A comparable-corpus based approach to the expression of obligation across English and French

A French-English comparable corpus of political discourse is used to investigate the expression of obligation across the two languages. The aims of the study are to look at the expression of obligation in the comparable genres in the two languages and to compare the contexts of use of deontic modal and semi-modal verbs, i.e. comparing their frequencies in contextual frames characterized by agentivity, polarity and event type. The focus is on the modal and semi-modal verbs must , need to , have to , devoir and falloir . While have to and falloir are more frequent than must and devoir respectively in spontaneous spoken language, the reverse is the case in the political speeches. The five verbs are found to occur in similar contexts within and across the two languages in the genre in question. The study highlights the potential impact of genre on frequency and distribution and the interactions between genre-based patterns and ongoing change in the wider languages.


Introduction
The increased ease of creating genre-specific corpora, together with the recent 'revival' of contrastive analysis over the turn of the 21 st century, have given a new impetus to contrastive genre analysis, shedding light on the functioning of genres across languages.
In his pioneering work on corpus-based genre analysis, Biber (1988:70) drew a crucial distinction between genre and text type1 .In essence, genres are defined by their sociocultural characteristics and text types by their linguistic characteristics.A genre is therefore to be conceived of, in an extensional definition, as the language actually used across events belonging to a recognized sociocultural event type.An event type in this context is some socially established, conventional task.It arises out of what Hyland (2009:211) calls "perceived repeated situations" and is characterised mainly in terms of form (the participants and the sequence of actions that compose the event), distribution (times, frequency, place) and purpose (objectives of the event).A text type, again extensionally, is a set of texts that share certain distributional patterns of linguistic features to a degree not found in other texts.This theoretical distinction between genre and text type is important in at least three respects.Most obviously, it is behind good practice in setting up the criteria for text selection when building genre-specific corpora; such criteria exclude linguistic features.A text belongs to a particular genre to the extent that its context of production and reception can be identified as an exemplar of a some socially-recognized event type.Secondly, if it is in principle possible to independently establish genres and text types, we can discover to what degree they line up, how and to what extent a given genre is linguistically distinguished from others, or is homogenous or heterogeneous.And thirdly, of course, it is sociocultural parameters that provide the situational tertium comparationis that enables contrastive genre analysis to be done at all.
The present study focuses on the expression of obligation in the political speech genre across English and French.The aims are threefold: (i) to look at the distribution of the commonest expressions of strong obligation in the comparable genres in the two languages, (ii) to examine the contexts of use of these modal and semi-modal verbs of obligation, and (iii) in the light of the polysemy of the quasi translation equivalents must and devoir, to compare the use of these verbs in the genre.
Patterns of usage may be genre-dependent.Where there is a 'marked' or atypical distribution of modal markers in a particular genre, there may also be an atypical distribution in the 'equivalent' genre in another language.Testing such a hypothesis would require comparisons among a wide range of matched genres that are beyond the scope of this paper.We shall look out for discourse-pragmatic similarities across the two languages English and French in one matched genre, that of political discourse, to see how they interact with linguistic choice.
In the light of the above discussion of 'genre', it is clear that each genre belongs to a particular discourse community or 'community of practice', and that we cannot therefore strictly speak of a 'political-speech genre'.For the purposes of the present study, the term 'language genre' will be used for convenience to refer to 'equivalent' genres in different language communities.
Section 2 outlines the notional area of obligation and looks in particular at its expression in English and French.The corpus on which the present study is based is described in section 3, and section 4 presents the findings of the study of three English verbs of obligation (must, need to and have to) and two French ones (devoir and falloir).Section 5 provides a conclusion.

