Comparing the functional range of English to be to German sein: a test of the boundary permeability hypothesis

1 Introduction

It is a truism that so-called translation equivalents may not have the same semantic range or extension in different languages. Take as an everyday example the English word girl, which is typically rendered by Mädchen in German. While Mädchen refers to females up to and including the age of puberty, girl can in principle be extended to address a woman in her eighties even though this use is of course less common than reference to a child. The case of girl versus Mädchen is certainly not exceptional. In fact, it does not seem exaggerated to claim that the majority of words differ in their extension from language to language. It may also be suspected that the difference in range varies on an item-by-item basis. That is, for some translation equivalents, the word from language A has a wider functional range than the corresponding word from language B whereas for other translation equivalents, it is the other way around. A case in point is the two pairs do – tun and make – machen in English and German. Nehls (1991) argues that do has a larger extension than its German congener tun while the opposite is true of make versus machen. Similar observations can be made on the basis of traditional word field analyses (e.g. time concepts (Jäkel 2003) and motion verbs (Snell-Hornby 1983)) as well as more sophisticated and more fine-grained methods of semantic cartography (e.g. Enfield et al. 2006 on body part terms and Majid et al. 2007 on cutting/breaking verbs).

All these (and a wealth of other) studies adopted an inductive approach. They almost certainly started from a casual observation which was felt to be promising enough to merit closer attention. While there is, of course, nothing dubious about this approach, it has certain limitations. It is unclear why a particular topic rather than another was subjected to scrutiny. As a consequence, the extent to which the results are generalizable is unknown. Moreover, the possibility of links between different word fields, not to mention wider organizational principles of the lexicon, remains unexplored.

The present contribution takes quite a different approach. It is deductive in that it starts from a general claim and examines one of the predictions that can be derived from it. Berg (2014) introduced the notion of boundary permeability as a typological parameter on which languages may vary systematically. The basic insight is that linguistic categories have more or less permeable boundaries. Strict boundaries make a language rigid and prevent the extension of a given unit to “new territory”. By contrast, soft boundaries make a language flexible and facilitate the wider use of a given unit. On the basis of an investigation of a considerable number of phenomena in English and German, Berg concluded that English is a soft-boundary language while German is a strict-boundary language. To illustrate, a good number of grammatical categories including causativity and resultativity involve semantic distinctions which are often formally expressed in German, but blurred in English. For example, the boundary between causative and non-causative is weaker in English than in German. The English verbs to drop and to melt may serve as examples. The causative use of to drop is illustrated in to drop a bomb, the noncausative use in the death toll dropped. Similarly, you may melt iron when the snow melts. In German, different verbs are typically required for the causative and the noncausative reading. Two of the various formal means are the marking of causativity by an additional verb (e.g. fallen lassen vs. fallen ‘to drop’) and a tonic vowel change (e.g. Eisen schmelzen ‘to melt iron’ vs. der Schnee schmilzt ‘the snow is melting’). While some overlap between English and German coding strategies cannot be denied, English verbs may be argued to be more functionally flexible than their German counterparts.

Two properties of this approach deserve highlighting. Following the noncategorical nature of boundary permeability, the difference between languages in general and between English and German in particular is a gradient one. Furthermore, while the notion of boundary permeability is generally applicable to entire linguistic systems, it is not claimed that the nature of the boundaries separating all categories of a language must be alike. Variation in boundary permeability within a language is to be expected. However, what is claimed is that in a representative dataset, a language like English will have on average more permeable boundaries for example, than a language like German.

While Berg’s conclusions mainly relied on an examination of grammatical categories, he also presented a brief look at phenomena at the interface of grammar and lexicon as well as some speculative remarks on the nature of the lexicon proper. While, as explained above, the hypothesis of a general difference between English and German does not claim to capture all conceivable aspects of a linguistic system, it invites thorough testing of areas not dealt with in the original article. In this paper, one such test will be performed. The single item which was selected for scrutiny is the verb to be and its German counterpart sein. These two verbs play a central role in their respective languages. What make them particularly germane to the notion of boundary permeability is their multifunctionality and their place at the crossroads of grammatical function and lexical meaning. They may serve an auxiliary function, as in the formation of composite tenses, and a copular function, as in predicative-adjective constructions, as well as cover some lexical ground related to the meaning of existentiality. It is irrelevant for present purposes where the line between grammar and lexicon is drawn, if it is at all necessary or appropriate to draw such a line.

The working hypothesis that will be put to the test follows directly from the above discussion: to be is predicted to be more flexibly used than sein. By virtue of the higher boundary permeability of English, to be is expected to exhibit a relatively wide range of uses whereas the lower boundary permeability of German restricts the use of sein. Note that the predicted difference is a relative rather than an absolute one: E. to be is predicted to be more versatile relative to German sein.

