Discourse particles in early English: Clause structure, pragmatics and discourse management

This article presents a case study on the shifting interaction between clause structure, information structure and discourse organisation in the history of English, as evidenced by the development from Old to Middle English of what we will call discourse particles: discourse-cohesive devices grammaticalised from adverbs. These include the Old English elements þa, þonne, nu, when used in a clause-internal position. We will show that these discourse particles have the following properties: 1) pragmatically: they express the response of the speaker to the context/common ground shared by speaker and hearer, and thus play a pivotal role in common ground management; 2) relation to clause type: the pragmatic meanings of discourse particles are co-determined by the illocutionary force of the clause types in which they occur, including main clause questions, imperatives, hwæt exclamatives, and correlative subclauses; 3) syntactically and information structurally: discourse particles occur in a fixed position in the clause that separates discourse-given from discourse-new information. Discourse particles thus form a subtle lynchpin between pragmatics, discourse management and clause structure. Some particles were lost in the transition to Middle English, but then and now continued to be attested in questions and imperatives, in the same clause-internal position as in OE. Towards the end of the ME period, we see a positional shift to clause-final position, though maintaining the discourse linking character of the particle. This change is due to a syntactic change tightening the use of strict SVO word order and narrowing the use of clause-medial material.


Introduction
This article presents a detailed treatment of what we will call "discourse particles" in Old English (OE), and it sketches their subsequent development over the Middle English (ME) period.
An example of a discourse particle is the use of þonne in questions as in (1) and (2)  The meaning of þonne as in (1-2) is not its literal meaning as a temporal adverb, since there is no temporal connection. The particle, we will claim, has a pragmatic meaning: it represents a speaker's response to the context: þonne expresses surprise on the part of the speaker (writer), and perhaps a degree of exasperation about the circumstances expressed in the context. We will argue in section 3.2 that the particles under discussion here have such a pragmatic meaning contribution. The position of particles is fixed: pronominal subjects and other discourse-given subjects typically precede it; nominal subjects, particularly discoursenew ones, generic and focused subjects, typically follow it ( van Kemenade & Los 2006; van Kemenade, Milicev & Baayen 2008; van Kemenade 2009; van Kemenade & Milicev 2012).
We will give a detailed analysis of discourse particles here that is inspired by the literature on their present-day Dutch and German counterparts (Dutch dan, German denn), which are representatives of what are often called "modal particles" or "discourse particles" (e.g. Abraham 1991;Zimmermann 2011;Coniglio 2012). These particles in Dutch and German come with special properties and restrictions: 1) they have context-sensitive meanings which are semantically bleached with respect to their adverbial counterparts; 2) they occur in some types of sub-clauses but dominantly in main clauses, each of them in specific clause types such as questions, imperatives, exclamatives and are thus taken to be related to the illocutionary force of the clause; 3) they divide the clause into domains for discourse-given and discourse-new information. We have chosen to call the elements under study here "particles", in line with the literature on the present-day continental West-Germanic languages on which we rely, because the Dutch and German cognates of the OE particles under discussion here are so similar in their behaviour. We note, however, that the properties of particles as identified here line up with the more general literature on discourse markers, and on the literature in the English studies tradition, in which elements of this type are more often called discourse markers (e.g. Enkvist & Wårvik 1987;Schiffrin 1987); pragmatic markers (e.g. Brinton 1996;2017); connectors (e.g. Lenker 2010). The work on Old English particles here is, however, a novel addition to this literature.
We will present a corpus-based study of the texts from 850-1050 of the parsed York-Toronto-Helsinki Parsed Corpus of Old English Prose (YCOE, Taylor et al. 2003) with a detailed description and analysis of three Old English discourse particles: þa, þonne, nu (we will motivate this choice in section 3). The article is organised as follows: section 2 will first define the term 'discourse particle' and discuss the properties of particles in present-day Dutch and German, as background and input for our analysis of Old English. Section 3 will be concerned with Old English particles. It will first outline how we arrived at our choice of particles, followed by a detailed discussion of the meaning of particles in relation to the clause types in which they occur, their position in the clause, and an analysis of particles in Old English clauses in relation to their discourse context. Section 4 will sketch the ME development. Section 5 will summarise, conclude and outline some further avenues of research. addressee on the relation between the propositional content of the clause and the common ground between speaker and hearer. Particles are used abundantly in spoken Dutch and German. They are presuppositional in the sense that they express the speaker's response to shared knowledge between speaker and audience in the common ground/context. Particles form a closed word class, they have an invariant form and are uninflected. They are typically unstressed, and they occur in fixed positions in the clause. The particles under study here are optional in the sense that they can be left out without causing ungrammaticality, but their use does add pragmatic meaning. A recent literature survey is Bayer & Struckmeier (2017). The terms "discourse particle" and "modal particle" are often used for the same elements by different authors (e.g. Zimmermann 2011;Coniglio 2012). 