Phraseology as indicator for translation quality assessment of museum texts: A corpus-based analysis

Phraseological competence, although traditionally left somewhat in the background in research on translation quality assessment, refers to a high level of knowledge of a language, and is therefore worthy of greater attention. Some studies, which are exceptions to this neglect, suggest that the level of phraseological quality of a text is directly related to its overall quality. This paper assesses these 2 types of quality in 14 original and translated texts from a parallel electronic corpus from museums and art centers in the city of New York, the aim being to test whether there is a correlation between phraseological and overall quality. In absolute terms, the results of the phraseological assessment are usually seen to be better than those from the overall assessment. On comparing the two analyses, and despite generally negative results and a few notable exceptions, there is a tendency toward this correlation. Subjects: Translation & Interpretation; Applied Linguistics; Lexicography


PUBLIC INTEREST STATEMENT
This article explores the translation of phraseological units (multi-word expressions, mostly collocations, idioms and proverbs) in museum texts from New York and places the review of these units in a central place within translation quality assessment. Although deemed one of the highest levels of command, phraseology has usually been a secondary component when assessing translated texts. This paper tries to prove whether a positive outcome in the assessment of phraseology means equally positive scores in the assessment of the elements that are typically assessed within professional environments: translation errors, omission and accuracy, terminology, grammar, orthotypography, punctuation, orthography and accent marks, style, register, typing errors, cultural adaptation, and coherence. Therefore, this paper constitutes a new approach, not only for aiming to find a possible correlation not widely explored so far, but also because museum texts, as crucial elements for intercultural and linguistic mediation, are becoming more relevant in recent specialized studies.

The phraseological issue within translation quality assessment
A number of studies have been conducted by translation scholars on the concept of the quality of translated texts. Since the early studies by such eminent scholars such as Newmark (1988Newmark ( , 1991 or House (1977House ( , 1997, and together with the contributions of other equally significant authors-some of the most outstanding, to cite just a few, being the work carried out by Lauscher (2000), Martínez Melis and Hurtado Albir (2001), Mossop (2001), Williams (2004), Colina (2008Colina ( , 2009, Angelelli (2009) or Reiss (2014)-we have come a long way in defining what is meant by translation quality and what parameters and factors are involved when it comes to measuring quality.
Likewise, and in view of the existing research, it is obvious that the concept of quality varies depending on the different perspectives from which this issue can be addressed. In this regard, then, it should be borne in mind that what is understood by quality in translator training (as dealt with by Way, 2008, or Galán-Mañas & Hurtado Albir, 2015 does not necessarily have to coincide with the quality sought in a professional translator, whether certified (cf. Koby & Champe, 2013;Stejskal, 2009) or not, where quality may be defined by the client (Mossop, 2011, p. 135). This is also how it is viewed by Gouadec, for whom "people and businesses have any number of criteria to judge and justify their judgment, ranging from 'punctuality' to 'proactivity' through 'compliance with the style guide' and 'initiative in upgrading the terminology'" (2010, p. 274). Additionally, according to some authors, quality in professional work maybe less decisive than other issues regarding money or time (DePalma, Pielmeier, Stewart, & Hegde, 2013, p. 64;Gouadec, 2010, p. 272). According to other sources, quality is taken for granted and is therefore not even open to debate: "quality is expected by clients and therefore not considered a differentiator" (PwC, 2012, p. 6).
In an altogether distinct but complementary sense, the way the quality of human translations is measured will necessarily have to vary with respect to the method used to measure that of machine translations (studied by, among others, Schwarz, Lacruz, & Bystrova, 2014;Temizöz, 2016) or even of computer-assisted translation (Jiménez-Crespo, 2009b). Likewise, there will be several approaches, depending on whether the quality assessment is performed by hand or either automatically or semiautomatically (see, for example, the studies by Babych & Hartley, 2009;Rabadán, Labrador, & Ramón, 2009). The aim of this study is to evaluate the phraseological quality of museum texts and compare it with their overall quality, with the intention of shedding some light on the apparent correlation in quality that exists between translated phraseological units and the remaining indicators subject to assessment, as pointed out by Buján Otero (2015, pp. 98-99). For the purposes of this work, phraseology is understood as the object of study of multi-word, not necessarily idiomatic, expressions, which constitute "(semi-)fixed, recurrent phrases" (Siyanova-Chanturia & Martinez, 2015, p. 549), such as collocations, locutions, proverbs, and speech formulae. Phraseology is considered to be a crucial element in language command-as Ferro Ruibal (1996, p. 104) stated, proficiency in phraseology is the highest level of proficiency in any language. Additionally, phraseology plays a critical role in translation competence and, more specifically, in the quality assessment of translated texts (see, for example, Colson [2008, p. 201], who claimed that "phraseology maybe one of the key factors in evaluating the quality of a translation").
