What happens in translation? A comparison of original and translated texts containing verbs meaning SIT, STAND and LIE in the English-Swedish Parallel Corpus (ESPC)

This article studies translation effects by comparing the use of verbs meaning SIT, STAND and LIE in original and translated texts in the English-Swedish Parallel Corpus (ESPC). Effects on both the frequency of use and the use of lexical and structural translation shifts are studied. Postural verbs have a much higher frequency overall in Swedish than in English. In Swedish translated texts, postural verbs are significantly under-represented in comparison to original texts, whereas postural verbs are significantly over-represented in English translations. At a more fine-grained level, it is possible to show that various categories are treated differently, in particular the types of subjects. The effect on frequency is stronger for Human subjects, which represent an unmarked category, than it is for Inanimate subjects, which are more marked. However, the pattern of overand under-representation presupposes a functional overlap across languages. Writing as subject, which is a category that is unique to Swedish in relation to English, follows a different pattern. (This type of subject appears in examples such as: Nyheten står i tidningen ‘The news is (literally: stands) in the paper’). The result is discussed both from the point of view of the research methodology used in contrastive studies based on translation corpora and from a theoretical point of view. For methodology, the conclusion is that frequencies can be considerably skewed, whereas a language remains true to its system of basic semantic contrasts in professional translations. Theoretically, the result can be related to theories of language contact and studies of second language acquisition and bilingual development.

situations of language contact, translated texts have been, and are, important forces behind language change -sometimes in conjunction with the influence from second language users of the target language.The present paper, focusing on postural verbs in original and translated texts in English and Swedish, will give only partial answers to some of these overarching questions, but it is important to keep the broader picture in mind.In addition to theoretical questions, an important aim of the paper is to test the methodology in contrastive studies based on translation corpora.The advantage of using translation corpora is that the expression of the same meaning in the same context can be compared across languages.On the other hand, the fact that translations do not always give a representative picture of original target language texts must be taken into consideration.In particular, it is important to identify which features are most vulnerable.

Why look at postural verbs?
The justification for looking at postural verbs in particular is that these verbs have been studied extensively from a typological (Newman 2002) as well as from a contrastive point of view (Lemmens 2002, Svensson 2005, Viberg 2013) and also from a second language acquisition perspective (Lemmens & Perres 2010).Of particular importance for the present study is the detailed and well-documented contrastive study by Kortteinen (2008) comparing the Swedish postural verbs sitta 'sit', stå 'stand' and ligga 'lie' with their French correspondents.That study contains a comparison of the use of the Swedish postural verbs in original and translated texts and shows that all three Swedish verbs are underrepresented in translations.
In French, the salience of manner is low in descriptions of motion and location.The closest correspondents to SIT, STAND and LIE are not simple verbs in French and are used with a much lower frequency than in Swedish.English and Swedish, on the other hand, share historical cognates that still are very similar in form: sit/sitta, stand/stå and lie/ligga.The relationship between the postural verbs in English and Swedish represents a very special case that is likely to lead to maximal cross-linguistic influence.Semantically, not only the core meanings in reference to human posture are similar, but there is also a semantic overlap in several cases when postural verbs refer to the location of physical objects (Viberg 2013).From a methodological point of view, this represents a suitable area for testing the validity of translation-based contrastive studies.

Earlier research
The answer to the question of what happens in translation can be approached from many different perspectives.The present study takes corpus-based contrastive studies (Johansson 2007) as a point of departure and confronts them with certain aspects of translation studies (Gambier & van Doorslaer 2010, Malmkjaer & Windle 2011, Munday 2008, Halverson 2003, 2010) and in particular corpus-based translation studies (Laviosa 2002).From a contrastive point of view, this study approaches what Johansson (2007, 32-33) refers to as translation effects, i.e features which in some way make translated texts different from original texts in the same language.Johansson's term is neutral in comparison to the more value-laden term translationese used in Gellerstam's (1986) seminal article.An important issue is how to cope with translation effects from a methodological point of view in contrastive studies based on multilingual translation corpora such as Viberg (2013).
One specific type of translation effect, which has been much discussed in translation studies, is translation universals (Baker 1993, Mauranen & Kujamäki 2004), features that are characteristic of translated texts in general, irrespective of the source language.Examples of possible such effects are normalization and explicitation (see Malmkjaer 2011 for a balanced review).Typically, studies of translation universals are based on comparison between originals and translations of the same language.The present study will primarily be concerned with translation effects characteristic of specific source-target language pairs and is closely related to studies of interference, transfer or cross-linguistic influence in bilingualism and second language acquisition research (see Jarvis & Pavlenko 2008 for a broad review, cf.Viberg 1998).In contact linguistics, translation effects are an important source of convergence and divergence between languages (see Braunmüller & House 2009 and, for translation, in particular Becher, House & Kranich 2009.)As is often the case in corpus-based studies, this study treats translation as a collective process without taking individual translators into consideration (see Johansson 2007, 197-215 for an exception).This is justified by the general importance of translations as a societal phenomenon.What happens in translation can also be studied from the perspective of the translator (the producer) in studies of the translation process, which are concerned with what happens in the translator's mind.Such research is based on methods such as think-aloud protocols, eyetracking and keystroke logging (see Englund Dimitrova 2010 for a concise review).A corpus-based study can also -albeit more indirectlyshed light on the translation process seen from the translator's perspective, in particular in studies of translating strategies (Gambier 2010).