4 The starting point for our discussion of the relation between particles and the Common Ground (CG) is Thurmair (1989), who gives a detailed discussion and inventory of the contextual meaning of individual particles in present-day German, with survey tables of the meaning components of particles, and their occurrence in clause types. Her general take on particles is stated on p. 2 (our translation): Modal particles essentially serve to tie in an utterance with the cohesion of the interaction. They serve to refer the interlocutors to shared knowledge, to assumptions or expectations on the part of the speaker or hearer; it can be shown that there is a reference to a previous utterance, or that it allows a characterisation of the importance attributed to the utterance by the speaker. It is to this extent that modal particles each specifically modify illocutionary types (Thurmair 1989: 2) 5 These general observations are best illustrated by some examples of individual particles. We discuss two of them, which will both prove particularly relevant for Old English as well. Thurmair (1989: 163) observes that the use of denn 'then', the cognate of present-day English then, Dutch dan, Old English þonne, marks reference to the immediate context, and is used when something unexpected is questioned (see also Bayer 2012). This is evident from the fact that denn is rare in questions that open a dialogue (Thurmair 1989: 165 The use of denn marks surprise on the part of Mia that Jo would have time to go to the pool. This can easily shade into disapproval/reproach, as in (4): 4 We used the term "modal particle" in previous work (Links et al. 2017). The results presented in this article show that the function of particles is primarily a discourse-cohesive one, which in addition expresses speaker response (though not necessarily attitude). This is why we now prefer to use the term "discourse particle". 5 The German version of the text is: "Im wesentlichen dienen Modalpartikeln dazu, eine Äußerung in den Interaktionszusammenhang einzubinden. Mit ihnen kann auf den Gesprächspartnern gemeinsames Wissen verwiesen werden, auf Annahmen oder Erwartungen von Sprecher oder Hörer, es kann ein bestimmter Bezug zu einer vorangegangenen Äußerung angezeigt werden, oder es kann der Stellenwert, den der Sprecher der Äußerung beimißt, gekennzeichnet werden. Insofern modifizieren die Modalpartikeln auf je spezifische Weise Illokutionstypen." (4) Du bist ja immer noch nicht angezogen! Wir wollten doch heute etwas you are prt always still not dressed we wanted prt today a little früher los. Hast du das denn vergessen? earlier off have you that prt forgotten 'You are still not dressed! We were planning to leave a little earlier today, didn't we? Did you forget?' Questions with denn are often rhetorical. This fits in with Thurmair's account, given that the answer to a rhetorical question is assumed to be obvious from the context. Foolen (1995; presents some observations on present-day Dutch that are based in part on Thurmair (1989), characterising Dutch particles, too, as having context-dependent meanings, and serving to navigate the common ground between speaker and hearer. They belong to a closed class of elements, are unstressed, and are positioned between given and new information. For the particle nu/nou (lit. 'now'), the cognate of OE nu, Foolen (1995: 65) observes that it indicates impatience on the part of the speaker, as in the imperative in (5): Kom binnen. vs. Kom nou binnen. come in come prt in 'Come in. ' 'Do come in.' Vismans (1994) notes that nou occurs in questions and imperatives, and he arrives at a similar meaning characterisation. The meaning of nu/nou in (6) is thus semantically bleached: a concrete indication of this is that nu/nou need not always refer to 'this moment', as it can be used in combination with another temporal adverb like morgen 'tomorrow', expressing future tense (Vismans 1994: 64). We will follow up the spirit of the literature discussed in this section in our approach to Old English, more especially because particles in Old English are relatively uncharted territory and require some careful description first.

Old English particles
Discourse particles in Old English are not particularly well-studied so far. Present-day English has no specific discourse particles in the sense studied here, even though there is an extensive literature on discourse/pragmatic markers, also called connectors (Schiffrin 1987;Brinton 1996;2017;Lenker 2010;Haselow 2011;Fraser 2009 and many others). Van Gelderen (2001) deals specifically with what she calls "mood particles" in the history of English, which in her treatment seem to include a broader category of adverbs that may express speaker attitude, in particular epistemic adverbs such as no doubt, realistically, truly, certainly. The elements we call particles here are distinct from epistemic adverbs in that they form a closed class of grammatical words with a specific position in the clause, and they are generally taken to be short forms derived (grammaticalised) from adverbs, as discussed above. This distinguishes them from epistemic adverbs. Haselow (2011Haselow ( , 2012 argues that utterance-final then in questions in present-day spoken English has many of the properties identified here for discourse particles in Old English (purportedly acquired since late Middle English), even though their position is essentially different. We will come back to this in section 4.