In spite of its central position, references to phraseology in connection with translation quality assessment are scarce. However, it is worth mentioning some contributions which are partially devoted to phraseology and translation quality assessment, such as the works by Lee-Jahnke (2001, pp. 266-267), Heltai (2004, p. 61ff), Peña Pollastri (2009, p. 259), Mossop (2011, p. 138), Sardelli (2014, pp. 202-204), House (2015, p. 81, 130ff) and Huertas Barros and Buendía Castro (2017). Moreover, Mossop stated unequivocally that phraseology is "perhaps the main reason why revisers (even more than translators) should be native speakers of the target language" (2001, p. 109). Although the translation of phraseological units has traditionally been disregarded when assessing texts, there is no real reason for such a lack of attention, since phraseology is one of the items that must be taken into consideration according to some well-known translation quality standards-as https://doi.org/10. 1080/23311983.2018.1442116 is the case with ISO 17100:2015 (ISO, 2015, p. 16)-which safeguard the quality of the translation process.
When designing the bases of this study, one of the issues that was considered was to determine what kind of study to conduct: Should it be a quantitative or a qualitative study? Should it be based on an analytical or a holistic method? Although in recent years the number of studies that pay attention to the qualitative aspect has risen, sometimes as the backdrop of functionalist or pragmatic approaches (cf. Colina, 2008Colina, , 2009Colina, Marrone, Ingram, & Sánchez, 2016;House, 2015;Jimenez-Crespo, 2011;Martínez Mateo, Montero Martínez, & Moya Guijarro, 2016;O'Brien, 2012, to cite but a few of the most recent), the truth is that the quantitative approach in the assessment of translations is still valid. This can be seen in the appearance of new studies (e.g. Tsai, 2014) and in the professional world (cf. Hague, Melby, & Zheng, 2011, p. 258), where the assessment of translations, despite a series of proposals more in line with functional and pragmatic models, is still quite frequently based on quantitative models. With respect to the coexistence of the quantitative and the qualitative, House stated (although referring to the use of corpora in translation studies) that "these two lines of inquiry, the qualitative and the quantitative, are not considered to be mutually exclusive, rather they should be regarded as supplementing each other" (2015, p. 108).
In the case of this study, and given the characteristics of the model that was made available for use in the overall assessment (cf. 3.2.), the analysis will be of a quantitative nature. Since the view taken for the analysis is that of the finished product, the evaluation that was performed is of the summative type, 1 which, according to the classification put forward by Waddington (2001), follows an analytical method based on error detection.

Compilation of a parallel electronic corpus of English-Spanish museum texts. Selection of the texts
Translation corpora, as House put it (2015, p. 107), "are an important source for translation quality assessment". Although several significant studies on quality in the translation of museum texts have been published in recent years-for example, the cases of the work by Neather (2008Neather ( , 2009 or Jiang (2010)-studies about quality based on corpora of museum texts are scarce. Some studies highlight the advisability of using these tools (Neather, 2012, p. 207), while others actually use corpora to analyze museum texts. This is the case of the interesting contributions made by Guillot (2014, p. 83) or Valdeón (2015, p. 364), although they do not offer details of the characteristics of the corpora that were employed. This research was conducted by means of a corpus-based study, and to this end an English-Spanish corpus was compiled that is specialized, parallel (or multilingual, type A, according to McEnery & Hardie [2012, p. 19]), and balanced. The corpus genre is tourism texts, specifically museum texts, that have been published online, originally written in English, and translated into Spanish.
The corpus, called MUSA16 (from the merging of museums and USA and the year it was compiled, 2016) consists of texts from the 77 museums and art galleries in the New York City metropolitan area listed in Wright (2016, pp. 599-600) and covers the period between the year 1999 and July 2016. As it is a parallel corpus, it consists of two subcorpora: • MUSA16EN (with texts originally written in English), which consists of 151 files and, according to the analysis performed with Sketch Engine (Kilgarriff et al., 2014), contains 381,074 words and 452,424 tokens, and • MUSA16ES (with the Spanish translations of the texts in the English subcorpus), which also has 151 files, amounting to 429,019 words and 499,895 tokens.