The remainder of this work is structured as follows. The next section introduces the methodology which underlies the empirical analysis. It is followed by a bidirectional translation study which examines the strategies employed in rendering to be and sein in the other language. Subsequently, the boundary permeability hypothesis is evaluated in the light of the results of this analysis. Suggestions for further tests of this hypothesis, in particular the comparative weight of English and German verbs, round off the paper.

2 Methodological issues

The most direct way of establishing equivalences in two (or more) languages is by means of a translation study. The major strength of this method derives from the obligation on the part of translators to render the content as faithfully as possible. This constraint guarantees, or is supposed to guarantee, semantic equivalence. Translators are granted more freedom in preserving or ignoring formal aspects of the original text. While they are constrained by the well-formedness conditions of the target language, they may often choose among alternative lexical items and grammatical structures. It is well-known from psycholinguistic research that form priming occurs in translating (or shifting) from one language to another (e.g. Davis et al. 2010; Loebell and Bock 2003; Maier et al. 2017; Schoonbaert et al. 2009). This priming effect biases translators to use the same (or similar) forms as were used in the original. If they prefer to depart from this model, they may be expected to have good reasons for doing so. This bias towards convergence makes cases of divergence especially remarkable when there is a choice between convergent and divergent forms.

The following criteria guided the construction of the database. It goes without saying that text selection was constrained by the availability of translations. Priority was given to the spoken language. This decision eliminated a mountain of literary translations. The dataset was required to resemble authentic language as closely as possible. While the search for large samples of spontaneous, unmonitored speech plus their translations turned out to be unsuccessful, the second best choice fell on television serials. Even though the dialogues in these programmes are at least partially scripted, they may be argued to emulate the living language reasonably well.

The two mystery serials Buffy and Dark were selected for scrutiny. Buffy was originally produced for an English-speaking audience in the U.S.A., first broadcast in 1997, subsequently translated into German and shown on a private German TV channel. It is about a 16-year-old girl who leads a double life of a high-school student at daytime and of a hunter of vampires at night. Only the first of seven seasons in total was subjected to analysis. Of the 12 episodes that make up this season, the ninth was ignored because its translation was unavailable. Each episode lasts 40 min. All of the data are from the year 1998.

Dark was originally produced for a German-speaking audience by the streaming provider Netflix and is one of the very few German serials which have been translated into English. The storyline is situated in a small town in which children disappear under mysterious circumstances every 33 years. The first two of three seasons, which contain 18 episodes in total, entered the database. Each episode lasts approximately 50 min. Dark is a fairly recent production, which started in 2017. The data are from the years 2017 and 2018.

The ensuing analysis is based on the transcripts of the dialogues available in Internet forums.^[1] Spot checks were carried out on the transcripts, which turned out to be faithful to the oral dialogues. No detailed information is at hand of the people (and their formal qualifications) who are responsible for the translations.

Table 1 provides a word count of the database. Although the two samples differ in size, it will be seen in the next section that the number of hits is virtually identical.

It may be noted as an aside that the translation into German involves a decrease in the number of words whereas the translation into English involves an increase in word number.^[2] It would seem that English requires more words to express the same content than German, at least as far as these TV dialogues are concerned. Whether this results from constraints on dubbing or reflects a genuine difference between English and German remains to be elucidated.

The bidirectional design is a crucial methodological aspect of the present study. It allows one to directly compare the translations of to be into German to those of sein into English. Comparability is further increased by the relatively large similarity of the two serials in terms of genre and speech style.

The ensuing analysis excludes all uses of to be/sein which are dictated by purely grammatical constraints and hence leave no room for variation. Unlike English, German forms the complex tenses of some verbs on the basis of sein. Obviously, this disparity has nothing to do with the semantic ranges of to be and sein. However, copular uses of these verbs were taken into account because they permit variation in that they may either be rendered by the counterpart or a different verb in the target language. German stative passives were also included because it was not clear at the outset how they would be translated into English. There were no cases of modal to be to in the data.

Each sentence or clause containing a form of sein/to be along with its translation was transferred to an Excel file. All tokens of sein and to be in the original text entered the analysis irrespective of how they were translated. This comprehensive analysis allowed us to determine the rate of translations in which sein and to be were rendered by other verbs in the two languages. For want of a better term, these other verbs will henceforth be called “lexical verbs”. While this label refers to all verbs other than sein and to be, it should be understood that sein and to be may also have more or less lexical uses (see below). The principal coding decision was whether sein and to be were translated by their counterpart in the other language (“category-preserving”) or by lexical verbs (“category-changing”). All relevant verbs in both the original and the translation were separately listed in the Excel file.