There is, to our knowledge, no treatment of the Old English particles as defined here that addresses their properties in a comprehensive, systematic, and corpus-based way. Mitchell in his monumental (1985) work makes no mention of the particle use, and he treats it as an adverbial use. The forms þa and þonne have at least three uses, however. Let's consider some examples in which all these three uses are attested: van Kemenade (6-7)) or conditional sub-clause (in (8)). Þa 1 (in (6)), þonne 1 (in (7)) and gif 1 (in (8)) are conjunctions introducing the adverbial/conditional subclause. þa 2 (in (6)) and þonne 2 (in (7-8)) are resumptive adverbs introducing the main clause; þa 3 (in (6)) and þonne 3 (in (7-8)) represent what we call the particle use here. The crucial property distinguishing particle use from the adverbial and conjunctional uses thus is that the particle is in a clause-internal position rather than a clause-initial one.
There is no watertight way of distinguishing particle uses from adverbial uses directly on the basis of historical texts. Clear evidence for particle status, however, is their uniform position in the relevant clause types. A further important argument is that the choice of particle clearly co-depends on the clause type, as we will see below. This would be entirely unexpected if particles were adverbs. Particle uses feature particularly in the special clause (illocution) types under consideration here, rather than in, for instance, subject-initial main clauses, which are often pragmatically neutral. Assessing the meaning of particles in Old English texts is of course to some extent a question of interpretation. We will discuss their meaning in detail with specific examples in context for each clause type.

The choice of particles and clause types
The two most obvious particles to consider are þa and þonne, as their particle status has already been established on syntactic grounds in previous work. We considered several rather practical criteria to come to an appropriate choice of particles beside þa, and þonne. The first is whether the potential particle has a cognate in Dutch and/or German which has particle uses. The second is whether we could pinpoint pragmatic meanings. The third is whether its position in the clause is similar to that of þa and þonne.
We did an exploratory pilot study in order to trace further potential particles in the Old English texts, considering four plausible candidates: eac (lit. 'also', cf. German auch, Dutch ook); nu (lit.'now', cf. Dutch nu/nou); þeah (lit. 'though', cf. German doch, Dutch toch); giet ('lit. 'yet' overlapping in meaning with German doch, Dutch toch). Nu does appear to meet these criteria, and it is thus included here -we will present more detail in section 3.2. Þeah is most often used in combination with other adverbs (in particular in the combination swa þeah 'as though, as if', often a conjunctional use). We therefore decided to leave it for further research. Giet used in isolation yielded very few examples; also, giet often co-occurs with þa (þa giet, lit. 'then yet, then still', with a strong temporal connotation). We leave this, too, for further research. Closer inspection of contexts in which eac occurs showed that it really only occurs in its additive meaning. A further addition might have been la 'lo', which we initially added because it is often thought to be an expressive element used in combination with the interjection hwaet. However, the use of la turned out to be much more varied than that of the other particles, pragmatically as well as syntactically, and it is not restricted to specific clause types. We thus decided to concentrate on the three particles þa (lit. 'then'); þonne (lit. 'then'), and nu (lit. 'now').
We now turn to discussion of the clause types included. Thurmair (1989) distinguishes the following main clause types for German: wh-questions and yes/no questions, imperatives, optatives, exclamatives and wh-exclamatives. These constitute the special illocution types in which we should be likely to find particle uses: the illocutionary force of inquiring for questions, ordering for imperatives, wishing for optatives, exclamation for the types of exclamatives (cf. Searle & Vanderveken 1985). We have simplified this set of main clause types as follows: 1) we did not distinguish the two question types in our searches, as the number of examples is relatively limited and the same particles occur in both, with the same readings as far as we can tell. 2) Exclamatives needed some special consideration. The total number of wh-exclamatives in the YCOE corpus is very limited and we therefore left them out. 6 The only way of achieving a minimally robust and internally consistent set of exclamatives is the set of declarative main clauses introduced by the interjection hwaet. Their exclamative character is often reinforced by a particle; we will see below that this particle is most typically þa. Walkden (2013) takes hwaet to be a wh-word in origin, and sees no real grounds for distinguishing the wh-word from the interjection. In terms of word order, hwaet-exclamatives waver between main clause and subclause behaviour. They presumably do not present one unified clause type, and may embed question and conditionals. We will come back to this in section 3.2.1. 3) We left out optatives, since these are not tagged as a category in the YCOE corpus and therefore cannot be distinguished in any straightforward way. 4) We added subject-initial clauses to get a clear view of the potential occurrence of particles in a clause type that does not (necessarily) have a special illocution type (this possibility cannot be excluded for subjectinitial clauses, but it is fair to assume that most of them are indicative and pragmatically more neutral than the clause types mentioned above).
We have thus included three types of main clause: questions, imperatives, hwaet exclamatives. We also included one type of subclause: correlative clauses which are known to feature þa and þonne quite prominently, as exemplified by (6-8) above. Potential correlatives are coded in the YCOE corpus as adverbial subclauses that directly precede a main clause. These include temporal adverbial clauses with þa or þonne as the conjunction, as in (6-7), as well as gif-conditionals with þonne, as in (8). They have a discourse-structuring function, which is reinforced by the use of particles (Links et al. 2017).