As regards the way in which the corpus was compiled, the process was performed in several phases, which consisted in: identifying the museums (April 2016); finding the files, starting out from the translated text (April, May, and July 2016); recording of metadata and archiving the documents (June and July 2016); and aligning the files, using SDL Trados Studio 2014 (September and October 2016).
The corpus thus compiled consists of texts of a wide range of types, from mere guides with basic information for visitors to texts that are much closer to research papers, but also includes press releases and a countless number of educational and pedagogic materials. The lengths of the texts also vary greatly: 6 of the registers in English have fewer than a 100 words, whereas, at the other extreme, there are ten texts with more than 10,000 words. Although the corpus is not excessively extensive, this does not necessarily have to be a problem-as Sinclair (2001, p. vii) puts it, "The type of work will certainly be constrained by the corpus size, but has nothing to do with the quality". Furthermore, MUSA16 has the advantage of being parallel and aligned, which makes it easier to perform searches in it.
For the analysis procedure, fragments, each 1,000 words in length, were selected from among the original texts. The reason for taking this number of words as the basis is twofold: on the one hand, because 1,000 words, as stated by Biber (1993, p. 249), include many of the features that are essential for the study of texts. On the other hand, and despite the fact that the assessment models used allow compensatory measures to be applied in order to put the results on the same level, such fragments were selected with the aim of making it easier to compare them in absolute terms. Due to space constraints, a second selection of the texts had to be performed, as 50 of the texts in the MUSA16EN subcorpus, from a total of 14 museums, had at least 1,000 words. The selection criterion indicated that, of these 50 texts, one should be taken per museum, so that the sample studied would be as broad as possible; in the case of the 8 museums that have several texts with more than 1,000 words, 1 single text was chosen at random with the help of a computer program. Hence, the text base utilized in this study consisted of 14 text pairs from 14 museums, from the ST of which the first 1,000 words were selected. Table 1 shows a more detailed description of each of the text pairs used in the assessment. This table specifies the code that was assigned, the museum it came from, the title in English, the year of publication, and a description of it. The 14 text pairs that were selected formed the basis of this study, which, in general lines, consisted in a bilingual review (that is, checking the TT against the ST) of the phraseological units and, later, a bilingual overall review of the text.

Phraseological assessment
The first phase of the phraseological assessment consisted in extracting the phraseological units from the ST with IdiomSearch (Colson, 2016, p. 143). This tool, which is still in its beta version, was unable to perform a totally satisfactory detection of the phraseological units: what IdiomSearch sometimes identified as phraseological units in fact were not. Likewise, it occasionally neglected some items that were phraseological units, as in the case of curb excess (BROMEN01), 2 in part (FRICEN05), vantage point (GUGGEN07), on view (BROMEN01, MADNEN01, MJNHEN01, NYHSEN19), daunting task (MJNHEN01), more and more (QMOAEN01) and long view (SKYMEN01), among many others.
To check the results from IdiomSearch, a broad conception of phraseology was adopted (cf. Burger, 1998, p. 1), which means that the possible phraseological units can belong to three different spheres (collocations, locutions, and phraseological sentences) and, furthermore, that idiomacity is not a defining element of a phraseological unit. Moreover, for this analysis a criterion was established by which, for any phraseological unit from the ST to be considered as such, it had to appear as an entry in English monolingual dictionaries of general language or of phraseological units, collocations or phrasal verbs.
After identifying all the phraseological units in the 14 texts, they were classified according to the translation procedures that had been applied, namely, équivalence, paraphrase, omission, calque,  Rojo, 2013), an electronic template for assessing translated phraseological units, which, after applying the corresponding subtractions and additions due to the presence of errors (at two levels of severity) and positive solutions (also at two levels), issues a numerical result with a maximum score of 10 points. For the purposes of this paper, the number of positive solutions has been omitted with the aim of making the methodology of this template as similar as possible to that used in the overall assessment (3.2.). In view of the findings of the study by Temizöz (2016, p. 31), in which she observed a significant difference in the results of the assessments depending on whether recurring errors are counted or not, the decision was made to take the recurring errors into account in both assessments.