Translation patterns
In this paper, the closest correspondent to a word (or multi-word unit) that appears in a translated text will be referred to as a translation (with a corresponding plural form: translations).Broadly, translations will be divided into direct translations (the corresponding postural verb in this case), lexical translation shifts, when some other word is used, and structural translation shifts such as restructuring and structural reduction.In addition, there are free translations, where the translation differs so much that it is impossible to identify any systematic structural relationship with the original text.A further category is no translation, which refers to examples where an extensive text has not been translated and the translation is lacking for reasons that are not structurally motivated as in reductions.A systematic description of all translations of a word along these lines will be referred to as the translation pattern of the word.

The corpus and its composition
This study is based on the English-Swedish Parallel Corpus (ESPC) compiled by Altenberg and Aijmer (2000) according to the same principles as the English-Norwegian Parallel Corpus (see Johansson 2007, 9-21).The ESPC corpus consists of original printed texts in Swedish and English together with their translations.The texts are divided into two broad genres: Fiction and Non-fiction with several subcategories.The sizes of the various sub-corpora of the ESPC are shown in Table 1 (see Electronic sources for more information).The corpus allows comparisons of various types.It can be used as a comparable corpus to compare originals in both languages in a contrastive study (see Section 2.1).Contrastive studies often also look at the translation patterns of the linguistic elements as a way of getting a better grasp of the meaning of the words that have been singled out for a study, for example all English translations of Swedish stå 'stand'.This means that originals in one language are compared with translations in the other language.The ESPC can also be used for corpus-based translation studies, which involves in particular the language-internal comparison of originals in one language with translations into the same language.The corpus also allows comparison between two genres (or registers): Fiction and Non-Fiction.This is relevant also for the present study, since the distribution of uses (or meanings) of the postural verbs varies dramatically in certain respects between Fiction and Non-Fiction.
For reasons of space this aspect will be treated briefly in Section 6.The rest of the paper is structured as follows: Section 2 presents global comparisons, which means that various uses (or senses) of the postural verbs are not taken into consideration.Section 3 presents a semantic classification of grammatical subjects on which the following quantitative analysis is based.In Section 4, this classification is applied in a quantitative study of the distribution of direct translations across subject classes.Section 5 is devoted to a fine-grained descriptive comparison of lexical and structural translation shifts, where grammatical constructions are also taken into consideration.Section 6 looks at the distribution of the major uses across genres (represented by fiction and non-fiction) followed by conclusions and discussion in Section 7.

Swedish and English originals
Table 2 compares the frequencies of the postural verbs in original texts.The result is also shown in Diagram 1.As can be observed, all of the Swedish verbs are considerably more frequent than their English correspondents.All of the differences are statistically significant (p< 0.0001).Statistical testing has been done with "The Significance of the Difference Between Two Independent Proportions" (Vassarstats, see electronic sources), which calculates the z-ratio for the significance of the difference between two independent proportions.This test as well as the following ones are based on raw scores, whereas the diagrams presented below show the frequencies per 100 000 words.Contrastive relationships: • In original texts, the Swedish postural verbs are much more frequent than their English direct correspondents in spite of the fact that all verbs are cognates which are transparently related in form and share core meaning.
Cross-linguistic influence: • In translations from Swedish, English sit, stand and lie are significantly over-represented.• In translations from English, Swedish sitta, stå and ligga are significantly under-represented.
At the global level, this points to a striking difference between originals and translations in the same language.However, this case cannot be generalized in a simple way.It turns out that other frequent lexical verbs do not always follow a similar pattern.The postural verbs were chosen as an example to test what happens in translation because they appear to represent a worst-case scenario from the point of view of contrastive studies based on translation corpora.(However, this applies only to mono-directional comparison.Bi-directional corpora like the ESPC can also be used to compare originals.) From a general theoretical standpoint, it is interesting to get a more differentiated view of what happens.Are all uses (meanings) of a frequent polyfunctional word affected in the same way, or do we find translation effects only for some uses?What are the correspondents, when the postural verb does not correspond to a postural verb in the other language?The result so far seems to suggest severe translation effects, which does call for caution when using translated texts for contrastive comparison, but, as the results will reveal, a language stays true to its basic system of contrasts in professional translations and a comparison based exclusively on originals in comparable corpora misses important contrastive relationships.However, first a specific methodological problem must be addressed in the next section.

Variation due to genre: Fiction versus Non-Fiction
Variation due to genre (or register variation in general) is reflected in the use of individual words, but rather few of the many studies that, like the present one, deal with the polysemy of frequent words have tried to address this as a problem.Diagram 3ab shows the distribution of the postural verbs across the two genres of the ESPC.There turns out to be very clear and significant differences.Both in English and Swedish, the postural verbs are considerably more frequent in fiction than in non-fiction.The results are in line with the characterization of fiction as a distinct register (Ebeling & Ebeling 2013) and in particular with Stubbs & Barth's (2003, 65) conclusion that "verbs of physical movement, mental activity, and nouns for part of the body" are characteristic of fiction.
2 Non-fiction is negatively defined in the ESPC and even in actual practice covers a very broad variety of texts that serve as an interesting contrast to fiction, but need to be broken down into more clearly defined categories.(See, for example Biber et al. 1999, who compare fiction with conversation, news and academic prose.)By comparing the frequencies in Diagram 3a and Diagram 3b, it is clear that each of the Swedish verbs is more frequent than its English counterpart in both fiction and non-fiction.This means that the conclusion reached in Section 2.1 concerning the contrastive relationship holds across registers (as far as we can tell from the ESPC corpus).As will be shown below, the register differences are more problematic when specific meanings or uses are taken into consideration, since certain uses are actually more frequent in non-fiction than in fiction.As a result, it is an open question whether fiction and non-fiction should be accounted for separately from the very beginning or not.In the coding of individual examples, the distinction is made, but the general account will present data from the whole corpus indiscriminately, and register differences will be discussed in Section 6.

