Verbs of letting in Germanic and Romance languages A quantitative investigation based on a parallel corpus of film subtitles

This study compares eleven verbs of letting in six Germanic and five Romance languages. The aim of this paper is to pinpoint the differences and similarities in the semasiological variation of these verbs, both across and within the two language groups they represent. The results of a Multidimensional Scaling analysis based on a parallel corpus of film subtitles show that the verbs differ along several semantic dimensions, such as letting versus leaving, factitive vs. permissive causation, modality and discourse function. Although the main differences between the verbs lend themselves to a genealogical interpretation (Germanic vs. Romance), a distributional analysis of constructional patterns in which the verbs occur reveals that these differences are in fact distributed areally, with a centre and a periphery.


Introduction
The aim of this paper is to compare eleven verbs of letting in six Germanic and five Romance languages. The Germanic verbs are the Danish lade, Dutch laten, English let, German lassen, instance, the VSV construction, where the second verb is an inflected infinitive that agrees with the Agonist (S), seems to exist only in Portuguese.
V + Clause with subjunctive < VSV < VOV < VVO Compare (9a)-(9d), four Portuguese sentences (Soares da Silva, 2004:297)  On the basis of our corpus observations, which will be introduced below, we can conclude that the Germanic verbs of letting exhibit less variation in the degree of syntactic integration, occurring predominantly in the VOV construction in the main clause. 6 6 The degree of semantic integration may be also expressed by the case form of the Agonist. For example, the preposition door "by" marks the agentive Causee in Dutch causative constructions with laten, whereas aan "to" marks the Causee in the role of an experiencer of mental events (Kemmer and Verhagen, 1994). These two prepositions indicate a lower degree of semantic binding than the bare NP (the default). Similar variation in case marking is also possible in Romance languages (e.g. Soares da Silva, 2004).

Dimension 5. Permissive or factitive causation
The difference between permissive and factitive causation (Nedjalkov, 1976: Ch. 3) can be defined as the difference between letting and making X do Y. 7 In many Germanic languages the verbs of letting can express factitive causation, as in (10) For instance, the Swedish auxiliary låta has been used to express both factitive and permissive causation since the earliest available texts (ibid.). Unfortunately, there is insufficient historical evidence to answer this question conclusively.

Dimension 6. Polarity
The distinction between permissive and factitive causation is cross-cut by another dimension, the presence or absence of negation. While letting verbs in positive contexts (e.g. She let him go) express non-impingement, they express coercion in negative contexts (Nedjalkov, 1976:24). For instance, the sentence She didn't let him go means that she made him stay. Even though making and not letting describe the same referential situations, the construals of these situations are different. Unlike positive constructions with factitive causation verbs (e.g. make, have, cause), negative expressions with let presuppose some potential world where non-occurrence or cessation of impingement might or should take place.

Dimension 7. Letting vs. leaving
This dimension is based on a subtle difference in construal that introduces a whole new family of semantic extensions. In constructions of the type X lets Y do Z, the Antagonist X is the primary Trajector, and the Agonist Y is the primary Landmark. However, the Agonist Y is also the Trajector of the action or process do Z, which represents the permitted event (cf. , 1991:409-410).

Langacker
The Agonist may also be construed as a Landmark only, a case of leaving, rather than letting. Typically, in such cases the Agonist's intrinsic tendency is towards rest (cf. Soares da Silva, 2007:176 Instead, French typically uses quitter "to leave, quit, part" and abandonner "to give up, abandon", whereas the Romanian equivalent is a părăsi "to abandon, neglect".  Yet another semantic modification of LEAVE is observed when the Agonist ceases to be an individuated participant and becomes an abstract entity, as in (16) In this sentence, the Antagonist is coreferential with the semantic object of the verb misleiden "mislead" (je "you"). Therefore, he/she is also a potentially affected entity. Such constructions demonstrate that the Antagonist is considered responsible for prevention of something bad that happens or may happen to him or her (cf. Loewenthal, 2003 for Dutch).
Thus, although the Antagonist is in fact an affected entity, it still has some properties of an Agent, most importantly, the control over the effected event.
Letting verbs in reflexive constructions can also be used to convey the semantics close to the middle voice. The Antagonist in such cases has some properties that facilitate the action specified by the effected predicate to be carried out (Davidse and Heyvaert, 2003 The Antagonist has no agentive properties and has no influence on the situation. In such cases, the participant in fact ceases to be a true Antagonist.

Dimension 9. Modality
Letting is closely related to modality. If X lets Y do something, this means that Y can or may do something. Some letting verbs can be used in a purely modal function to express the speaker's desire, encouragement, surprise, etc. One example is the hortative use of let and its Germanic cognates, as in (21): Another use is optative, as in (22), whereby the speaker expresses a wish: (22) Let no one be dead today. "He said that he would drink ten Duvels, and imagine, he did that!"