A note on the data
The data are based on detailed searches in the syntactically annotated YCOE corpus. We have included the texts from the O2 period (850-950), the O3 period (950-1050), and those dated O23 (texts composed in O2, but the manuscript dates from O3). They are specified in Appendix 1. This choice was made in order to be able to search a sizeable part of the corpus, including all the major 9 th and 10 th century prose texts, while avoiding the O4 texts (1050-1150), because it is unclear in many cases whether they represent "real" Old English, as they are translations and/or adaptations from earlier texts. The particles studied here occur robustly in the 9 th century texts translated at the court of King Alfred the Great. We believe, however, that they are a thoroughly Germanic phenomenon: they have cognates with very similar properties in the present-day West Germanic languages, and section 3.2 will show that they are Germanic in their historical origin as well as their meaning and behaviour. For instance, particle use features particularly prominently in the Old English Boethius, which famously represents a non-slavish translation from Latin with frequent gaps, so that the Vorlage can often not be identified. Its robust use of particles may also be reinforced by the fact that it contains a lot of dialogue, i.e. spoken language, the favoured context of particles.
We used xml versions of the texts in the YCOE corpus, searching them with XQuery in CorpusStudio (Komen 2009;, creating features for clause type, period, particle, subject type, and the relative position of particle and subject. We queried for clause types as follows: 789 • main clause questions: CP-QUE which does not contain a complementiser C. • imperatives: IP-MAT (matrix clause) whose verb is coded as imperative (VBI).
• hwaet exclamatives: IP-MAT introduced by an interjection phrase INTJP that dominates a W-pronoun. • correlatives: adverbial subclauses CP-ADV, which dominate a P, which in turn dominates gif, þa, or þonne as the conjunction introducing the sub-clause. We included only those adverbial subclauses that precede the main clause they are part of. • Subject-initial clauses: IP-MAT which has the subject as the first constituent.
For each of these clause types, we searched for particles within the clause: þa, þonne and nu are coded as ADV^T (temporal adverbs). Our searches yield the results presented in Table 1. The total numbers for the clause type are raw numbers. The number of particles within the clause type was established after some manual correction, eliminating invalid results due to the internal complexity of some clauses comprising a large number of adjuncts, appositions and the like. Table 1 gives the numbers for particles per clause type. To get a clearer view of the frequency of occurrence of particles per clause type, Table 2 gives the number of particles per clause type normalised to their occurrence per 100 clauses. This was done in order to give an approximate comparison of the occurrence of individual particles across clause types. It should be noted that normalisation renders an approximate comparison, as the total occurrence of each clause type varies quite substantially, as seen in the second column in Tables 1 and 2. The cells in Table 2   ing preferences between particles and clause types. Thus, questions and imperatives clearly favour nu and þonne; exclamatives, preposed adverbial clauses and subject-initial clauses show a strong preference for þa; preposed conditionals favour þonne.
We will discuss the results in detail in the next section.

The meaning of Old English particles in context
We will now consider the various particles in turn and discuss their contribution to the meaning and the discourse cohesion of the clauses they occur in, including the question of the choice of particle in relation to clause type. Tables 1 and 2 show that individual particles have clear preferences for clause types. These preferences would be entirely unexpected if they were adverbs rather than particles. This provides initial support for our hypothesis that their meaning contribution is non-lexical and relates to illocutionary force. Let us start with a semantic comparison of þa and þonne, which is important because they are etymologically related and have identical literal meanings. Yet, Table 2 shows that they are largely complementary in their choice of clause type: þa occurs dominantly in temporal correlatives and hwaet exclamatives, whereas þonne occurs in questions, imperatives and in preposed conditional clauses. According to the Oxford English Dictionary (entry thenne, then, adv.), their base form is the demonstrative pronominal stem þa; þonne (and related Germanic forms) is derived by the addition of particles such as ne. This shared etymology in a deictic form fits naturally with the notion that particles refer to the context. The added particles contribute a further deictic meaning component: the German cognate form dann is taken by the Grimmsches Wörterbuch (available at http://dwb.uni-trier.de/de/) to derive from the accusative of the Gothic demonstrative sa. Persson (1893) and van Helten (1910) reconstruct an instrumental or deictic form for the suffix -ne. It is tempting to hypothesise that the -ne part of þonne relates to a negative element, or was perhaps at some point reinterpreted as a negative element; this makes semantic sense as the clause types in which þonne occurs are negative polarity contexts (Gertjan Postma p.c.), including questions, conditionals, and imperatives (Giannakidou 1998). This is questioned for imperatives by an anonymous reviewer. Links (2019) shows, however, that imperatives in Old English also occur as the protasis of conditionals, as in (9): (9) (CP:51.395.33.2689) Berað eowre byrðenna gemaenelice betwux iow ðonne gefylle ge godes ae. Bear your burdens in common between you then fulfil you God's law 'If you bear your burdens jointly between yourselves, then you will fulfil God's law.' Whenever these imperative conditionals are correlative as in (9) they trigger the use of þonne as a resumptive pronoun, rather than þa. Giannakidou (1998: 130 ff.) explicitly discusses imperatives in terms of non-veridicality: their meaning cannot rely on truth conditions. This confirms, we claim, that the key semantic property shared by þonne contexts is non-factuality. The idea that þonne includes a negative element in its etymology thus suggests an important clue to the distinct distributions of þa and þonne. We hypothesise that þa finds its origin in a (spatial) deictic meaning that became temporal, and was subsequently bleached to an exclamative meaning, and, in temporal correlatives, to an action marker, drawing attention to the important new information to come (cf. Links et al. 2017). A key point is that the temporal origin of þa implies that the events it sequences have actually happened; they are factual: Mitchell (1985) refers to them as actions completed in the past). Þa is, by the same token, descriptive, and it structures narrative discourse, in line with observations about þa as an adverb in the literature, e.g. Enkvist (1986);Wårvik (2011;; Links et al. (2017). Þonne is used in contexts that are nonfactual: questions, conditionals, and imperatives. We will exemplify this in detail in the following subsections.