(4) since tax revenues are affected (SKYMEN01)/ya que afecta la recolección de impuestos  The joint study of the data shows a surprisingly high number of errors made in the 14 translations, involving 32% of the total number of phraseological units detected. Likewise, one striking finding is the fact that the procedure that contains the highest number of errors-without counting pseudoéquivalence, the occurrence of which is not representative-is calque ( Table 2): 77% of the cases of calques are wrong. Omission also has a high rate of error: 70%. Table 3 shows the errors detected in each text pair. The minor and serious errors have been grouped with the aim of making the data easier to read.

Overall assessment
The second phase of the analysis consisted in the overall assessment of the 14 text pairs, using an assessment model that allowed the results to be compared with those from the phraseological assessment.
Initially the researcher considered using models applied in the professional setting: LISA QA Model 3.1., SAE J2450, QAT, or the TAUS Dynamic Quality Evaluation Model, but this option had to be ruled out due to their limited availability and because it was considered that they did not fit the needs of this study satisfactorily. The next option considered was to use in-house templates employed by translation services providers. Despite the difficulties that this usually entails, due to restrictions resulting from the need for confidentiality, the researchers managed to obtain authorization to use the assessment model employed by a translation services provider, Hermes (Spain), for the revision and review of all the assignments for translation into Spanish performed in this firm by both inhouse and external translators. The model used follows, although only partially, the LISA QA Model 3.1. and the SAE J2450 (Arevalillo Doval, 2015, p. 314), which is a tendency that, as observed by O'Brien (2012, p. 56), is common in firms in the sector, although in reference to the first of the models. The LISA model is one of the most frequently employed in the industry, especially in localization (Lommel, 2007, p. 43;Stejskal, 2009, p. 299), despite having received perhaps more than its fair share of criticism, for example because of the limited number of error categories it offers-especially those related with language: just one with seven subcategories (cf. SDL TMS, 2015)-and also owing to the overlapping of some of the error categories (Jiménez Crespo, 2009a, p. 74;Thelen, 2009, pp. 203-204). Likewise, it has been applied, either alone or together with other models, in previous  (2016).
Of the eight error categories in the LISA QA Model 3.1., the Hermes model uses errors present in only three of them (document language, document formatting, and help formatting). These errors are defined with greater precision, in agreement with the aspects included in the standard ISO 17100:2015 on the translation process (ISO, 2015, p. 10). Consequently, the model utilized consisted of the following 13 error categories: Translation error, Omissions and accuracy, Terminology, Grammar, Orthotypography, Punctuation, Spelling and accent marks, Style, Register, Typing error, Cultural adaptation, Coherence and, lastly, Non-compliance with the brief. This results in the more accurate specification of the linguistic aspects, albeit only partially, while omitting the errors detected in LISA QA 3.0, which are more closely related to extralinguistic aspects of localization.
The main drawback of SAE J2450, the other standard taken as the basis of the method, is that it is only designed to detect linguistic errors and overlooks any issues related to format or style (O'Brien, 2012, p. 61;Stejskal, 2009, p. 299). Although it was initially created for the US automotive industry (SAE J2450, n.d., p. 1), it is also currently used in other fields (Martínez Mateo, 2014, p. 78). The Hermes assessment model coincides to a certain extent with the SAE J2450 model, as regards both the levels of severity (it reflects minor and serious errors, but not the critical errors of the LISA model) and the weighting of those errors (SDL TMS, 2015), although the coincidences are only approximate. Moreover, the Hermes model includes a series of values that are set by the reviewer and affect the scoring of the review: time taken, average amount that can be reviewed per hour, and difficulty of the translated text. In order to make the results comparable with those from the phraseological assessment (which does not offer these customization options in its template), the same values were applied in these fields for the 14 text pairs. The interpretation of errors and levels of severity was performed following an internal document from Hermes which explains the way the model works. Furthermore, the typology of errors proposed by Jimenez-Crespo (2011, pp. 321-328) was also a valuable aid, since the text pairs studied share many of the features of the localization of corporate websites examined in that study. Lastly, the year of publication of the texts that were assessed was taken into account (see Table 1), in order to determine whether the new grammatical, orthographic, and orthotypographical rules of the Real Academia Española should be applied in the assessment (2009,2010).