3 Data analysis

The high textual frequency of to be and sein ensured a considerable number of hits. Table 2 presents a binary classification of the data into translations in which to be/sein is preserved and those in which to be/sein is changed into a lexical verb. It should be noted that the infinitives to be and sein are understood to be umbrella categories which include all inflected forms of these verbs.

It is plain to see that the strategies employed in translating from German to English differ substantially from those employed in translating from English to German. Whereas in the vast majority of cases, translations into English select to be as the counterpart of sein, translations into German make use of a lexical verb in more than 1 out of 4 cases. Category-changing cases are four times more frequent in translations from English to German than those from German to English. This difference is highly statistically significant (X²(1) = 228.6, p < 0.0001; Cramer’s V = 0.28).

The preceding analysis suggests a major disparity between to be and sein but remains silent on the nature of this disparity. In particular, it says nothing about the semantic range of these verbs. The next step involves a count of the verbs by which to be and sein were translated. It will be assumed that the higher the type frequency of the lexical verbs, the wider the semantic range of to be and sein. This is only a first approximation because a quantitative analysis of verb types does not reveal anything about the number of semantic fields to which the verbs belong.

Verb types were defined in terms of lemmas. For example, verbs such as to go and to go on were treated as separate types. The results were as follows. Whereas 62 different verbs were used in translating sein into English, as many as 131 different verbs were considered appropriate for translating to be into German. This difference is highly significant (binomial, p < 0.0001). Thus, the number of German verb types corresponding to to be is significantly larger than that of English verb types corresponding to sein. On the assumption that different verbs code different meanings, it may be provisionally concluded that the semantic range of to be is wider than that of sein.

And yet, the number of verb types is only an indirect measure of semantic range because it ignores the extent of the semantic range. While it would seem unlikely that a rather large number of verbs fall into a low number of semantic categories, the only convincing way of ruling out this possibility is to compare the number of semantic categories to which the lexical verbs can be assigned.

Semantic categorization is a notoriously difficult task. The difficulty lies not so much in setting up categories but in deciding on the most insightful level of granularity. If categories are too narrowly defined, the degree of semantic generalization remains low; if, in contrast, the categories are too broad, possible differences between the languages are levelled out. Comparability was ensured by targeting the same (intermediate) level of granularity. This decision led to the establishment of classical categories such as verbs of motion and change of possession. However, the meaning of quite a few verbs was so particularistic that grouping them with other verbs seemed unreasonable. We therefore ended up with a set of semantic categories as well as a set of miscellaneous verbs. In an attempt to limit the effect of hapaxes, only verbs with a token frequency of at least two were included in the ensuing analysis.

All in all, 16 categories were set up for the English–German translations but only 8 for the German–English translations. The number of unclassified verb types amounted to 19 in the English–German translations but only 7 in the German–English translations. It is certainly possible to shift the boundary between semantic categories and unclassified verbs. For instance, the requirement that each category should have more than one verb may be relaxed. Additionally, token frequency may be accorded greater weight. Whatever constellation and weighing of criteria was chosen, the number of both semantic categories and isolated verbs always remained about twice as high in the translations from English to German as those from German to English.

The critical comparison concerns the relationship between the categories created for English and German verbs in the target language. As it turns out, all 8 categories in the German–English translation were also found in the English–German translation (see Table 3). That is, the English categories form a proper subset of the German categories. This is the strongest piece of evidence so far adduced in favour of a greater semantic range of to be than that of sein.

The preceding quantitative analysis will be followed by a qualitative analysis. The more detailed examination is in three parts, all of which are intended to map out the extension of to be and sein. The first two focus on individual verbs and are distinguished by the relationship between the lexical verbs in German and English. The first part (Section 3.1) investigates semantic domains which occur as translations of to be but are missing as translations of sein. The second part (Section 3.2) considers areas of semantic overlap where the translations of to be and sein create similar or identical categories in the target language. The focus on the word level is a necessary starting point in order to get a handle on possible differences between the two languages. However, it is obvious that these differences cannot all be captured at the level of individual words. Verbs are embedded in larger structures, which are here labelled “constructions” without any commitment to Construction Grammar. Since the use of some verbs cannot be adequately described without reference to the constructions they are a part of, it is necessary to address the constructional level separately. This will be done in Section 3.3.