Dimension 10. Discourse function
In some contexts, the original conceptual meaning of verbs of letting is bleached, and a verb of letting may be used as a part of a discourse marker, most commonly for making metalinguistic comments. Such discourse markers may introduce a minor break in discourse, as in (24), or clarification and elaboration, as in (25). Another function is signalling the speaker's intention to perform some action, as in (26) Section 4 will investigate which of the dimensions of semantic and pragmatic variation described above play prominent roles in describing the cross-linguistic variation in the use of the eleven verbs of letting. Before doing so, one needs to describe the data and methodology, which are presented in Section 3. The format of the files is SubRip (.srt), which includes information about the time when a caption should appear on screen and disappear, as well as optional formatting information.

Data and method
More complex structures are possible, as well, as a segment from the same file demonstrates in Example (28)   It seems that the frequencies of let in the soap operas and subtitles are comparable. One can also see that the subtitles and soap opera scripts have higher proportions of let than TV and radio transcripts from the spoken segment of the COCA. This may be due to the number of instances of let's and letting of cooperative dialogue, stemming from the fact that films and soap operas involve a great deal of action and interaction. The frequency of let in the translated subtitles (27.38 per 10,000) is somewhat lower than in the original subtitles (30.83 per 10,000). One may wonder whether this difference is due to the fact that laisser has relatively low frequencies in the original French subtitles (see Table 1  The second test concerned qualitative differences in the distribution of let in the subtitles and in the corpus of American soap operas. We constructed a profile vector with relative frequencies of the constructional patterns of English let in soap opera scripts from the . Although other languages need to be investigated as well, the working hypothesis of this paper is that subtitles reliably represent the general patterns in the use of verbs of letting.

The data set
The data come from a multilingual parallel corpus for typological comparisons (ParTy) compiled by the author. We selected subtitles from five films of different genres and original languages, representing the eleven languages given in Section 1. The films are listed in  Table 2. Films used in the study. The resulting two-dimensional solution (stress = 0.094) with some individual points ( Figure   6) will be discussed below. Note that the semantic dimensions that we are trying to detect need not correspond to two automatically obtained dimensions of the MDS; rather, they have to correspond to the patterns and clusters that emerge when one compares the semantics of the points on the map. An examination of the points in the middle of the map reveals that this area contains exemplars that express non-occurrence of impingement (i.e. extended letting); one such example is (30), which corresponds to Exemplar 223 in Figure 6. This area also contains exemplars that express removal of blockage (i.e. onset letting); one such example is (31) central, it is difficult to conclusively determine which is more basic.

Original title Title in English
The area at the top of Figure 6 contains different instances of leaving; for example, leaving someone in a specific location, as in Exemplar 64, presented in (32) The lower portion of the map in Figure 6 contains examples for which verbs of letting are accompanied by another verb. Exemplars on the left-hand side include numerous instances of cooperative dialogue and thinking (e.g. Exemplar 87, presented in (34)), as well as hortative uses (e.g. Exemplar 102, presented in (35) To summarize, the main distinctions revealed by the map are as follows: letting vs. leaving, modal/cooperative vs. non-modal/non-cooperative uses, and factitive vs. permissive causation. These are therefore the most important semantic dimensions of the cross-linguistic variation of the verbs.
Let us now see how the verbs are mapped onto the semantic space. Figures 7 and 8 display the distributions of the Germanic and Romance verbs, respectively. Each black point represents one instance of a given verb of letting.

A hierarchical cluster analysis of the verbs
In Section 4.1, we compared the verbs of letting by looking at their semantic functions. This section offers a quantitative comparison based on a hierarchical cluster analysis. First, the Gower distances between all verbs were computed (see Section 4.1). These scores represent the proportion of overlapping values ('Yes' or 'No') between the verb columns in the matrix shown in Figure 3. The greater the proportion of overlapping values between a pair of verbs, the smaller the distance. Next, a hierarchical cluster analysis was performed. The resulting clustering tree is shown in Figure 9. Notably, the Romance and Germanic verbs of letting do not constitute separate clusters, as one might expect on the basis of the genealogical information. Instead, the German lassen, Dutch laten and Swedish låta belong to the large cluster on the right, which also contains the Romance verbs. Most of the lower-level genealogical subgroups (cf. Figure 1a and 1b) are not supported, either. For example, Portuguese clusters together with Romanian and Swedish, whereas the other Ibero-Romance language, Spanish, is in the same cluster with French and Italian. English, a West Germanic language, is in a cluster with Danish and Norwegian, which are both North Germanic languages. These conclusions support the observations made in Section 4.1. Although the tree does display some genealogical patterns (e.g. French, Italian and Spanish cluster together), this is not systematic. This suggests that there could be other factors that explain these cross-linguistic differences, such as geographical proximity and language contact.