Obligation markers in contrast 2.1 The notional area of obligation
This section considers the nature of obligation as an area of meaning, or semantic category, and the issues that arise in attempting cross-linguistic comparisons of the linguistic expression of obligation.
Obligation is traditionally studied under the rubric of modality.However, there is no general agreement among linguists on how to cut up the semantic space of modality into types of modality, or even on the boundaries of modality.A distinction is traditionally made between root modality, which pertains to the degree of necessity of the proposition in an utterance, and epistemic modality, which pertains to the degree of probability of the proposition (table 1).2Other perspectives on modality have flourished, however, such as a division into internal and external modality, or into subjective and objective modality (v.van der Auwera and Plungian 1998, van der Auwera 2001, Palmer 2001, Heyvaert 2003:90ff;Depraetere and Reed 2006 provides a brief and useful summary).Likewise, within the traditional framework, there is no generally-agreed approach to the sub-categorization of root modality.Nevertheless, there is a fair consensus that it is useful to identify an area of modality that involves human-generated precepts about the desirability, or otherwise, of people bringing about particular states of affairs, and this area is usually referred to as deontic obligation and permission3 .

Deontic necessity Obligation Permission
Non-deontic necessity

Dynamic modality
Epistemic modality

Evidentiality
Table 1.The place of obligation in a traditional schema of types of modality From the perspective adopted here, obligation and permission can be seen as poles at either end of a continuum of desirability, stretching from moral necessity (obligatory) to moral acceptability (permissible).'Moral desirability' can be thought of as a broad space ranging from legal requirements and widely accepted social norms to personal ethics, opinions and wishes.Obligation is then a scalar category, ranging from strong obligation to absence of obligation, and the other side of the coin is permission.
Deontic modality as a semantic category can be defined, according to Nuyts, "as an indication of the degree of moral desirability of the state of affairs expressed in the utterance, typically, but not necessarily, on behalf of the speaker" (2006:4).The 'deontic source', i.e. the creator of the obligation, has traditionally been considered an important parameter in modality studies (v. Lyons 1977:825ff).The deontic source may be 'internal' (the obligation is created by the speaker) or 'external' (the obligation is imposed by a third party and reported by the speaker).A difference in source is potentially reflected in a difference in linguistic expression, i.e. grammaticalized or lexicalized.The speaker can choose her expression to indicate the degree to which she assumes responsibility for the obligation.Thus, the imperative is normally understood to express a speaker-created obligation.The English modal verb must tends to be associated with speaker-created obligation, and the semi-modal have to with external, speaker-reported obligation.However, the internal-external distinction is often blurred, and the correlation with must and have to is far from systematic.
For van der Auwera and Diewald, the investigation of "which markers express which meanings and why", and the discovery of how much variation there is across languages is the most important reason for studying modality (2012:123).