Nu is found in the same clause types as þonne, but its original temporal meaning 'at this moment, straightaway' ensures that it occurs predominantly in interaction contexts in the first and second person (questions and imperatives). We hypothesise that it was grammaticalised to a reinforcer expressing surprise or irritation in questions, and impatience in imperatives, which may shade into requests, similar to present-day Dutch as discussed above.

Þonne
The reader will recall that the German counterpart of þonne is denn and occurs in questions where it marks surprise/unexpectedness on the part of the speaker. This reading sits well with the robust use of þonne in questions in Old English. An example is (10). Þonne in (10) expresses surprise on Maria's part at being pregnant, since she had no contact with a man.
There is a similar reference to context in the imperative exemplified in (11): þonne refers to the follow-up of the (nonfactual) context: the perception is the expected result of the preceding discussion. This example allows an almost causal reading. Why, you will say he is even more unhappy than the other, because of the added evil. Am I not bound to think so? Then he said: then say that it seems so to you:] Ongit þonne mid innewearde mode þaet ða yflan habbað symle hwaethwugu see prt with inner mind that the evil have always something godes on gemong hiora yfle; good in among their Evil 'Mark then with your inner mind that the evil always have something good among their evil;' Đonne is the privileged particle in gif-conditionals. (12) illustrates that þonne introduces the (possibly negative) expected result which needs to be acted upon, and what needs to be done in that case; it seems to strengthen the condition and could be paraphrased as 'in case that': van Kemenade  Hwaet exclamatives also feature þonne to some extent, even though its use is vastly outnumbered by that of þa. In part, this is no doubt due to the fact that quite a few exclamatives are part of a series of questions or conditionals. An example of an exclamative that is part of a conditional is given in (13): (13) (CP:45.335.11.2260) Hwaet se ðonne unryhtlice talađ, se đe talađ đaet he sie unscyldig, gif Lo he prt wrongly argues, who that argues that he is innocent if he đa good đe us God teo gemanan sealde, him he the good(s) that us God to common use gave, himself synderlice aegnađ. especially appropriates 'He argues, therefore, wrongly, who argues that he is innocent, if he appropriates especially to himself the good things God gave us for common use.' The examples in this section show that þonne refers to the context in its preferred clause types; it marks surprise/disapproval in questions, and it establishes a causal link in imperatives and conditionals, indicating the consequence suggested by the context. The context is non-factual, however, and the potential result in the follow-up context is dependent on whether the situation in the context actually obtains.

Þa
Þa is quite robustly attested in hwaet exclamatives, and its meaning in that context is derived from its temporal meaning: (14) on the one hand shows the temporal sequencing, but it also draws attention to the new information that is the result of the message of the Magi: Þa is overwhelmingly preferred in preposed temporal adverbial clauses. (6), repeated here as (15), is a temporal correlative with the first þa a subordinating conjunction, the second a particle, and the third a resumptive adverb. Such correlatives have a temporal, discourse-sequencing, effect: the subclause is in initial position; the particle reinforces van Kemenade and Links: Discourse particles in early English Art. 3, page 11 of 23 the temporal sequencing of the whole event, while the resumptive adverb introduces the main clause presenting the new information: the subsequent building of the monastery. Links et al. (2017) show statistically that adding a particle to the temporal adverbial subclause promotes the use of the third þa, the resumptive adverb. This construction thus has a strong discourse-structuring function, promoted by the use of a particle.
(15) (Bede_4:4.272.28.2779) Þa he þa in þӕt ealond cwom, þa getimbrede he þӕr mynster, … then he prt in that island came, then built he here a-monastery 'On coming to that island, he erected a monastery there, …' We conclude that particle þa refers to a factual, temporally sequenced context in exclamatives and correlatives, drawing attention to the following events in these constructions.