Distributional analysis based on constructional patterns
To examine the differences between the verbs in more detail, all examples in every language were coded for a number of constructional patterns. Constructional information serves here as an operationalization of semantic differences. This approach has been applied fruitfully in neo-structuralist approaches to semantics (cf. Geeraerts, 2010). For example, Levin's (1993) verb classes are based on the syntactic alternations in which the verbs can participate.
Alternatively, Levshina and Heylen (2014) use syntactic information, e.g. subcategorization frames and syntactic relationships, as well as information about the lexical neighbours, to identify relevant semantic classes of nouns and verbs for modelling near synonymy in Dutch.
Although the constructions are not functionally identical in the languages under consideration, they are similar enough to enable a cross-linguistic comparison. The constructions are shown in Table 3. As one can see, the classification is quite coarse-grained. This is done for two reasons. First, there needs to be a sufficient number of tokens per type to draw robust conclusions. Second, more general patterns have higher chances of conveying similar semantics across this group of languages than more specific ones.

Discussion
The aim of this study was to find similarities and differences in the semasiological variation of eleven verbs of letting in Germanic and Romance languages. Using a parallel corpus of film subtitles, we employed exploratory statistical techniques to visualize the common semantic space of letting and to establish the distinctive and overlapping functions of these verbs with regard to the most important semantic dimensions. We found the most important 13 Adding a third dimension leads to a total explained inertia of 81.9%. However, this dimension separates the Danish verb from all others, which is not particularly informative.
dimensions of cross-linguistic semantic variation for these verbs to be letting vs. leaving, factitive vs. permissive causation and modality/discourse function.
We also observed some general genealogical patterns that relate to these dimensions of  (2007), the Germanic cognates let, laten, lassen, etc., as well as the Gothic lētan illustrate the evolutionary course from the sense of LEAVE to the semantics of LET/PERMIT. 14 The results of our quantitative study suggest that, in the contemporary Germanic languages, the semantics of leaving is marginal. Instead, prefixal derivatives are used as verbs of leaving, such as the German verlassen "to leave, abandon" or the Dutch nalaten "to leave, bequeath", as well as other verbs that may be etymologically related to the verbs of letting, such as the English leave or Swedish lämna "to leave".
The development of the Romance letting verbs can be traced in greater detail than that of the Germanic ones. The meaning 'to let go, release' of the Latin etymon laxare became prototypical around the second century CE. Later, the meaning 'to go away, abandon' developed (replacing the verb relinquere, which previously expressed this meaning), as well as the meaning 'to let, allow', a development that made laxare a semantic successor of sinere (Soares da Silva, 2007:185). With time, the historical prototype of active letting with previous impingement 'to release, let go' has become less prominent in the Romance languages. As 14 This assertion is not a unanimous one, however. For example, The Etymological Dictionary of Proto-Germanic (Kroonen, 2013: 332) says that the meaning of the proto-Germanic *lētan was 'let, allow'. noted by Soares da Silva (2003), the French laisser and Romanian a lăsa have more salient verbs for expressing the sense of 'let go' (see Section 2). This sense implies an initial physical contact and a spatiotemporal contingency of the letting and permitted events, giving the latter a high degree of semantic integration. At the same time, the meaning 'let', as opposed to 'leave', has become overall less prominent in the Romance verbs under consideration.
Considering this historical information, we can formulate a hypothesis for diachronic research. It seems that the Germanic and Romance verbs of letting have been developing in opposite directions. The Germanic verbs previously denoted both 'let' and 'leave', but now specialize mostly in letting. They have also acquired a higher degree of semantic and syntactic integration with the caused event than have the Romance verbs, and overall seem to stand further on the grammaticalization cline. In contrast, although the historical prototype of the Romance verbs had a prominent component related to 'let', these verbs are now frequently used as lexical verbs of leaving. At the same time, when expressing letting, some Romance verbs have a relatively high proportion of finite clausal complements, which suggests a lower degree of syntactic and semantic integration of the letting and permitted events in comparison with the infinitival construction. Thus, it seems that the position of the Romance verbs (or at least, that of their particular meanings) has become lower on the semantic binding hierarchy.
Importantly, the languages in the middle of the geographical continuum (especially German and French) tend to combine both tendencies, while the languages on the periphery tend to more strongly exemplify one of the tendencies. Of course, this hypothesis requires a thorough diachronic investigation based on solid corpus evidence.