The expression of obligation in English and French
Obligation in both English and French is semi-grammaticalized, being typically expressed by a range of grammatical and lexical means, most notably modal auxiliaries, semi-modals (such as English have to, have got to or need to), modal verbs, modal adjectives and adverbs, as well as by imperatives and speech act verbs of the type demand or require.
Obligation, as seen above, stems ultimately from someone's desire or will that some state of affairs (event or situation) be (or not be) realized, and it therefore concerns unrealized situations, especially -arguably exclusively --future ones.As Lyons points out, "there is an intrinsic connexion between deontic modality and futurity " (1977:824).Once the desired action or event is realized, the modality is lost.A distinction is to be made, then, between an affirmation of obligation and a deontically modalized proposition of obligation.This distinction is partially grammaticalized in English by the choice of verbal expression, as can be seen in the following example constructions: [Subject had to V] (affirmation, past), dynamic, such as an accomplishment.
[Subject ought to have V-en] (counterfactual, past) and [Subject must V] (deontically modalized, future).But there is no clear-cut correspondence between linguistic expression and the realized/unrealized divide.In French, the distinction may be made by the choice of verbal form, as can be seen in the following example constructions: [Subject a dû V]' (affirmation, past), [Subject aurait dû V] (counterfactual, past) and [Subject doit/devra V] (deontically modalized, future).
A major difficulty in contrasting expression of modality across languages is that of maintaining a coherent notional tertium comparationis (TC).The natural TC in this case is the notional space of deontic obligation, but, as is clear from the discussion above, the domain lacks clear-cut boundaries.First, both the more grammatical (e.g.modal verbs) and the more lexical expressions (e.g.modal adjectives) most closely associated with obligation tend to range over other domains too, and so are potentially ambiguous or vague.Such vagueness is often exploited by speakers to save face or to implicate something without committing themselves.Speakers can thus use so-called modal-evaluative adjectives such as It's essential to.., It's important to .., or modal-adverbials such as You had better .. and so on to convey an obligation.Second, speakers can employ expressions having as yet scant 'obligation' sense at all, such as want to and need to, but which, in the right context (as in You want to get your hair cut), the speaker can use to impose an obligation on the hearer.Such strategies may be used from considerations of politeness: expressing an obligation can be face-threatening, so that a speaker may choose to use a weak-obligation expression, a neutral one or even a permissive, for what she perceives as an obligation.Examples include such constructions as, in English, be for <someone> to + V, involves V-ing, the time has come to +V, and similarly in French à <quelqu'un> de +V (roughly 'up to <someone> to V'), il s'agit de +V ('it is a matter of V-ing'), le moment est venu pour +V ('the moment has come to V').An example of this kind of inferred obligation is given in (1). (1) L'avenir consistera pour l'Europe à assumer des responsabilités difficiles.[Alliot-Marie 2002] 'The future for Europe will consist of taking on difficult responsibilities.' Depending on the contexts and the expressions, the 'obligation' implicatures, in principle defeasible, can be strong.The communicative intent in (1) is clearly to convey that an obligation exists.Deontic modality, then, is not only expressed through lexemes whose main or sole function is modal, but also through a variety of constructions not usually associated with modality.These usages, of course, can presage incipient grammaticalization of lexical constructions into markers of modality.Within deontic obligation, distinguishing internal from external obligation is especially tricky.In English and French, where the same constructions serve for both, the source of the obligation can only be pragmatically inferred, and is frequently interpreted as vague.The speaker's will is often in harmony with some external will, so that the source is simultaneously external and internal.For example, Je dois arriver avant midi ('I must arrive before noon') may mean that the speaker has a self-imposed (internal) obligation to arive before noon or that she has an externally-imposed obigation to do so, or again that she predicts that she will arrive before noon.But it can also mean all of these simultaneously.This paper focuses on the 'central', semanticized modal markers of obligation, not on whether obligation figures to a similar extent in the two language genres.The verbs have been chosen because (a) to the extent that they carry roughly the same 'weight' of obligation they are often interchangeable and (b) they are dictionary equivalents across French and English.Expressions of mitigated obligation such as should or ought to in English or il faudrait or devrait in French are therefore excluded, as are modal adjectives and adverbs.The aim is to look at high-frequency verbal expressions of obligation in each language and to examine their distribution in the texts, with a view to comparing their contexts of use in the two language genres, and the ways in which the aims of the speakers may be reflected in the linguistic patterns of the genres.
The high-frequency verbs most closely associated with obligation are, for French, devoir and falloir and for English must, have to and need to.All of them are also very frequent in the political speech language genres sampled here.The conceptual space occupied by these five verbs is wide: they can all express a range of contiguous modal meaning and lend themselves to wider or narrower pragmatic interpretations depending on context. 4ccurrences that correspond solely or mainly to deontic obligation, therefore, are not always straightforwardly distinguishable in discourse.Figure 1 maps the five expressions on to a partial semantic map of modality.A semantic map is "a geometric representation of meanings or, if one likes, uses, and of the relations between them" (van der Auwera and Plungian 1998:86).Semantic maps are useful for graphically representing polysemy, partial synonymy and cross-linguistic mapping.In fact they have been most extensively used in typological studies to map grammatical functions and word senses across languages.Such mapping allows typologists to build up a representation of conceptual space based on (a) the ways in which it is grammaticalized and lexicalized by different languages and (b) regularities in semantic shifts over time.5 (1998:90-97).The background area represents a postulated conceptual space, on to which we can map the (prototypical) senses of linguistic expressions.Lower case labels refer to semantic notions; upper case labels refer to (prototypical uses of) lexical items of English and French.The central area corresponds to the 'root modality' row of table 1, that is, to the modal notional area of root necessity.The large circle (centre top) represents deontic necessity6 , comprising that emanating from both internal and external sources.The small ring (centre bottom) represents physical, non-deontic necessity (exemplified in ****).The arrows indicate the directions of frequently-attested diachronic developments (semantic shift or semantic expansion and grammaticalization).Thus, the three rings to the left show non-modal notions which cross-linguistically often share a lexical item with a root modality notion.The two rings to the right show the non-root modal notions of epistemic judgment and future time. 7Again, the arrows show that these notions too often attach to a lexical item expressing a root modality notion.The semantic map therefore provides a rough way of visualising how the senses of lexical items map on to conceptual space and how they relate to each other.For the lexical items that concern us here, we can see in figure 1 that all of them range over two or more notional areas: they are polysemous as a result of semantic expansion.The map also graphically represents cross-linguistic overlap: devoir is used across a wider range of notions than must, for instance.