Semantic classes of subjects
In order to test whether the global differences between originals and translations depend primarily on any specific meaning or are more evenly distributed across meanings, a functional-semantic coding of all examples was carried out.A natural starting point was the semantic category of the subject.The coding was done at different levels of granularity to make it possible to identify classes that were general enough to allow statistical comparison.At the coarsest level, four types of subject were distinguished: Human, Animal, Inanimate (concrete) and Abstract (see Section 3.1).Different types of more fine-grained coding will be discussed in Section 3.2.

The four major classes
In its most frequent use, SIT, STAND and LIE have a human subject and refer to the posture of a human functioning as subject of the verb.Often, a locative complement is added.The Swedish verbs can also freely be used in presentative constructions (see ( 1)), which have the basic structure det 'it' + Verb + NP indefinite .All the three Swedish postural verbs can appear in this structure with the same postural meaning, as in the corresponding non-presentative clause.

3
(1) Det står en man här ute som nödvändigtvis vill tala med dig.(HM2) [It stands a man] "There's a man out here who insists on talking to you." (HM2T) In example (1), a man is the subject.To be exact, the classification is based on the located entity in cases like this one, but since the located entity in the majority of cases is realized as grammatical subject, this has been used as a general term in order not to complicate the terminology.In a few cases such as (2) and (3), when metonymy is involved, the choice of verb is decided by a category that is clearly different from the grammatical subject.In the Swedish original version there is a human subject, but the verb actually refers to the location of that person's car and boat, respectively.A similar type of metonymy is involved in: I'm parked outside.In example (3), she obviously refers to a boat in the English translation.
(2) Kan en av er komma med mig och se om jag står bra där jag står.Animal subjects were so few that it is not meaningful to include them in most of the comparisons.It is tempting to collapse this class with human into an animate class, but the naming of animal postures follows different rules and certain animals have a default verb (birds sit in trees, snakes lie in the grass etc.).
Both English and Swedish can use the postural verbs to denote the location of inanimate concrete subjects.When the verbs meaning STAND and LIE are used, the basic semantic distinctions are similar, but there is a rather strong tendency to express location with the copula in English, whereas such uses are very restricted in Swedish.In both languages, LIE basically refers to an object placed in such a way that it has a maximal extension in the horizontal dimension, whereas STAND refers to a maximal vertical extension.In examples (4) and ( 5), the closest Swedish equivalent is used as a translation.
(4) She came forward to open a folder that lay on the counter.(AT1) Hon gick fram och slog upp en pärm som låg på disken.(AT1T) (5) The father took a golf-club that was standing in the corner.(RD1) Hennes pappa tog en golfklubba som stod i ett hörn.(RD1T) In Swedish, a special case is represented by physical objects (in particular artefacts such as a plate) that have a distinctive upper side, when they are used in a canonical way.The location of such objects is referred to with stå, even when the horizontal extension is dominant (see example 6).
Functional upper side appears to cover the same cases as base in other accounts (e.g.Lemmens 2002) but is based more on the typical human interaction with physical objects.

(IB1T)
[There is a tense shift in the translation from past to present.] This special use is referred to as functional upper side in Table 3, which shows the basic contrasts in Swedish.Even when the subject is inanimate, the located entity is sometimes not explicitly referred to due to metonymy, for example, contents for container as in the Swedish original in ( 7), which literally means: 'It stood wine on the table'.The choice of verb in the Swedish example is based on the orientation of the bottle and is coded accordingly.In the English translation, the bottle is referred to explicitly.

(KOB1) [it stood wine on table-the]
There was a bottle of wine on the table, and a salad.(KOB1) SIT represents a special case in both languages.According to Newman (2002:19), English sit basically has two functions with inanimate objects: (1) Non-activity sit "with a nuance suggesting that the entity is underutilized, out of use, useless etc." and (2) Good-fit sit, a position which represents a tidy arrangement or "good fit".Shape and orientation are irrelevant.An example of "good fit" sit is given in (8) and of "non-activity" sit in (9).In both cases, the translation is a verb that basically describes a different posture.In Swedish, sitta refers to an attachment of some kind.Both stå and ligga refer to situations where a physical object rests on a supporting surface, whereas sitta is typically used when such support is missing.The attachment can be of various types such as adhesion (see 10) or a tight fit that "holds" an object in its place (see ( 11)).The role played by the concept of attachment is very language-specific in Swedish and is not lexicalized in the same way in closely related languages such as German and Dutch (Viberg 2013: 160-163).
Both the English and the Swedish postural verbs are frequently used with abstract subjects (Inflation stands at three percent), but such uses are based on conceptual metaphors, which are often lexicalized as more or less fixed phrases.For that reason they follow different types of rules and require separate treatment.To a certain extent uses with such subjects are shared between English and Swedish, but very often translations contain verbs from more abstract fields and the translation is in many cases rather free, as in ( 12).Various types of subject are more or less marked or prototypical.Human subjects represent the unmarked or prototypical category.The use of postural verbs with Human subjects is conceptually basic and the use with other types of subject is derived via metonymy and/or metaphor.

