก /khẂk.khák/ and สดใส /sòt.sǎj/ in clar and conn. As such, มีชีวิตชีวา /mi:.chī:.wít.chī:.wā:/ was selected as the final translation for vibrant.
Except for undr, no other criterion shows statistically significant differences in performance. Since น่าเบื่อ /nâ:.bẀa/ outperformed the rest in terms of undr with , น่าเบื่อ /nâ:.bẀa/ was selected as the final translation for monotonous.
|English Attribute||Final Thai Translation|
In this work, we proposed a structured and quantitative framework for the translation of soundscape attributes from the standardized English terms in the ISO 12913 series of standards to a local language. By considering the inter-attribute relationships of soundscape attributes on the circumplex model, a set of questionnaire items and corresponding evaluation criteria was developed to evaluate the linguistic-cultural suitability and psychometric equivalence of the candidate translations. The proposed framework was then applied to the soundscape attribute translation process for the Thai language as an initial study.
The translation process involves two phases, the first being parallel translations by linguistic experts and a group discussion to obtain a shortlist of candidate translations. The second phase applied the proposed quantitative framework to assess the translation candidates and select the final set of translations based on the evaluation scores. In total, 31 participants who are bilingual in Thai and English were recruited to participate in the pilot validation questionnaire. The proposed evaluation metrics were computed based on the raw questionnaire responses and statistical tests were performed to identify the most suitable translation for each of the soundscape attributes.
The use of a quantitative framework has greatly facilitated the process of identifying and verifying the strengths and weaknesses of each candidate translation, and the data can continue to provide insights on the linguistic and psychometric properties of the translations for the experimental validation stage and future studies based on this set of translations.
The proposed quantitative framework is also currently being used for the translation of soundscape attributes to Bahasa Melayu, which is a language used in both Malaysia and Singapore. The quantitative nature of the assessment data would allow for statistical comparison of cross-national differences and similarities in the linguistic and psychometric properties of the candidate translations.
Although the proposed framework has been constructed specifically for the ISO 12913 circumplex model, it is also applicable to other octant-based circumplex models. Moreover, it is also possible to adapt or extend the current framework for the translation of other psychoacoustic descriptors or models where the preservation of the inter-descriptor construct is crucial. All in all, the authors hope that the proposed framework will continue to assist other researchers in the field of soundscapes and psychoacoustics working on a translation, and act as a step forward in transforming the traditionally subjective and heavily expert-reliant process into one that is more robust and verifiable.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
This work was supported by the Google Cloud Research Credits program (GCP205559654).
The authors would like to thank Dr. Francesco Aletta, Dr. Tin Oberman, Andrew Mitchell, and Prof. Jian Kang, of the UCL Institute for Environmental Design and Engineering, The Bartlett Faculty of the Built Environment, University College London (UCL), London, United Kingdom, for coordinating the SATP project and providing assistance for the Thai Language Working Group.
We would like to also thank Pulaporn Sreewichian (School of Computing, Engineering and Built Environment, Glasgow Caledonian University, Glasgow, Scotland, United Kingdom), Kan Jitpakdi (Thai Student Society, Singapore; National University of Singapore, Singapore), Phumrapee Pisutsin (Thai Student Society, Singapore; Nanyang Technological University, Singapore), Nerinat Yongphiphatwong (Samaggi Samagom, United Kingdom; University of Cambridge, Cambridge, United Kingdom), and Siddha Kumwongwan (Samaggi Samagom, United Kingdom; University College London, London, United Kingdom), for their assistance with the distribution of the quantitative validation questionnaire.
Appendix A Results of Statistical Tests on the Evaluation Scores
-values of the Kruskal–Wallis tests and the posthoc Conover–Iman tests. Double asterisks (**) and single asterisk (*) indicate statistical significance at and , respectively. For Conover–Iman tests, the pairwise tests are listed in the order of increasing -value, and the candidate translation with the higher average score in the respective criterion is always listed on the left of each pair.
|Eng. Attr.||Criterion||Statistical Test||-value|
|จืด ๆ||/cẀ:t.cẀ:t/||v||เนือย ๆ||/nW̄aj.nW̄aj/||0.042||*|
|จืด ๆ||/cẀ:t.cẀ:t/||v||เอื่อย ๆ||/PẀaj.PẀaj/||0.944|
|เอื่อย ๆ||/PẀaj.PẀaj/||v||เนือย ๆ||/nW̄aj.nW̄aj/||1.000|
|[End of table]|