French 'devoir' and 'falloir'
French makes a basic grammatical distinction between the semi-modal devoir and the impersonal verb falloir, insofar as deontic devoir typically takes a human subject while falloir can only be used with dummy subject il, as in il faut que 'it is necessary that' or il faut + V 'it is necessary to V' (originating from the sense 'want' as in 'be lacking').But this personal/impersonal distinction does not carry over neatly to the semantics or the pragmatics of the two verbs.Falloir can be personalised by the use of one of two constructions: il + oblique personal pronoun + faut/faudra as in il vous faut V ('you have an obligation to V') and il faut que + person +V subjunctive_mood as in il faut que Jean V ('John has an obligation to V').
In usage, devoir is said to be more solemn or more insistent than falloir, while falloir is more often used in 'subjective contexts' where devoir might be interpreted as epistemic or as expressing futurity (Larreya 2004:748-9).In fact, devoir has grammaticalized into a wide polysemous network of uses, so that ambiguous or vague uses abound.Two main areas of usage beyond Obligation have developed: (i) epistemic modality (deduction, along similar lines to the development of English must) and (ii) future marking (roughly equivalent in sense to the English [be [expected] to V] or [be due to V] constructions).
Overall, deontic devoir is more associated with formal registers, while falloir is more colloquial.'must', 'have [got] to' and 'need to' By contrast, English must and have to are said to be distinguished along internal/external obligation lines, must being associated with speaker-created obligation, and have to with externally-imposed obligations (v. Palmer 2001:75).While must and have to are both deontic, need to primarily expresses 'objective' modality (non-deontic necessity) in contexts such as You need to be tall to see over this fence (v. Quirk et al 1985:226).This situation, however, is rapidly changing, so that the most salient fact about these obligation markers is the collapse of must over recent decades and its partial replacement by have to and, more recently, need to.

English
The decline of must in spoken English seems to have affected the root and epistemic uses equally, root uses staying at around one third of all uses through the period 1960s -1990s (Close and Aarts 2010) This affects both written and spoken English, and both epistemic and root must8 .For written English, Mair and Leech (2006) document a sharp decline in the use of must between the 1960s and the 1990s: -29% for written British English and -34% for American -while have to shows a very modest increase and need to increases in their data by a massive 249% for British English and 123% for American.Johansson (2013) shows that for American English need to has now overtaken must.Close and Aarts (2010) find a halving of the frequency of must in spoken British English 1960s-1990s.
It has been suggested that social factors, notably politeness, may be behind the decrease in the use of the 'subjective' forms such as must (Smith 2003).Yet the notion that must is subjective may need revising.Collins claims that "deontic must is more often used objectively than subjectively" (2009:37).His data suggest that 'objective' must conveys weak obligation, as in agentless passives "with an unspecified deontic source having no necessary connection with the speaker, where must merely expresses what is thought to be desirable" (2006:38).An example taken from the corpus of political speeches is given in (2).
(2) majority voting must apply across the board [Rifkind 1997] And Larreya (2004:743n) points out that must is particularly frequent in very formal registers such as political speeches.He suggests that announcements of the decline of must are therefore premature.But, as will be seen, it may be rather that the high frequency in political discourse and other formal registers is due to occurrences of must in particular construction types, and reflects a usage characteristic of the genre, resulting in local pragmatic effects.
In both languages, there is scope for a great deal of ambiguity and vagueness in the use of these modal and semi-modal verbs, due largely to wide-ranging polysemy in the case of French and to major ongoing change in the modal verb system in English.

The corpus
The study is based on a comparable corpus of political speeches from the United Kingdom and from France.The speeches were given by serving government ministers, including prime ministers and, in the case of France, presidents, of the countries.In each case, ministers belonging to governments of different political persuasions are included.The corpus is outlined in table 2.