More specific categories
It is possible to draw finer distinctions and to distinguish a number of further subcategories, but only a few more specific categories will be discussed.In many other cases, the examples are too few to make a quantitative comparison meaningful.The categories that will be identified are "robust" and can be coded and counted in a straightforward way.The category inanimate (concrete) subject can be divided into a number of subcategories.The basic contrasts represented in Table 3 apply most clearly to physical objects that can be held in the hand and manipulated, such as bottle or plate.Another important subcategory is building.
According to Newman (2009, 46-52), stand and sit are used in English to refer to the location of buildings, and both of these alternatives occur in the ESPC corpus.In Swedish, ligga is the dominant alternative as in the translation of example (13).Stå can also be used, when the subject refers to a building, but that is a less frequent alternative.
(13) The house itself was medium-sized, unexceptional to look at, standing on a street of such houses in an older part of Baltimore.(Animal is too small to be meaningfully discussed.)Note in particular the under-representation of the category Writing, which is relevant only for Swedish stå.
The relationships between originals and translations in English are shown in Diagram 4b.Human and Inanimate subjects are overrepresented.Writing, on the other hand, which is a category that is unique to Swedish (with respect to English), only occurs in one translated example (0.13 per 100 000 words), and this means that calques or loan translations are practically non-existent in this category.Human and Inanimate subjects, which have overlapping functions in English and Swedish, differ with respect to frequency of use in originals and translations.In Swedish, no translation effect was found for Abstract subjects and as will be evident, Abstract subjects do not follow the general pattern of other verbs either (for some discussion, see Section 6).Swedish than in English originals.In spite of that, both categories of use must be said to be clearly established in English.The overlap of functions appears to be a major explanation for the strong cross-linguistic influence on the translations.

LIE
Human subjects follow the same pattern for LIE as for STAND.They are under-represented in Swedish translated texts and over-represented in English translated texts.The same applies to Inanimate subjects taken as general category.However, in this section an attempt will be made to see what happens if different subcategories of Inanimate are distinguished.
The reason for this is that ligga can be regarded as a default category for describing the location of inanimate subjects in Swedish.The frequency of occcurrence per 100 000 words for inanimate subjects of ligga is 56 in originals, which is higher than for human subjects of ligga and higher than for inanimate subjects of stå and sitta.
An inspection of the Swedish originals in Diagram 5a shows that physical object, building and place are well established in Swedish.In translations, Human is significantly under-represented.As for the subcategories of inanimate subjects, only Building and Other are significantly under-represented in the translated texts if tested individually.(Inanimate as a whole will be treated in 4.4.) It is important to note that all the subcategories of Inanaimate distinguised in the analysis of Swedish also appear in the English original texts (see Diagram 5b), but all categories have a considerably lower frequency than in Swedish (compare Diagram 5a).There is substantial functional overlap between the systems, but salient differences with respect to frequency of use.In English translated texts, Human and the Inanimate subcategory Physical object are significantly over-represented, whereas Place and in particular Building (p just above 0.05) appear to be under-represented, but the data are inconclusive with respect to categories other than Physical object due to the low frequency in both originals and translations.

SIT
The categories of Swedish sitta are presented in Diagram 6a.With respect to the use with inanimate subjects, sitta represents the most language-specific category among the postural verbs (see comments to Table 3

Contrast and convergence
Repeatedly, the clearest difference has been found between Human and Inanimate subjects.In order to illustrate this, the frequencies of occurrence of human and inanimate subjects are compared for all verbs in Diagram 7. In originals (Diagram 7a), there is contrast.Both human and inanimate uses are much more frequent in Swedish than in English, and this applies to all verbs.All of these differences are statistically significant except for SIT when it has an inanimate subject since it is too infrequent in English, but the difference is still clear.(The raw scores are 58 for sitta vs 4 for sit).
In translations (Diagram 7b), there is a significant statistical difference between the frequencies of human subjects only for STAND, and even in that case the difference is proportionally much smaller than for the inanimate subjects, which all are significantly more frequent in Swedish than in English. 4There is a tendency for Human to converge in translations, whereas there is still a clear difference between the two languages in the Inanimate category. 4The reason why it is significant for SIT in translations but not in originals is that the raw scores are 37 vs 6 and that the test requires both scores to be equal to or higher than 5. words using the categories discussed in Section 1.4.For reasons of space, only STAND will be systematically described.The general patterns are similar for LIE and SIT (see also Viberg 2013 for a contrastive study of the Swedish postural verbs).