The global results of the overall assessment yielded the following data: 735 errors were detected, of which 27 were serious and 708 were minor. Table 4 shows the results of the overall assessment by number of errors. The serious and the minor errors are shown jointly in order to make the data easier to read. The category "Non-compliance with the brief" was not included in the assessment, as this was not applicable for this study.
From Table 4, several conclusions can be drawn prior to the comparison of the two assessments. First, of all the errors detected, 735, a high percentage of them (33%) belong to the category "Orthotypography" (10). This percentage increases to 45% if three others related to the foregoing category are added: "Punctuation" (11), "Orthography and accent marks" (12) and "Typing error" (13).  (13) From the more than 400 entries received, nearly 100 award-winning shorts, features, and documentaries will be screened (NMAIEN03)/De las meas de 400 obras recibidas, cerca de 100 premiados cortos, largometrajes y dcumentales se presentarán (NMAIES03) [más, documentales] These are followed in number of errors by the categories "Grammar" (14) and "Omissions and accuracy" (15), which account for 16% and 13% of the total, respectively. The category "Terminology" (16) yields proportionally fewer errors than could be expected-8% of the total, which may be due to the low level of specialization generally observed in the 14 text pairs.

Analysis and comparison of the scores of the two assessments
The two assessment models, phraseological and overall, allow a maximum score of 10 points but with no minimum score, as results below zero can also be obtained. The results of the ratings are shown in Table 5, so as to allow the analysis of the scores obtained in the two assessments. In this table, the texts are presented in descending order: on the left, according to the score obtained in the phraseological assessment; and on the right, they are ordered according to the score from the overall assessment. Furthermore, in order to make the data easier to interpret, two horizontal lines have been drawn to separate the first five texts (six on the left due to two of them having the same score) and the last five texts.
The analysis of Table 5 and its comparison with Tables 3 and 4 allow some relevant conclusions to be drawn for this study, which are detailed in the following.
Firstly, on analyzing only the scores from the phraseological assessment with PhrasQA (left-hand side in Table 5), it can be seen that they exceed 5 points in 10 of the 14 text pairs. Yet, although the phraseological quality is more or less acceptable in general terms, the results are low, as this score only takes into account one single assessment criterion. Moreover, it should be remembered that, according to the data shown in Table 3, the number of errors is high even in the texts that obtain better scores: 7 and 9 errors in the 1,000 words reviewed from the pairs MOMA17 and BARR21, respectively. The three text pairs that yielded the poorest results are the ones in which, moreover, the greatest number of phraseological units were detected: 71 units in BROM01, 62 in AMNH08, and 63 in SKYM01. Although it could be claimed that the more units there are in a text, the greater the possibility of committing translation errors will be, this does not appear to be an entirely decisive factor in the texts analyzed in this study: the pair with the fourth poorest score, NMAI03, ranks second to last in terms of the number of phraseological units it contains: 38. Conversely, the number of phraseological units in the best text pair, MOMA17, is slightly higher than the average, as it contains 51 units.
Secondly, the scores from the analysis with the Hermes model, for the overall assessment (righthand side in Table 5), offer very negative results in general terms. It is interesting to note not only the fact that MOMA17 is the only text pair that obtains a score above 5 points, but also that 9 of the 14 pairs obtain negative scores. Moreover, a review of the number of errors recorded for each text pair (Table 4) shows the large number of errors committed, even in the best pair, in which 13 errors were detected. As pointed out earlier, many of the errors belong to categories related with orthographic and orthotypographical aspects; nonetheless, this should not lead us to think that the reason for these scores lies only in the poor orthotypographical quality of the texts. If such aspects had not been taken into account, the results would have still been unsatisfactory: only 3 out of the 14 text pairs (MOMA17, GUGG07, and METM16) would obtain a score above 5, whereas the last in the results shown in Table 5, FOLK02, would continue to rank last and, again, with a result below zero.