Table 2. The composition of the comparable corpus of French and English political speeches
Although the corpus is balanced in terms of word count, there are fewer English language speeches, due to their being on average noticeably longer (2,922 words) than the French language ones (2,594 words).This is already an indication that there is no total match between the genres, and that the typical function of the political speech may vary across the different contexts.Epistemic occurrences of must, have to and devoir were excluded.Must being modal only, comparison is restricted to deontic modal contexts (i.e.excluding needed to, would have to, a dû, etc.) 9 .Discourse-marking and highly idiomatic uses, such as I must say or dois-je le rappeler were also excluded.For devoir, clear future-time uses were excluded; however, many uses were ambiguous between expectation and obligation and these were included on the grounds that an obligation reading was pragmatically possible in the context.
The figures for the occurrences of the relevant modal and semi-modal verbs of obligation that were extracted from the corpus using WordSmith Tools (Scott 2012) are given in Table 2.These are the occurrences that were analysed.
The occurrences were coded for the type of subject (of the proposition in the scope of the modal marker); voice; polarity; negative effect (i.e.whether the realization of the proposition affects anyone negatively) 10 .

Overall frequencies of the five verbs
Table 3 shows the raw and normalised figures for obligation occurrences in the corpus of all five verbs of obligation.These are the occurrences that are taken into account in the analysis.The frequency of deontic (root) must is strikingly high, bucking the well-documented trend described in section 2. It is more than ten times higher than that found by Close and Aarts (2010:176) for spontaneous spoken English of the 1990s.Another salient finding is that need to is twice as frequent as have to.Must, then, looks like a preferred choice for obligation.The situation is similar for French.In a study of the use of modal verbs in the speeches of successive French presidents of the fifth Republic, Labbé and Labbé (2013) found very different patterns from both spoken and literary French.While falloir was more than twice as frequent as devoir in their 2.3million-word spontaneous spoken language corpus (2013 table 15) 11 , the reverse was the case on average for the presidential speeches, where, for the four commonest [modal + V] combinations, devoir was more than twice as frequent.The findings of the present study of ministerial speeches are consistent with those of Labbé and Labbé for presidential speeches.These frequencies may simply reflect a conservative style, but if the distribution (the contexts) also differs significantly from that of other genres, it may be seen as genre-specific rather than conservative.9 There is only one occurrence of the form have got to in the corpus, perhaps because of its association with informal speech.

English
10 Politicians recycle sections of previous speeches into new speeches and such passages occur in the corpus.
Where a whole sentence or clause complex is reused in a second speech, only one occurrence is counted; where a sub-sentential unit is reused in a different context, two occurrences are counted.11 Around a fifth of the corpus in question was Canadian and Belgian French.

The five verbs and speaker attitude
For French devoir and English must, obligation is by far the most frequent use: 90% of occurrences of devoir and 96% of occurrences of must.This is markedly different from what has been found for other registers of English.For written language, Collins (2009) found on average that one third of occurrences of must were epistemic, and for spoken language Close and Aarts (2010) found that more than half were epistemic, and around 40% were deontic (root modality).The situation for devoir is complicated by the overlap between the obligation and future-time uses of the verb, so that many occurrences are ambiguous or vague between what is due to happen and what the speaker desires to happen.Only about 7% of occurrences were epistemic in the sense of speaker estimation of a truth value.
In their cross-linguistic study of obligation, Myhill and Smith rejected speaker attitude towards the proposition as an appropriate parameter for categorising obligation utterances in favour of 'negative effect', that is, whether the carrying out of the obligation will have a negative effect on anyone.The contexts of use of must in the data are overwhelmingly those where the speaker not only urges the rightness of bringing about the state of affairs, but where the state of affairs is couched in terms of universal desirability.Have to by contrast collocates with states of affairs that will affect some party negatively.This may be a necessary evil externally imposed (3), or a speaker-created obligation that will be unpleasant for the agent to carry out (4).In about a third of cases (36%) the speaker can be interpreted as favourable to the obligation.This finding for have to echoes what Myhill and Smith found for their latetwentieth century English drama data and which they term "negative effect" (1995:247ff).They did not include must in their study, but found that negative effect was a significant parameter separating have to from other expressions of obligation.It is in the nature of political discourse to focus on the positive and the desirable and exclude the unpleasant; this in itself may go some way to explaining the relative infrequency of have to in speeches. (3) Anyone alleging a violation has to take his case to Strasbourg to obtain a remedy [Irvine 1997] (4) These issues have to be addressed -to ignore them would stoke up fear [Taylor 1996] By contrast, need to, like must, almost always correlates with the speaker being favourable to the obligation, which is by implication external even where we is the subject (5).
(5) a.There are some old ghosts which need to be laid to rest [Mowlam 1998] b.We also need to sort out proper arrangements to appoint Special Envoys [Rifkind 1997] By comparison, the French verbs cover a wider range of obligation types, and their contexts are very heterogeneous.Thus, devoir occurs in contexts of both very strong and very weak obligation, and in both speaker-created and external obligation, with much vagueness.The speaker is generally favourable to the obligation.
Likewise falloir gives rise to vagueness, but again the speaker is usually favourable to the obligation expressed.Falloir differs, of course, from the other four verbs in that it takes an impersonal construction with dummy subject il.
Negation, for the three verbs (must, devoir and falloir) that can be negated to produce a prohibition or obligation not to do something, is rare (less than 5%); speakers focus on the positive 12 .

