A deontic possibility modal in Latvian: Personal vs. impersonal uses in a corpus.

A corpus-based study of a dedicated deontic possibility modal in Latvian focuses on its impersonal variety with a non-canonical subject in the dative. Normally, drīkstēt ‘may’ and other possibility modals have nominative subjects, while dative subjects are found with expressions of necessity. As distinct from other constructions where non-canonical dative subjects are experiencers, the modals are also used with inanimate subjects. A frequent ellipsis of lexical verbs in the impersonal uses of drīkstēt not only reflects the informal style of the construction but also points to the Russian možno / nel’zja as a possible source, especially when combined with an object in the accusative referring to food. The Russian construction has a meaning of deontic possibility, but its use is restricted to animate subjects. The article claims that the animacy restriction was lifted in Latvian under the influence of the necessity modals in contexts of prohibition.

1 Impersonal modals Drīkstēt, which is a dedicated deontic possibility modal in Latvian, has an impersonal variety that has not so far made its way into the existing descriptions of Latvian modals. This article aims at filling the gap by presenting a study of the construction as it is used in the Latvian Web Corpus (lvTenTen14) of 530 mln words. The main question I am trying to answer is why drīkstēt is the only Latvian modal that has impersonal as well as personal uses.

On impersonal constructions
Impersonal uses of modals here are understood as those with the experiencer-like subject in the dative rather than nominative, compare (1) and (2). According to Malchukov & Siewierska's structure-based classification of impersonal constructions (2011, 2), (1) represents the type "with a subject that does not display canonical subject properties". Another type of impersonal construction, that "with an argumental subject which is not fully referential", as in (3), will only be mentioned in connection to the former.

On expressing modality in Latvian
Necessity is conveyed by the verb vajadzēt 'need' and the verbal category of debitive, see Holvoet (2007;2001, 9-62) for more detail. Both expressions are roughly synonymous and cover a wider range of meanings including dynamic, deontic and epistemic modality. (4)-(5) are examples 1 with the dynamic meaning, but our main interest lies with another feature they all have in common, i.e. the experiencer in the dative. A similarly wide range of meanings in the field of possibility is expressed by the verb varēt 'can', which, in contrast to the expression of necessity, is a personal construction with a subject in the nominative. Out of the three possibility verbs only drīkstēt is also found in an impersonal construction with the dative, which is the focus of this article.

On dative subjects in Latvian
Apart from displaying non-canonical marking, dative subjects of Latvian modals also lack agentivity. On the whole, it is not uncommon for an experiencer to receive dative marking in Latvian. Apart from the necessity constructions mentioned above, see also examples from Holvoet & Nau (2014, 21), with some of the verbs oscillating between nominative and dative marking of their first arguments in a manner reminiscent of competing patterns found with drīkstēt.  Holvoet (2013;2015) applies the notion of quasi-subjects to dative experiencers and states that in case of more than one participants, as in (15), subject properties are spread over several NPs. The NP in the dative appears to be the least oblique as it is usually in topic position at the beginning of the clause and animate (Holvoet & Nau 2014, 24-28 However, not all dative subjects are necessarily given the experiencer role by the verb. Modals select an infinitival complement, bringing us to the problem of raising vs. control verbs. Traditionally, epistemic modality is associated with a raising structure, and root modality with a control structure. As Holvoet (2007, 147) points out, in case of the epistemic use of vajadzēt this would involve raising the subject of the embedded clause to the position of dative complement, as in (16)  An alternative solution suggested by Holvoet is to treat the dative subject as belonging to the infinitival clause because it is not unusual for infinitival embedded clauses to have overt dative subjects in Baltic, as in example (17) from lvTenTen14. A linear position of the dative subject is not indicative of its place in a syntactic structure because of free word order in Latvian.

123
(17) Ir bīstami presidentam lietot tādu be.prs.3 dangerous.adv president.dat.sg use.inf such.acc.sg jēdzienu. notion.acc.sg 'It is dangerous for a president to use such a notion.' Holvoet suggests that the deontic example (18) might also contain a dative subject in the infinitival clause. His argument is that (18)  Holvoet does not say it explicitly but it can be inferred from his text that the dative subject gradually becomes differentiated from the experiencer as a modal extends its scope from dynamic and deontic to epistemic use, which is a development accompanied by an increase in the degree of grammaticalization. Holvoet (2007, 148-149) gives the corresponding facts as an argument against radical differences in the syntactic structure between deontic and epistemic uses, and views modal verbs as "an area of inter-determinacy between 'raising' and 'control'".
Separation of the experiencer from the subject of a modal verb corresponds to what Barbiers (1995, 141-150) calls directed and non-directed deontic modality depending on whether the subject of the sentence is also the person who is given permission/assigned an obligation. Non-directed deontic modality is thus grouped together with epistemic modality as both operate on the entire proposition. From the syntactic viewpoint, these considerations provide grounds for treating non-directed modals as raising together with epistemic modals, and directed modals as control verbs together with dynamic modals (de Schepper & Zwarts 2009). This article does not further investigate the question whether the dative subjects are raised from the embedded clause or not. But the semantic difference between directed and non-directed modality corresponding to a higher degree of grammaticalization is useful when explaining inanimate subjects that cannot be experiencers and are therefore never found with non-modal verbs like sāpēt 'hurt', etc., as cited above. The examples in (20)-(21) are from Daugavet (2018) I will look more closely into the differences between dative and nominative subjects in the next sections.