Lexical shifts
The lexical shifts related to English stand when it is translated into Swedish are shown in Table 4. (The structural shifts will be commented on in Section 5.2.)As would be expected, direct lexical translations dominate.Since it turns out that direct lexical translations are much less frequent when the postural verbs are used figuratively (Table 4), literal and figurative uses are accounted for separately when the subject is Human or Inanimate as in ( 22) and ( 23).It should be noted that Figurative refers to the interpretation of the whole sentence.The subject in ( 22) still literally refers to a human.In ( 23), the subject tank-trap is used in a figurative sense, but the subject is classified as inanimate rather than as abstract since the choice of postural verb is still decided by the type of concrete inanimate referent that is referred to when tank-trap is used literally.Metaphorization applies to the whole proposition 'tank-traps lie ahead'.
The major lexical shifts shown in Table 4 are ställa sig upp and resa sig upp, which refer to change-of-posture in Swedish.In English, changeof-posture is primarily formed by combining sit, stand and lie with a directional particle (e.g.lie down, stand up.See Newman 2009: 36-46 for discussion and some exceptions).One major difference between English and Swedish is that Swedish sitta, stå and ligga are not generally used to denote change of posture.Instead, a reflexive form of a related causative verb is used as in (24) or in some cases as in (25) a reflexive form of another action verb (resa sig 'rise', a reflexive of resa 'raise').The directional particles upp and ner 'down' are optional in Swedish with a few exceptions as indicated in Table 5, which shows the dynamic system of Swedish posture verb (state/activity -postural changecausative).There are also a few examples where a Swedish postural verb is used in combination with a spatial particle to describe change of posture.This occurs in particular in the imperative form (Stå upp! (LH1) -On your feet!(LH1T)). 5 Table 6 shows what happens when Swedish stå is translated into English.A direct translation is the most frequent alternative when the subject is human or inanimate and does not have a figurative meaning.Direct translations are used with a very low frequency in figurative uses with human subjects and with abstract subjects.Such translations require separate treatment and are only briefly discussed in this paper.
When the subject is human or animate, the second most frequent translation is be, used in various functions.This verb thus represents the most frequent type of lexical shift.The use of be to indicate the location of the subject is referred to as "Locative be" in Table 6.Proportionally, locative be occurs more often with inanimate than human subjects.The same applies even more to there is (in various inflected forms), which is treated separately and has been classified as a lexical shift.Sometimes it also involves a structural shift, when the Swedish verb does not appear in a presentative construction.(Although relevant, this is not shown in the table, since perspicuity would be lost if too many subcategories were to be introduced.) In ( 26), the Swedish posture verb is used in a copula-like construction with an adjective or past participle (see 27) as a predicative complement.Such constructions occur both with human and inanimate subjects, which are coded separately but will be briefly commented on together in this section.
( In most cases, English uses the copula as a translation ("Copula be" in Table 6), but English postural verbs can also appear in a copula-like construction.Example ( 28) is taken from an English original text.

Structural shifts
Structural shifts can have an impact on the occurrence of lexical correspondents in the translation.This section will discuss the types of structural shifts that directly affect the occurrence of postural verbs in the translations, especially various types of reductions that are characteristic of translations from Swedish into English.The clearest example is the reduction of relative clauses to prepositional phrases (or occasionally to some other type of place adverbial): som (Relative marker) + Postural verb + PP Locative > PP Locative (for short: Rel > PP).Relative clauses are typically introduced with the indeclinable relative marker som, as in ( 29).
(29) En av orsakerna var hans far som bodde ensam i ett hus som låg på slätten strax utanför Löderup.(HM2) [that lay on the plain] One of the causes was his father, who lived alone in a house on the plain just outside Löderup.(HM2T) There are 18 shifts of this type when stå is translated into English.This kind of reduction is mirrored by expansion when English is the source language as in ( 30).(There are 12 such expansions with stå.)The correspondents of sitta behave in a similar way as the other postural verbs (10 reductions of the type Rel > PP in translations from Swedish into English and 17 expansions PP > Rel in the opposite direction).Somewhat surprisingly, translational shifts involving relatives work in a direction that reduces the characteristic degree of over-and underrepresentation.In English translated texts, for example, postural verbs are over-represented in spite of the fact that reductions cut down the number of postural verb correspondents.The shifts involving relative clauses probably reflect general structural differences between English and Swedish, since relative clauses are very easy to form in Swedish and probably also are more frequent than in English. 6 Another general structural difference affecting the translation patterns of the postural verbs are various types of participial constructions.Sometimes Swedish relative clauses are reduced to participial constructions in English, as in (34).
(34) Trots att han försökte leva så anonymt som möjligt, i detta hus som låg strax öster om Ystad, hände det att nyfikna människor höll honom under uppsikt.(HM1)[this house that lay just east of Ystad] Even though he tried to live as anonymously as possible in this house located just east of Ystad, sometimes curiosity-seekers spied on him.(HM1T) Such constructions are not completely ruled out in Swedish but usually sound very formal (ett hus beläget strax öster om Ystad).There is another type of structural shift that involves a past participle, where a postural verb is translated with a resultative expression as in ( 35).