Types of subject and agentivity
Deontic obligation is typically conceived of as agent-oriented, so that "examples of root must in English would normally be classified as prototypical examples of agent-oriented modality" (Coates 1995:57).Some source of obligation (typically the speaker) desires some agent to effect some state of affairs (typically a dynamic event).The prototypical example given by Coates is (6).( 6) You must finish this before dinner.[from Coates 1995:57] The three elements that form the deontic obligation 'frame' are thus a source of obligation (typically the speaker and if so implicit), an agent (typically the addressee) and a desired state of affairs (typically expressed by a dynamic verb phrase).In (5 these three elements are the speaker, 'you' and 'finish this before dinner'. The aim of examining the types of subject is to look at how the agentivity of obligation is conveyed in political speeches.For all five obligation verbs, second person subjects and first person singular subjects are very rare.Four main types of subject account for the vast majority of occurrences.In order of animacy these are first person plural, collectivities (countries, institutions, companies, professions, and so on), agentless passives, and abstract or inanimate subjects (such as deverbal nouns, mental activities, measures, rules, processes, time periods, and occasional metaphors).Figure 2 shows the percentages of each verb that are accounted for by these patterns.Figure 3 shows the same patterns as occurrences per 100,000 words.
A first person plural subject ( 7) is strikingly frequent across all five verbs, accounting for between 17% (devoir) and 49% (need to) of occurrences (fig.2).For devoir, must, and need to 'we' represents between 54 and 74 occurrences per 100000 words (fig.3) 13 .The referent(s) of 'we' are rarely specified and by implication range from the speaker and their immediate entourage through governments, professions and socio-economic sectors, countries, international fora, Europe and the world.This pattern with 'we' conforms to a typical agent-oriented obligation structure (self-obligation); it firmly includes the speaker in the source of obligation. (7) a. that is not a discourse marker of the type il faut bien le dire ('it has to be said').