Subject-initial clauses also feature a substantial use of þa (12,3 per 100 clauses). We give two examples here, which illustrate that its use in that context appears to be primarily temporal: was then frequently shown to-his disciples 'The Lord then frequently appeared to his disciples, …' The temporal character of þa in subject-initial clauses implies that, in the absence of any special illocution type, þa is simply a temporal adverb. This in turn reinforces our case that particle use of þa occurs in special illocution types.

Nu
Nu occurs largely in questions and imperatives. An example of a question with nu is (17), where nu expresses irritation, yielding a reading like 'is this all?', which in this example is reinforced by the exclamative hwaet la hwaet, and by the fact that two further parallel questions with nu are added in the following context.
(17) (Bo:3.9.19.111) [And if you so are willing to repent of your folly, I will begin to carry you immediately and bring you to heaven.] þa andsworode him þaet unrote Mod and cwaeð: Hwaet la hwaet, sint then answered him the sorrowful Mind and said: excl prt excl are þis nu þa god and þa edlean þe þu ealne weg gehete þam these prt the benefits and the rewards that you all way promised the monnum þe þe heorsumian woldan? people that you obey would 'Then the sorrowful Mind answered him and said: "Lo, are these now the benefits and the rewards which you always promised to those people who would obey you?' In imperatives, nu at first glance appears to have a purely temporal reading. Imperatives, however, while basically directive, often shade into requests, and the use of nu seems to contribute to this. A parallel Dutch request might be Ga nou mee, lit. come now along 'please come along' as in (5) above, which is a request rather than a directive. An Old English example of this is (18), which allows a reading 'at this moment', but it could also be interpreted as context-linked request, allowing a reading like 'since you can ask anything of God, please ask that I may know Greek'. This reading is reinforced by the fact that the latter seems a hypothetical situation rather than a factual one. A concrete indication that the meaning of nu need not always refer to 'this moment' is that, like in present-day Dutch (Vismans 1994: 64), it is occasionally accompanied by another temporal adverb, sometimes expressing future tense (we found today; tomorrow; for six days; as long as, and henceforth as in (19)): (19) (AEHom_2:53.276) [Afterwards, the Lord saw the man in the temple and said to him: "even now that you are healed,] Heald þe nu heononforð, þaet þu ne syngie, þy laes þe þe sum keep yourself prt henceforth that you not sin lest that that some þing wyrse gelimpe; thing worse happen 'Keep yourself from sinning henceforth, lest something worse should happen to you;' Nu in this context thus seems more like a friendly reinforcer of the imperative.
(20) (Bo:14.32.32.586) ["Know this for a truth, that no good thing harms the one who possesses it.] Hwaet, þu wast nu þaet ic þe ne leoge, truly you know prt that I you not lie 'Truly, you know that I do not lie to you' In correlatives, nu occurs with some frequency in gif-conditionals. Its meaning is very close to that of þonne in gif-conditionals: it introduces the potentially negative condition which, in case it is not met, has an ensuing result.
(21) (AELS_[Sebastian]:69.1252) [Then he said to the kinsfolk who were seducing the martyrs,] gif nu þas gebroðra be eowrum benum gebugað fram heora haelende to if prt these brothers by your prayers turn-back from their Saviour to eowrum haeðenscype, þonne beoð hi mid eow on sceortere blysse. your heathenism then be they with you in shorter bliss 'If now these brothers, at your prayers, turn back from their Saviour to your heathenism, then they shall be with you in a short bliss.' van Kemenade and Links: Discourse particles in early English Art. 3, page 13 of 23 Summing up, nu in questions marks surprise and/or irritation on the part of the speaker. Nu in imperatives is a reinforcer of the order, which may shade into a request. In correlatives, nu introduces the condition that needs to be met, very like þonne.
To conclude this section, we have shown that particles are typically responses to the context, and their meaning interacts with clause type. In questions, þonne expresses surprise/disapproval, whereas nu expresses irritation. These bleached pragmatic meanings support our hypothesis that they are particles rather than adverbs in contexts with special illocutions. Đonne and nu seem to be very similar in meaning and use in conditional correlatives, roughly meaning 'in case that' the condition is (not) fulfilled. In imperatives, the meaning of þonne and nu is distinct, þonne yielding a result reading, whereas nu reinforces the directive (adding impatience), or toning it down, shading into a request. Exclamatives and temporal correlatives prefer þa to affirm the exclamation, drawing attention to the new information following þa, whereas þonne is occasionally used in exclamatives, for instance when they embed a conditional.
The use of þa in subject-initial clauses is adverbial.