(GAPG1) [The wine table stands]
The wine table is placed in the bedroom of Karl XIV Johan.(GAPG1T) Some expressions included into this structural shift such as be located (see ( 36)) and be situated, which are used primarily as translations of ligga with places and buildings as subjects, are so conventionalized and have such a bleached resultative meaning that it would be justified to regard them as purely lexical (multi-word) shifts.
(36) Den hette Ercildoune och låg utanför Centerport (JMY1) [lay outside Centerport] It was called Ercildoune and was located outside Centerport (JMY1T) Non-finite ing-clauses and ed-clauses in various syntactic roles (Biber et al 1999, 199-200) can very often not be translated with the corresponding present and past participles in Swedish without sounding unidiomatic.This lack of correspondence relatively frequently leads to reductions and expansions of various types.In ( 37) and ( 38) the postural verb disappears in the Swedish translation.Another option is to expand the non-finite clause into a relative clause (if the structural context allows as is done with the ed-clause in (39) in a way that introduces a postural verb into the translation.
(39) (…) music so loud it made the transistor radio balanced on the stoop steps buzz and tremble.(NG1) (…) transistorradion som stod balanserad på förstubron (NG1T) [that stood balanced] The structural shifts characteristic of the translations of non-finite clauses are a general phenomenon that affects the grammar more than the lexicon, but still the effect on the translation patterns of postural verbs appears to be rather great.

Swedish pseudo-coordination
The most discussed characteristic of Swedish postural verbs is their use in so-called pseudo-coordination (see Darnell 2008 for an overview of earlier research and a detailed study).This structure can be regarded as a case of emergent grammaticalization and has often been compared to the general tendency across languages to develop durative aspectual meanings such as the progressive and the continuative (Newman 2002).Recently, Blensenius (2015) has challenged the view that pseudocoordination with postural verbs expresses progressive meaning in Swedish.They are "instead locative and, in some cases, episodic" (Blensenius 2015, Abstract).
In pseudo-coordinations, the postural verb is de-accented and coordinated with a lexical verb, as in (40).There are at least four recurring types of English correspondents of pseudo-coordination.The correspondences go in both directions, which will be demonstrated by presenting first an example from a Swedish original text and then from an English original.The most frequent and also the closest correspondent, according to my intuition, which is not as strong for English as it is for Swedish, is the construction V Postural + V-ing, which appears in (40).The construction is referred to as "simultaneous conjunction" and is regarded as "a potential root of grammaticalization" in Newman & Rice (2004, 371).
(40) När jag låg och smälte frukosten uppenbarade sig ronden.(PCJ1) [lay and melted the breakfast] While I lay digesting my breakfast, the round appeared.(PCJ1T) The same type of correspondence can be observed when English is the source language (see 41).
(41) He would drop the spoon in the sink and stand sipping from his mug while the cat wove between his feet.(AT1) Sedan skulle han släppa ner skeden i diskhon och stå och läppja på kaffet medan katten slingrade sig ut och in mellan fötterna på honom.(AT1T) However, there are several other relatively frequent correspondences.One option is to use a single lexical verb in the progressive, as in ( 42) and (43).This happens 21 times when stå is translated and 17 times when sitta is translated (from Swedish originals) but only twice when ligga is pseudo-coordinated.Further options are to use a co-ordination also in English, as in ( 44) and ( 45).There are also examples where a single verb in non-progressive form appears in the English version as in ( 46), but this option has a rather low frequency.
(46) As he was borne upwards in a series of disconcerting jerks Dalgliesh reflected that success, (…) (PDJ1) Medan hissen ryckvis förde Dalgliesh uppåt stod han och tänkte på att framgången, (…) (PDJ1T) Pseudo-coordination has been commented on rather extensively because of its theoretical interest, but it will not be studied in detail in this paper.
In order to be meaningful, a study would require a detailed discussion of the problem to identify all pseudo-coordinations in an actual corpus, which appears to have been underestimated. 8In the quantitative tables, postural verbs appearing in pseudo-coordinations are simply counted like other postural verbs, since they still refer to posture in this construction.

The problem with genre, in particular Non-Fiction
The comparisons between originals and translations in Section 4 are based on the whole corpus.As mentioned in Section 2.3, postural verbs have an overall frequency that is much higher in Fiction than in Non-Fiction.This section will show what happens when Fiction and Non-Fiction are analyzed separately.Only raw frequencies will be presented, since the frequencies in many cases would be extremely low if presented per 100 000 words.(As in Section 4, the statistical testing is based on the raw frequencies.)

distribution of meanings across genres
The differences between genres are not restricted to the overall frequency, but also involve the distribution of meanings or uses.As can be observed in Table 7, Non-Fiction contains a greater proportion than Fiction of human subjects in sentences with a figurative meaning such as I can't stand him and an even greater proportion of abstract subjects (except for sitta, which is primarily used literally).Direct translations dominate when the subject is human and the overall interpretation is literal, reaching a proportion as high as 71% if the result is summed up for all three postural verbs.For figurative uses of Humans subjects and for Abstract subjects, the proportion of direct translations is very low.Table 9 shows the result for LIE.In this case, Non-Fiction also presents some problems for the generalizations arrived at in Section 4. In Swedish, human subjects are significantly over-represented in Non-Fiction (see Table 9d).However, it must be noted that the number of examples is small.The proportion of figurative uses is also high (8 out of 18 in originals and 11 out of 30 in translations).Inanimate subjects also behave in an exceptional way and are significantly under-represented in English Non-Fiction and close to significantly over-represented in Swedish Non-Fiction.A closer look at the data reveals that the number of physical objects is small and the major difference is related to the category of subjects coded as Place (see Table 9e).SIT as shown in Table 10 is characterized by a dominance of human subjects, which are used literally (as shown in Table 7).Apart from that, both Fiction and Non-Fiction follow the pattern for the whole corpus and do not present any problems for the generalizations presented in Section 4.    To sum up, the comparison across genres shows that the generalizations arrived at in Section 4 very clearly hold up for Fiction.For Non-Fiction, the picture is unclear.One reason for that could be the low proportion of literal uses.Another reason is that Non-Fiction is negatively defined.It actually contains a number of different types of texts such as biographies, academic prose, written versions of public speeches, legal texts and annual reports.For lexical studies, this variety is valuable for giving examples of a wider range of uses than what would have been possible with a more narrowly defined selection (given the size of the corpus).On the other hand, this variety makes it more difficult to achieve comparability across languages and makes quantitative comparisons as the ones presented here less reliable.