Fig. 3. Types of subject and frequencies per 100,000 words
The second most frequent subjects are collectivities.These include the political collective to which the speaker belongs (e.g. the government), the speaker together with his/her addressees, and/or wider social units such as sectors of the population, the country as a whole, Europe, or society generally.But there is a clear division between the unspoken source of the obligation in the speaker and the collective agent enjoined to carry it out.Excepting The third and fourth types of subject reflect two strategies for depersonalizing the obligation by removing the agent.
One is the use of the agentless passive construction (9).The other is the use of an inanimate, abstract noun as subject of the modal verb.In a large proportion of the political speech occurrences of must and devoir not only is the source of obligation barely recoverable, but the agent is missing and unidentifiable from the context, and instead of a dynamic event, we find some situation.The data reveal four sub-types of abstraction.
One way of delinking the obligation from an agent is by nominalizing the state of affairs and using a copula or stative verb.Both devoir and to a lesser extent must lend themselves to this construction (10).
(10) a. une meilleure coopération doit exister entre les acteurs [Voynet 1999] 'better cooperation must exist between the participants' b.Cooperation in the fight against crime must be as instinctive as it is in foreign and defence policy [Major 1994] This strategy can result in an unusual form of redundancy (11).
(11) L'exercise des missions de police doit également s'exercer au plus proche des citoyens (kind of tautology) [Bartolone 1999] 'The carrying out of police duties must be carried out close to the public.' A second sub-type of depersonalization consists of an abstract noun in subject position, such as an idea, process, rule, time or metaphor (12).The only element of the deontic obligation frame that is expressed is the verb itself which prompts the hearer to recreate an obligation scenario, filling in the missing elements.In (12a), hearers will infer a source as usual (such as the speaker and others) but also an agent (political actors, perhaps including the speaker and the hearers) as well as an event that can bring about the situation where flexibility is a component of the future European construction.The speaker's linguistic choice is to evoke obligation with must, but in an atypical mustconstruction, so that the usual frame can only be retrieved by pragmatic inferencing.The effect, and presumably the aim, is to depersonalize the utterance, so that the source can be everyone and the agent no one in particular.This is a face-saving strategy, perhaps, that does not involve abandoning the use of must.Similar strategies are also found with the other modal verbs.
A third strategy is to employ what looks like a reversal of semantic roles: rather than the usual Subject-Agent + Verb + Object-Patient we find Subject-Patient + Verb + oblique or evoked agent, as in ( 13).
(13) a. the lead must come from the real experts [Taylor 1996] b. cette question aussi devra trouver une réponse [Sautter 1999] 'this question will also have to find an answer / an answer will also have to be found to this question' c.Et je demeure convaincue que la culture doit détenir une place particulière dans notre réflexion.[Trautmann 1999] 'And I remain convinced that culture must hold a special place in our deliberations.' Finally, there is a pattern in which the 'role reversal' involves an animate beneficiary.Sequences such as devoir pouvoir and must be able to or have to be able to are associated with this pattern, as in ( 14).In each case, the subject of the obligation verb is the beneficiary of the desired state of affairs.
(14) a.People have to be able to protect their children [Taylor 1996] b.We must instil confidence.All of these patterns occur in other genres too, of course.But the parallels between the English and the French in their frequency and distribution are quite salient: 'prototypical', agentive expressions of obligation occur regularly only with the first person plural, where the speaker is part of both the source of obligation and the agent.Otherwise, non-agentive constructions of the four types outlined are preferred in both languages and fairly consistent across different verbs of obligation (fig.2).

Conclusion
Devoir in French and must in English are both associated with conservative or formal language, as seen in section 2. Both show unusually high frequency in the political speeches.Could it be the case that the very high relative frequency of must in the speeches at a time when must is declining is due to this formal register being conservative and slow to change?It looks unlikely.Perusal of some nineteenth-century political speeches reveals a different pattern of use of must altogether: in addition to discourse-marking uses (I must say), must is used widely across epistemic (predictive) and logical necessity contexts as well as for obligation, with no apparent connotations of positive speaker attitude, rather the opposite.Although this data is inadequate for any conclusions to be drawn, a plausible hypothesis is that must has undergone semantic shift (amelioration) and further subjectification (speakeroriented desirability).Present-day usage of must in political discourse appears to be a recent, possibly genre-related development rather than a hangover from a period before must started to wane.
Both French and English have, on the face of it, a more subjective modal expression for deontic obligation in devoir and must, and a more objective one in falloir and have to, which are both also associated with non-deontic necessity.In both language genres, speakers favour the use of the supposedly more formal and subjective expressions, but use them atypically in impersonalized constructions.Moreover, the desired states of affairs tend to be presented in both langauges through stative or resultative constructions as situations to be brought about rather than as events to be accomplished.Another striking similarity across all five verbs is the proportions of first person plural subjects and rarity of other persons.
Political discourse is characterized by a number of special features: it is often very carefully crafted, every nuance being analysed, and is designed for a wider audience than the immediate hearers; it aims to impress and persuade and may have a hortatory function; it has a ceremonial function that favours rhetorical routines; and above all it deals largely with unrealized states of affairs.Speakers are projecting a vision of realizable states of affairs and therefore have recourse to modal expressions.It is perhaps not surprising, then, that the uses of the main deontic modal expressions in political speeches are similar across English and French.Johansson (2013) found strong genre effects in American English, with need to almost twice as frequent as must in spoken informal language, but the order inversed for academic prose.The contextual overlap in the political speeches between must and need to may therefore reflect some colloquialization or modernization in the English speeches (rather than a flight from a potentially face-threatening expression).Labbé and Labbé (2013) also found a strong genre effect for French presidents' speeches, which showed both less personalisation and a greater use of modal verbs than literary French.Our observation that il nous faut accounts for a fifth of falloir occurrences and appears in the same contexts as nous devons may reflect a move towards a more colloquial style.At the same time it may be a way of emphasizing a strong obligation: the semantic expansion of devoir towards weak obligation and future time results in vagueness regarding degree of obligation.
Overall, we conclude that there are salient similarities in the ways in which each language genre handles the expression of obligation.Such similarities are masked when register-diversified corpora are used to identify ongoing changes in the frequencies and distributions of linguistic patterns, or to compare some notional area or semantic field across two or more languages.It is therefore interesting to undertake contrastive-linguistic studies at a relatively fine-grained level of social context matching: a genre may have its own twist on an evolutionary dynamic, stemming from the strategies and situational constraints guiding its speakers.