Clause types: Particles in main and subclauses
A final point we wish to address is the question of the distribution of particles in main clauses vs. subclauses. In Present-day Dutch and German, they are used a good deal more frequently in main clauses than in subclauses. We have seen that the particles þa and þonne occur in relatively high proportions in the subclause of correlatives, as can be seen in Table 2. The use of particles in correlative clauses is closely related to the special status of these clauses as paratactic: Links et al. (2017) show that they are paratactically adjoined to the main clause; Haegeman (2003; and Quirk et al. (1985) argue that this is still the case at least for what Haegeman calls peripheral adverbial and conditional clauses in present-day English (2012: 160-61). 10 Correlatives show multiple uses of þa and þonne, as discussed above in connection with the examples (6-8): they are introduced by þa/þonne as conjunctions, and when they feature a particle use of þa and þonne, this robustly triggers the use of þa and þonne as a resumptive adverb, as shown by Links et al. (2017) on statistical grounds. A plausible analysis of this state of affairs is that correlatives are more like main clauses in that they have a more extended functional articulation that supports the use of particles, even though they are clearly subclauses in terms of word order. Particle use may be further promoted by the fact that correlatives have a discourse-structuring function: the preposed adverbial or conditional subclause creates the context against which the main clause is to be interpreted. It is against this backdrop that the threefold use of þa an þonne, as conjunction, as particle, and as resumptive adverb, provides a coherent rationale to the discourse sequence.

The position of particles
The position of particles is rather fixed, as discussed before. This was established for subclauses by van Kemenade & Los (2006), van Kemenade, Milicev & Baayen (2008): Table 3 presents the results of van Kemenade & Los (2006), as corrected for excluding questions in 10 We do not at this point wish to hazard any claim that Haegeman's treatment of peripheral adverbial and conditional clauses in Present-day English carries over straightforwardly to Old English correlatives. We leave this matter for further research. Table 3: Order of subject and particle þa, þonne in embedded clauses in Old English.  van Kemenade & Milicev (2012). Table 4 shows that pronominal subjects near-categorically precede the particle in subclauses; nominal subjects are more mixed in their position and at some 36% precede the particle, with some 64% following the particle.

Subject-particle Particle-subject Total
It was also established that nominal subjects preceding the particle are most typically discourse-given subjects, such as demonstrative pronouns, definite NPs, and specific indefinite NPs (and optionally pronominal objects). They also include the pronoun man, which van Kemenade & Milicev (2012: 250), following up on van Bergen (2003) argue is an impersonal pronoun rather than an indefinite one. Nominal subjects following the particle are indefinite, generic or focused NP subjects ( van Kemenade & Milicev 2012). van Kemenade, Milicev and Baayen (2008) show that these generalisations are robustly significant statistically. We will consider in this section whether they extend to the other clause types examined here, in particular questions and exclamatives. Subclauses, including correlative subclauses, are covered in the results of van Kemenade and associates, in particular in van Kemenade, Milicev & Baayen (2008) and van Kemenade & Milicev (2012). Imperatives are not included because the subject is usually not realised there.
Let us first take a look at questions. Table 4 gives the numbers for questions with þa, þonne and nu. The objects included in the tables are pronominal objects, which also may occur, optionally, on the left of the particle.
The results in Table 4 are nicely in line with earlier findings and thus extend these to questions. We give examples with a nominal subject preceding þa in (22) (with specific reference to the context); with pronominal subject preceding nu in (23), and with a (generic) nominal subject following nu in (24).
(22) (AECHom_II,_3:23.150.555) Hwi com se halga gast ða on fyres hiwe ofer ðam apostolon, and why came the holy ghost prt in Fire's colour over the apostles, and ofer Criste on his fulluhte on culfran Gelicnysse? over Christ on his baptism in to-dove likeness 'Why did the Holy Ghost come in the colour of fire and in the likeness of a dove over the apostles and over Christ upon his baptism?' (23) (CP:42.308.6.2065) Wene ge nu ðaet ic aenigre leohtmodnesse bruce, …? Think you prt that I any levity possess 'Do you really think that I employ any levity, …?' (24) (ApT:3.10.32) Hwaet is nu mare ymbe þaet to sprecanne buton þaet cyningas what is prt more about that to speak except that kings aeghwano coman and ealdormen for ðam ungelifedlican wlite from-everywhere came and aldermen for the incredible beauty þaes maedenes, of-the maiden 'What more is there to say except that kings and princes from everywhere hurried there because of the girl's incredible beauty'  Table 5 gives the results for exclamatives introduced by hwaet. This presents a picture that is more mixed than that in other clause types.
The results for pronominal subjects are in line with the observations for subclauses and questions in Tables 3 and 4, but those for nominal subjects are not. Remarkably, the 251 instances of subjects following the particle include quite a few given subjects, beside predominantly names.
An example including a given subject is (25), where the subject se sceocca 'the devil' is mentioned in the immediate context. Nominal subjects are far more likely to follow þa in hwaet exclamatives (Table 5, 86,5%) than in questions (Table 4, 41,3%) and in subclauses (Table 3, 64%). We tentatively suggest that, since þa as a particle draws attention to the new information to come, this is more emphatically the case in a hwaet exclamative clause, which in a sense highlights an event by drawing attention to it. We leave this for further research. 11 We conclude that the position of particles is to a large extent consistent across clause types. The earlier finding for subclauses, viz. that the particle typically follows discoursegiven subjects and pronominal objects, and precedes discourse-new nominal subjects, is further confirmed for questions and, in part, for hwaet exclamatives. This in turn confirms the clause structure assumed in earlier work, as in (26) In main clauses questioned constituents and exclamative hwaet are in the first position, Spec, CP. The finite verb in questions and imperatives is in C, the position occupied by conjunctions in subclauses (cf. Den Besten 1983). FP represents the domain that hosts given information, and is part of the C-domain. The higher subject position in Spec, FP precedes the particle (þa, þonne, nu), whereas the lower subject position follows the particle.