4 Theoretical discussion

The heuristic principle that has been employed in this study in order to get at the meaning of verbs is translation: meaning is inferred from how a verb is translated into another language, which itself is the result of an inferential process on the translators’ part. On the basis of the difference in boundary permeability, category change was predicted to be significantly more frequent in the English–German than the German–English translation. This prediction was impressively borne out by the empirical data. While the use of sein is largely limited to copular and existential functions, that of to be goes well beyond these functions and reaches into areas which are more semantically loaded. Thus, to be displays a wider lexical range than sein does.

The notion of variable boundary permeability provides a natural explanation for this cross-linguistic difference. Because category boundaries are more permeable in English, to be can (relatively) easily bridge the gap between grammar and lexicon and cover different areas which would traditionally be separated into grammatical and lexical domains. Inversely, because category boundaries are stricter in German, the use of sein is relatively constrained and more anchored in the grammatical domain.

Needless to say, the difference between to be and sein is one in degree, not in kind. After all, nearly one third of the category changes occurred in translations into English. This shows that sein can also be variably used. However, its semantic range is rather limited. The data analysis revealed a difference between the unidirectional cases discussed in Section 3.1 and the bidirectional cases discussed in Section 3.2. Generally speaking, the unidirectional cases involve a greater semantic distance while the bidirectional cases involve a smaller semantic distance between to be and its German translation. To be more specific, German verbs are generally less lexical in the bidirectional than the unidirectional cases. Semantic distance thus functions as a predictor for the translation patterns. If the distance is small, to be and sein may both be translated by a lexical verb; if, however, the distance is large, only to be may be translated by a lexical verb. This implies that to be has a wider semantic range than sein.

The high versatility of to be can be seen not only in the semantic but also in the constructional domain. The data analysis showed that to be partakes in a number of constructions of which at least one is of extremely high frequency and all are more frequent in English than in German. Nothing even remotely comparable can be said about sein. While presentational constructions also exist in German, they are less common and do not usually draw on sein. Why is to be the verbal basis of these highly abstract constructions rather than another verb? The boundary permeability hypothesis provides a natural answer to this question. Given that English is characterized by soft boundaries, to be may not only span the grammatical and the lexical domain but also serve in units of various sizes. More particularly, it may act on its own or be part of a larger structure which has a highly abstract, schematic meaning.

This perspective allows one to paint a unified picture of to be. Its range may extend in two opposite directions. As documented by the German translations, it readily assumes lexical functions. At the same time, it encroaches upon the constructional domain by forming the verbal basis of the above constructions. Significantly, the boundary permeability hypothesis views both these properties of to be as related. Because permeability describes the nature of boundaries as such, boundaries are permeable towards both the lexical and the constructional end. The lexical translations of to be in German and the use of to be in English constructions may therefore be regarded as two sides of the same coin.

It is not suggested here that the considerable semantic range of to be and the high frequency of the there is construction are causally linked in the sense that they must co-occur. The frequency of a construction is certainly determined by a variety of factors of which the verb is just one. However, there is no doubt that the high versatility of to be is compatible with its preferred use in constructions, which themselves may have a high token frequency.^[5]

The above analysis of the difference between English and German might give rise to a serious misunderstanding. A number of examples discussed in the previous section may have given the impression that the English original is somewhat vague or even loosely worded while the German translation is more precise or “crisp”. Such a value judgement is alien to the notion of boundary permeability. It is more in the spirit of this hypothesis to argue that to be is more versatile and hence can be more flexibly used than sein.

The claim that to be has a relatively wide semantic range raises the issue of how its specific meaning in a concrete communicative act is ascertained. Basically, there are two options. Either all the different meanings are memorized or listeners have to rely on the context in which to be is embedded during the comprehension process. While memorization undoubtedly plays a major part, innumerable psycholinguistic studies have shown that listeners take full account of the micro- and macro-context to construct a semantic representation. On this basis, it may be predicted that the meaning of to be is distributed more widely across the neighbouring words than the meaning of sein. Notably, Hawkins (1986, 2019 has presented a generalized version of this proposal. He contends that English verbs are vaguer or more polysemous than their German counterparts and that this vagueness or polysemy is overcome by “word-external properties”, i.e. by information from the surrounding context. In a word, English is more distributive than German. It is obvious that the outcome of the present study is in perfect harmony with Hawkins’s claim.

If more context is required in English than in German, it may be inferred that more words are needed, i.e. that English sentences are longer than German sentences conveying the same content. At this point, a subsidiary result of the above analysis becomes crucially relevant. It was noted in connection with Table 1 that the English translations were longer than the German originals and that the German translations were shorter than the English originals. This unexpected result now begins to make sense in the light of the more distributive nature of English.