Lastly, the main aim of this study was to determine whether, in the texts from the MUSA16 corpus that were analyzed, there is a correlation between the phraseological quality and the quality of the other levels that are usually assessed in translations. On observing Table 5, it can be seen that the text with the best results from the assessment performed with PhrasQA, MOMA17, is also the best in the results from the Hermes model. Likewise, and having observed that the results of the overall assessment are negative, another three of the six best texts as regards phraseology are also the best in the overall aspects (METM16, FRIC05, and GUGG07). In contrast, it can be seen that two texts with good scores in the phraseological assessment obtained a very negative score in the overall assessment. In the case of BARR21, this is largely due to the very high number of inaccuracies and grammatical errors (see Table 4). In the case of MADN01, again there are a number of grammatical errors, in addition to a large number of translation errors (the second poorest result of the 14 text pairs), which has a negative effect on the comprehension of the text.
The relative correlation observed in the upper part of Table 5 is even more apparent in the lower part: four of the five poorest text pairs in the phraseological assessment were also the poorest in the overall assessment (NYHS19, NMAI03, SKYM01, and AMNH08). Conversely, the case of BROM01 also stands out from the rest, as it obtained the poorest result in the phraseological assessment but the fifth best result in the overall assessment, albeit with a modest score. Therefore, the assumed correlation that this study aimed to confirm does occur, but only relatively. Although several striking exceptions have been detected, there does seem to be a certain tendency toward the phraseological quality of the translated text being matched by an at least similar degree of quality on the other levels of what, in this study, has been called the overall assessment. Unfortunately, the highly unsatisfactory result of the assessment of general aspects does not favor being able to draw more categorical conclusions, although it does allow us to note the tendency indicated above.

Conclusion
Throughout the previous pages it has been seen that phraseology, despite being a recurring element in practically any translation, usually goes unnoticed in assessment models and in studies conducted on translation quality assessment. In research that does pay attention to it, however, its importance is highlighted almost unanimously.
The corpus used in this study was MUSA16, made up of 151 texts written in English with their corresponding translations into Spanish. They are texts from the museums and art centers in the city of New York that were published between 1999 and 2016. For this study, 14 text pairs on a variety of topics were selected: three of them provided information about activities, three gave information about exhibitions, four were press releases, and four others were educational material.
In general, the phraseological analysis of the 14 text pairs yielded unsatisfactory results: as a consequence of the high number of wrongly translated phraseological units, the results of the phraseological assessment were low in most cases. Additionally, the procedures of calque and omission, which are not negative per se, displayed a large number of errors in the translations analyzed.
As far as the overall assessment is concerned, the results are poorer than those of the phraseological assessment, partly due to the fact that a greater number of parameters were subjected to assessment. A detailed look at the errors detected, moreover, reveals that a high percentage of them are related with orthographic and orthotypographical elements. Errors concerning grammar, as well as omission and accuracy, also show negative figures, whereas the incidence of terminological errors is lower, probably due to the low level of specialization of the 14 text pairs that were analyzed.
On comparing the two assessments, it can be seen that the best text in the two assessments is the same one: the fragment from the Museum of Modern Art. Furthermore, the fact that the texts that achieve the best results in the phraseological assessment also generally do the same in the overall assessment and, more clearly still, the poorest at one level are also the poorest at the other, allows us to conclude that the correlation between phraseological quality and overall quality does, in general terms, occur in these 14 text pairs. Nonetheless, several significant exceptions are observed that warrant further study.
Although it was not a primary aim of this study, the analysis of the 14 text pairs has revealed that, generally speaking, they do not have the same level of quality as the corresponding source texts or the quality that is presupposed of texts that are intended to reflect the culture, thinking, and reflection that today's museums seek to transmit. With museums and art centers holding exhibitions and arranging activities not only inside their buildings, but also "anywhere and at any time through websites and social software" (Monti & Keene, 2013, p. 27), it is clear that more care needs to be given to the quality of these texts than that evidenced in this study. https://doi.org/10.1080/23311983.2018.1442116 in examples 4-21 possible solutions to the translation errors are given between square brackets. 4. Due to the nature of the research, one procedure was left out of the study: that of compensation, which consists in using a phraseological unit in the TT as a translation option to deal with a lexical unit or nonphraseological syntagm. This does not mean that it is not used with relative frequency in the translated texts.
In this respect, of the numerous cases of erroneous phraseological compensation found, one example that is particularly striking is the following, in which the verb spawn has been translated as the verb phrase dar a luz ("give birth"), but it is introduced in the wrong place in the sentence, without the preposition a, which is mandatory in the Spanish prepositional system: high rents that spawned downtown's highrise buildings (SKYMEN01)/alquileres altos y dieron a luz las torres de "downtown." (SKYMES01).