Conclusions and discussion
The major result of the present study is that there is a rather strong tendency in translations towards convergence in the usage patterns (word frequencies) between languages on points where functions/meanings overlap.Unmarked categories (Human subjects) tend to be more affected than marked categories (Inanimate subjects).In spite of this, the system of semantic contrasts tends to be preserved.
As a test of the methodology used in corpus-based contrastive studies, the study shows that special care must be taken when monodirectional translation corpora are used and the characterization of some language(s) is based on translated texts alone (cf.Johansson 2007: 31-33).Word frequencies may be strongly affected in particular when the functions overlap across languages.On the other hand, the system of semantic contrasts is preserved, judging from the present study, which probably represents a particularly problematic case.The preservation of the semantic contrasts and the avoidance of translations that can directly be spotted as semantic calques presuppose that translations of high quality are used.Problems do not arise when bi-directional corpora such as the ESPC are used and translation patterns are used only as a complement to the comparison of original texts in both languages.From this, it appears that the inclusion of original texts from all the languages compared is ideal.However, translations make it possible to compare how a specific meaning is expressed in exactly the same context, so the exclusive use of original texts may also be problematic due to lack of comparability.When many languages are compared, as is the case in typologically oriented studies based on corpora, the desirability of having original texts in all languages must be balanced against the desirability of having access to translations, since resources in terms of time and effort are limited.
As a contribution to general linguistic theory, the result of the study of translation effects should be related to theories of bilingualism and language contact.Of special interest from the point of view of contact linguistics and language change is the comparison by Kranich, Becher & Höder (2011) of the contact between Latin and Old Swedish and the contact between English and German today and their discussion of the importance of factors such as the degree of standardization overall of the target language and the establishment of the genre in the target language.Old Swedish lacked a written standard.The influence of Latin could be traced at all linguistic levels, and Latin served as a model for many new genres to the extent that formal equivalence became a preferred strategy in translation at the end of the period (for example, in religious literature).In German, where two relatively new genres were studied (popular science and business communication), the influence was restricted to pragmatic and stylistic features.
However, the convergence seen in translations (especially) of fiction in the present study is related neither to the lack of standardization of the languages involved nor to the lack of an establishment of the genre.In most cases, it is probably completely unconscious as far as the translators are concerned, and it does not fill any expressive gaps; nor is it likely to affect the structure of original texts for the time being.On the other hand, it is a latent feature probably present in most situations where translations are produced, and when the external conditions are right, this type of translation effect can lead to contact-induced language change.(This type of change is discussed in Thomason 2001, Chapter 4).
It is also interesting to compare translation effects with data from bilingual speakers and L2 learners.At a general level (with important differences at a more detailed level), the convergence in translations by professional translators noted in Section 4 has a parallel in secondlanguage acquisition and in bilingual development.In a recent study of Turkish-German contact varieties, Goschler (2013, 127) concludes: "Most of the time, bilinguals neither simply transfer constructions from their L1 to their L2, nor do they behave exactly like monolinguals.Instead, they use L2 patterns that occur in their L1 as well".(Cf.bilingual "compromise systems", Obler 1982.)Professional translators are bilingual speakers with a very advanced level of competence in both their languages.Ideally they translate into their strongest language.When comparing with L2 learners, it is important to take the level of proficiency into consideration, since acquisition often involves several different steps (see Viberg 2002 for a summary of the development of the causative counterparts of the postural verbs in Swedish used as placement verbs, sätta/ställa/lägga 'causesit/stand/lie').In a longitudinal study of Swedish as L2, Viberg (1999) found that postural verbs belonged to the type of frequent verbs that had a tendency to be under-represented in the speech of the L2 learners in comparison to native controls.This was evident even in recordings carried out two years after the first recordings.(After four years, sitta was still under-represented, whereas the situation was unclear for stå and ligga.)More recently, Lemmens & Perez (2012) have presented a study of French learners of Dutch as L2 based on oral picture descriptions.Learners were divided into three proficiency groups and compared to a group of native speakers.In general, the learners had a clear tendency to under-use Dutch postural verbs.Given that both Swedish and Dutch use postural verbs much more than French, this is a clear parallel to the under-use of Swedish postural verbs that Kortteinen (2008) found in translations from French (see Section 1.2).However, the French L2 learners also had a tendency to confuse the different posture verbs.Such confusion lacks a clear parallel in the translation data.There was also a tendency among learners at a more advanced stage to over-use some of the Dutch verbs.In general, there was a U-shaped development, since confusions increased in frequency as the learners started to use posture verbs more often.
Thus, what happens in translation sheds interesting light on what happens in language contact and in bilingual development (cf. Hyltenstam & Viberg 1993).If professional translators are regarded as highly proficient bilinguals, translation effects represent cross-linguistic influence at the upper end of bilingual development.In a similar vein, language contact can be regarded as the study of bilingual language development at the societal (in distinction to the individual) level.From this perspective, the translation effects found in this study represent a kind of language variation that is limited to translated texts, but that ̶ if the social circumstances are the right ones ̶ can rise to the surface and be established as part of the norm of the language.
(8) (…) and one of the great attractions of this house was that it sat within the boundaries of a National Park.(PM1) (…) och något av det mest lockande med det här huset var att det låg i en nationalpark.(PM1T) (9) Their wedding picture used to sit on the piano in their living room.(JSM1) Deras bröllopsfoto brukade stå på pianot i deras vardagsrum.(JSM1T)