Impersonal uses of drīkstēt in the corpus
Examples with the experiencer in the dative are very rare, which explains their absence from literature on Latvian modals. Indeed, out of 99 4 randomly selected instances of the affirmative drīkstēt from the Corpus only two were found to contain it, and only one such example was found in the sample of 98 instances of the negated nedrīkstēt. The corresponding figures for the experiencer in the nominative are 60 and 44, respectively, which reveals that the impersonal construction is a rather peripheral option.
Additionally, the two samples showed that the present tense (ne)drīkst is a prevailing (about 80%) form both with and without negation, which was a defining factor in shaping other samples.
In order to provide data for the research, sequences of drīkst and nedrīkst preceded by a noun or a pronoun in the dative were extracted from the corpus. After manually selecting those examples where the dative corresponded to the experiencer-like subject of the modal, the following figures were obtained ( Table 1). The higher frequencies of datives with the negated nedrīkst are easily explained by the overall higher frequency of nedrīkstēt in comparison to drīkstēt in the Corpus (168 vs. 71 ipm).  In what follows, these samples are compared with four other samples containing (ne)drīkst in combination with a preceding nominative form of a (pro)noun, obtained by a similar procedure from a list of randomly selected sequences.
The data revealed differences between the dative samples and the nominative sample with respect to animacy of (quasi-)subjects and ellipsis; although negation, too, is an important factor.

Animacy
All (quasi-)subjects were divided into four groups comprising animate nouns (humans and animals), inanimate nouns (for example, nosaukums 'name', māja 'house') and collective nouns (karaspēks 'army'). Because of the small size of the samples, presenting the share of (in)animate nouns as a percentage of all subjects in a sample seemed uninformative. Instead, a ratio between inanimate nouns and animate nouns was chosen for comparison, ignoring the collective nouns altogether. The main factor behind the distribution of animate vs. inanimate subjects is negation. In combination with nedrīkst, inanimate subjects in the nominative are almost twice as frequent as the animate ones. The usage is typical of instructions and regulations, the most frequent verbs being būt 'be ' and pārsniegt 'exceed', as in (24). Without negation, the ratio is reversed, with inanimate nominative subjects comprising only 0.2 compared to animate nominative subjects. The same tendency is maintained by dative subjects, as their ratio to the number of animate subjects is 0.4 with negation and 0.1 without negation. Probably reflecting a less formal character of the source texts, the verb pārsniegt is only found once in the dative sample of nedrīkst. The example in (25)  [This is the floor temperature on coldest winter days. Normally, the floor temperature is lower.]' But even with nedrīkst, the number of inanimate dative subjects never exceeds the number of animate dative subjects, and one may conclude that the impersonal construction is more preferable. This conclusion would be in full agreement with the fact that the dative is associated with the experiencer semantics, clearly absent from inanimate subjects. Pronouns show the same tendency, although with smaller numbers because of a much higher frequency of animate referents in general.

Ellipsis
With both dative and nominative subjects the main verb can be omitted, with various degrees of recoverability. Obviously, the verb is easiest to recover if it is present in context, often within the same sentence. The table reveals a tendency which justifies space devoted to ellipsis in this article that almost two thirds of the dative data are those with ellipsis, which is a striking contrast with less than 10% in the nominative data. In other words, when a speaker uses (ne)drīkstēt with the dative, the chances are high that the lexical verb is omitted, and it is only in 16 out of 60 ellipsis examples that the lexical verb is present in context. For the nominative data, the verb is almost always preserved, and when it is not, it is still found in the context in half of the sentences where ellipsis is found. Evidently, these figures reflect a stylistic difference between the dative and the nominative construction, the dative construction being characteristic of informal style.

Sources
In its original meaning 'dare', still extant, the verb (ne)drīkstēt combines with a nominative subject, see example (35) from Holvoet (2007, 157). Therefore, the use of a dative subject with (ne)drīkstēt must be a new development. The question then is what made this change possible, and why it did not take place with other possibility verbs, that is, the universal varēt and the dynamic spēt.
Since inanimate subjects are much more likely to appear with negation, inanimate dative subjects might have been introduced from the expressions of necessity, namely, vajadzēt and the debitive. These two constructions are only possible with the dative and, when negated, they are capable of conveying prohibition which is also the meaning of nedrīkstēt. Introducing inanimate dative subjects into the drīkstēt construction makes the latter susceptible to the issues that remain unresolved for vajadzēt and the debitive, that is, whether drīkstēt can be interpreted as a raising verb.