Figure 1
Figure1is inspired by van der Auwera and Plungian's discussion of directionality in modal meaning change(1998:90-97).The background area represents a postulated conceptual space, on to which we can map the (prototypical) senses of linguistic expressions.Lower case labels refer to semantic notions; upper case labels refer to (prototypical uses of) lexical items of English and French.The central area corresponds to the 'root modality' row of table 1, that is, to the modal notional area of root necessity.The large circle (centre top) represents deontic necessity 6 , comprising that emanating from both internal and external sources.The small ring (centre bottom) represents physical, non-deontic necessity (exemplified in ****).The arrows indicate the directions of frequently-attested diachronic developments (semantic shift or semantic expansion and grammaticalization).Thus, the three rings to the left show non-modal notions which cross-linguistically often share a lexical item with a root modality notion.The two rings to the right show the non-root modal notions of epistemic judgment and future time.7 Again, the arrows show that these notions too often attach to a lexical item expressing a root modality notion.The semantic map therefore provides a rough way of visualising how the senses of lexical items map on to conceptual space and how they relate to each other.For the lexical items that concern us here, we can see in figure 1 that all of them range over two or more notional areas: they are polysemous as a result of semantic expansion.The map also graphically represents cross-linguistic overlap: devoir is used across a wider range of notions than must, for instance.

French
a quarter and a third of occurrences of the five verbs specify a collective agent (8).(8) a. .. new democratic institutions have to be founded on agreement and consent.they have to command the support of both communities here [Mowlam 1997] b.L'ESA doit rester au coeur du dispositif spatial européen [Schwartzenberg 2000] 'The ESA must remain at the heart of European space activities.' Les problèmes doivent être évoqués ici, là où ils prennent leur source [Sarkozy 2002]'Problems must be raised here, where they originate' b.The fighting there must be brought to an end.[Major 1995] c.First, the fighting needs to be brought to an end[Hurd 1995]

(
12) a.That is why an essential component of the future European construction must be flexibility.[Major 1994] b.Le concept de police de proximité répond évidemment aux évolutions sociales, mais il doit avoir un contenu plus fort [Sarkozy 2002] 'The concept of neighbourhood policing is of course a response to social change, but it must have a stronger component.' [Moscovici 2001] 'It's a real problem for European citizens, to which we must find solutions.'e. Il nous faut simplifier nos structures nationales pour les faire mieux coopérer au niveau européen.[Fabius 2001] 'We need to simplify our national administrations to make them cooperate better at European level.' First we have to get agreement in the talks [Mowlam 1997] b.We must achieve a sustainable consensus on pensions policy [Harman 1997] c.We need to pursue agreement with the Russians [Hurd 1995] d.C'est un vrai problème pour les citoyens européens, auquel nous devons trouver des solutions.