( 12 )
Unfortunately this theory does not stand up too well either.(DM1) Tyvärr fungerar inte heller detta särskilt bra.(DM1T) [Unfortunately, this does not function very well either]

( 24 )
Kevin stood up again and scouted for a watchman.(RDO1) Kevin ställde sig upp igen och spanade efter vakten.(RDO1T) (25) Archbold stood up when Wexford came in.(RR1) Archbold reste sig när Wexford kom in, (RR1T) stood open.(RR1)Ytterdörren stod öppen.(RR1T) Say is the most frequent translation of stå when the subject refers to Writing (see examples 20 and 21 above).Other relatively frequent alternatives specific to this category are read and write (usually in passive form: be written).

( 37 )
Lying just south of the Thames in west London, Richmond Park is the most "natural" and largest of the London Royal Parks (SUG1) Strax söder om Themsen i sydvästra London, (…) (SUG1T) [Just south of the Thames…] (38) Jim Rawlings spent the hour between nine and ten that night sitting in another, smaller rented car outside Fontenoy House.(FF1) Jim Rawlings tillbringade timmen mellan nio och tio den kvällen [Zero] i en annan, mindre hyrbil utanför Fontenoy House.(FF1T) [spent the hour ---in another ---car] 7

Table 1 .
The composition and size of the English Swedish Parallel Corpus (ESPC)

Table 2 .
Frequency of SIT, STAND and LIE in Swedish vs. English original texts in ESPC Swedish and English Postural verbs.The frequency in original texts of Swedish sitta, stå and ligga compared with English sit, stand and lie.Frequency per 100 000 words.The Frequency of the English postural verbs: sit, stand and lie in originals and translations.Frequency per 100,000 wordsIn translated texts (Diagram 2a), English postural verbs are significantly over-represented (if original texts are regarded as the norm), whereas the Swedish equivalents are under-represented (Diagram 2b).
Över vintern låg jag i Dragör, som hade en av de få hamnarna i Öresund där det var liv och rörelse året om.(BL1) [Over the winter, I lay at Dragör] At present, she was laid up for the winter at Dragor, that being one of the few ports on the Sound which remained astir whatever the season.(BL1T) Such examples are coded as inanimate subject, since that is what decides the choice of verb.(In canonical position, cars stand and boats lie in Swedish.)There are not very many examples of this type, but they are theoretically interesting.

Table 3 .
The basic contrasts between Swedish Posture verbs indicating Inanimate Location.
Writing has a special semantic status combining concrete and abstract readings (see Viberg 2013: 157-159 for discussion), but for the purposes of this paper, it is enough to say that examples of this type are treated separately and not included in the abstract (or inanimate) category.
Major categories of English stand in orignals and translations.Frequency per 100 000 words.By comparing Diagram 4a and 4b, we can also see that Human subjects and in particular Inanimate subjects are considerably more frequent in Categories of English lie in originals and translations.Frequency per 100 000 words.
above).Although well established, the use of sitta with inanimate even in comparison to the other postural verbs in Swedish, and is considerably less frequent than the use of stå and ligga with inanimate subjects.Per 100 000 words, its frequency of occurrence is 9, which should be compared to 34 for stå and 56 for ligga.Abstract subjects are infrequent with sitta in Swedish.Diagram 6b shows the distribution of the categories of English sit.The most striking characteristic is that human subjects dominate completely over the other types.The other categories have such low frequencies that it is not meaningful to compare originals and translations.Human subjects are significantly over-represented in translations (p<0.0001).

Table 4 .
The translation pattern of English stand 1: Direct lexical translations and Lexical shifts:

Table 5 .
The dynamic system of Swedish Posture verbs (with Human subject).

Table 6 .
The translation pattern of Swedish stå: Direct lexical translations and Lexical shifts Actually, it is possible to find parallel examples where a postural verb without a correspondent in the Swedish version appears in a relative clause in English, as in (33), but there are very few of them.For lie, there is this example in the translated texts but there is none in the original English texts.

Table 7 .
The distribution of meanings across genres in Swedish original texts

Table 8d
Swedish stå in originals and translations, non-fiction

Table 9a English
lie in originals and translations, fiction

Table 9b
Swedish ligga in originals and translations, fiction

Table 9c
English lie in originals and translations, non-fiction

Table 9d
Swedish ligga in originals and translations, non-fiction

Table 9e
The major subcategories of inanimate for English lie and Swedish ligga, nonfiction.

Table 10a English
sit in originals and translations, fiction

Table 10b
Swedish sitta in originals and translations, fiction

Table 10c
English sit in originals and translations, non-fiction

Table 10d
Swedish sitta in originals and translations, non-fiction