5 Is English a speaker-friendly language?

Let us assume that the difference in semantic range between to be and sein generalizes to an unspecified extent across the English and German lexicons. A wider semantic range is beneficial to speakers in that it facilitates lexical access. If words cover more semantic ground, speakers can be less “pedantic” about finding the correct address. Thus, lexical retrieval is less cognitively costly in production. The opposite holds for comprehension. If individual words do not provide enough semantic cues, listeners have to pay attention to more than one word simultaneously and integrate these words into a larger structure. This is cognitively costly. As a matter of fact, there is empirical evidence for this hypothesis. As Wittenberg and Piñango (2011) showed for German and Piñango et al. (2006) showed for English, light verbs incur extra processing cost arguably arising at the stage of semantic integration.

On the basis of the difference in semantic range, English may be characterized as a speaker-friendly and German as a listener-friendly language. Of course, a single property does not justify such a general conclusion. However, additional evidence can be adduced in favor of characterizing English as a speaker-friendly language. We will limit ourselves to three salient features of English, each from a different subsystem. The first is the morphological process of conversion. The ease with which words may change their class without any formal change makes English stand out among the languages of the world. Clearly, using one and the same lexeme in different syntactic functions is to the speaker’s benefit because affixation and other morphological processes can be dispensed with.

The second relevant feature is a phonological one. The unstressed vowel schwa occurs in both English and German. While schwa is part of the English phoneme system, its phonological status is unclear in German (see Wiese (1996) for some discussion). The fact that clear cases of minimal pairs are lacking is one of the reasons why both Moulton (1947) and Wurzel (1991) assign it allophonic status. It may be worth adding that schwa is the most frequent vowel of English (Denes 1963). There is no doubt that schwa is a speaker-friendly vowel, which involves the least displacement of the articulators from their rest position.

The third salient feature is of a syntactic nature. As Rohdenburg (1974) showed, English subjects can accommodate a large number of thematic roles. Whereas German subjects are largely restricted to the agent, English subjects run the full gamut from time and location to source and instrument. This flexibility facilitates sentence planning because a single syntactic structure can be used for expressing a variety of semantic relationships. At the same time, this one-to-many mapping puts a strain on listeners, who cannot rely on the syntax in their attempt at deriving the meaning of an utterance.

This set of features could be enlarged considerably. However, it may suffice to lend credence to two major claims. One is the viability of the distinction between speaker- and listener-friendly languages. What is to the advantage of the speaker may be to the disadvantage of the listener. The other claim is more specifically about a general characteristic of English. There is preliminary evidence to suggest that English is biased towards satisfying the needs of the speaker to a greater extent than those of the listener.

6 Outlook and conclusion

The present study has exclusively focused on a comparison of the range of to be and sein. However, this is only one topic among numerous others. The generality of the notion of boundary permeability invites a large number of further comparisons, of which only a few can be mentioned here. The second most grammatical verb pair is to have – haben. Does it evince the same difference as to be and sein? There is preliminary evidence that it does. English has numerous V + NP structures such as to have a bath and to have a drink, whose translations do not use the corresponding verb in German. It is possible, then, that like to be and sein, to have has a wider range than haben. Only a careful analysis can tell.

In-between the grammatical and lexical verbs we find light verbs such as to go – gehen and to come – kommen as well as to take – nehmen and to give - geben. The boundary permeability hypothesis leads us to expect “lightness” to play a larger role in English than in German. There are three ways in which this prediction can be made more precise. First, there may be lighter verb uses in English than in German. Second, light verb uses may have a higher token frequency in English than in German. Third, light verbs may on average be lighter in English than in German. A serious test of the third option faces the considerable challenge of quantifying weight. However, such a comparison would certainly repay a careful examination.

The preceding analysis raises a more general theoretical issue. Can a language (or rather, its users) afford to have a lexicon in which all (or the great majority of) entries have a wide functional range? In other words, can a wide range be a lexicon-wide property of a given language? Or is it necessary to balance out the wider functional range of some words by a narrower functional range of others? In fact, there is no evidence for such an equalizing mechanism. Besides, it would be completely mysterious as to how such a mechanism could arise in the first place. It would have to be accorded tremendous computational power for no obvious reason.

However, the unlikelihood of a compensatory mechanism within the lexicon does not rule out the possibility that other parts of the linguistic system may directly or indirectly respond to semantic range. Hawkins (2019) argues for a link between the distributive nature of the lexicon and the fixed word order of English. If listeners have to rely on units larger than single words, their task of meaning construction is facilitated by predictable word order. This would imply that the rigidification of word order preceded the increased content dependence of lexical items in the history of the English language.