There is a good deal of independent evidence for the structure in (26), which has been analysed as the key structure underlying the position of secondary negators (Haeberli 1998;van Kemenade 1999;2000;; the position of þa in Beowulf ( van Kemenade 2002); particles þa and þonne in subclauses (van Kemenade and associates, op. cit.); the mixed nature of finite verb placement in main clauses (van Kemenade 2012;van Kemenade & Westergaard 2012); the agreement asymmetry in the Northern Subject Rule (de Haas & van Kemenade 2015), and it is further confirmed for questions and (in part) for hwaet-exclamatives here.

Discussion
We have so far addressed three key properties of particles: 1) their discourse-linking character; 2) their (pragmatic) meaning and expressive function in relation to the clause type in which they occur; and 3) their form and syntactic position. We now attempt to explicitly unravel the interrelation between these three. We assume that the syntactic structure is an independent piece of grammar that does not in and of itself contain expressive or pragmatic information, or indeed information-structural notions such as topic and focus. We thus do not assume a clause structure along the lines of Rizzi (1997 and subsequent work), in which the full potential of information structural and pragmatic notions are universally encoded as syntactic structure, although we accept that this is in whole or in part correct for languages that have the morphosyntactic properties to support this, such as topic or focus particles (see e.g. Aboh 2004;. Syntax in our view reflects a grammaticalised version of robust structural generalisations available in the input during language acquisition. Particles such as the ones under discussion here may be part of that syntax, as adverbs grammaticalised to functional heads. Grammaticalised functional heads by their nature have undergone loss of lexical meaning; their meaning is bleached to match the context in which they are grammaticalised, relating to e.g. modality, or illocutionary force. These meanings are typically underspecified and context-dependent (see e.g. van Kemenade 1999, on the semantic effect of grammaticalisation). This is the point at which illocutionary force co-defines the contextual meaning of the particle: we have seen above, for instance, that there is a clear distinction between affirmative/assertive clauses (subject-initial clauses, exclamatives, temporal correlatives) and the use of þa, which then takes the shape of an assertive particle. On the other hand, clauses whose truth conditions are open (questions, imperative, correlative conditionals) feature þonne and nu. It is thus interesting to see how the source meaning of the adverb is traceable in its function in OE. This is in line with recurring patterns in grammaticalisation as observed in Bybee et al. (1994). We have also seen in the previous sections that particles encode expressive meanings, including the speaker's response to the context, such as surprise, disapproval, irritation, exclamation, assertion. We have so far seen evidence for speaker's response, though not for speaker attitude, which is why we hesitate to call the OE particles under discussion here modal particles.
Finally, we have shown in the previous section that particles are discourse-cohesive devices that respond to the common ground shared by speaker and audience. The common ground also includes discourse-given information (pronominal subjects and objects, discourse-given nominal subjects), which typically precedes the particle, which in turn demarcates the right edge of the discourse-given domain in the clause. Pronominal subjects (and optionally, pronominal objects) and discourse-given nominal subjects typically precede the particle and are part of the given domain. Discourse-new subjects follow it. We assume, following Biberauer & van Kemenade (2011), that discourse-given NPs are "big DPs" in the sense of Zeller (2008), featuring a pronominal head which is either absent in new-information DPs or represented by an expletive pronominal element (which typically facilitates new-information structures). We assume, pending further research, that this DP may be moved to a left-peripheral phase edge, triggered by an edge feature. Discourse management then entails that a discourse link is established between the given element and its (extra-clausal) antecedent. Particles thus form a class of grammatical elements that are at the interface of clause structure and discourse management.

Middle English and beyond
We now turn to a brief consideration of what happened to the particles under discussion here after the Old English period. The transition from Old to Middle English saw massive losses of the Old Germanic discourse-cohesive devices, as discussed in van Kemenade (2009) ;Los & van Kemenade (2018). Importantly, the earlier Germanic series of demonstrative pronouns and adverbs was largely lost. This included the se paradigm of demonstrative pronouns and determiners, which expressed definiteness and deixis and had provided a separate strategy for pronominal reference in Old English. It also affected the etymologically related set of time, place and manner adverbs (þa 'then', þonne 'then', þaer 'there', þus 'thus', swa 'so', swylc 'such') (Los & van Kemenade 2018). þa was lost altogether, according to Fischer (1992) because of its highly polysemous use, which van Kemenade & Los (2006); Links et al. (2017) identify as a conjunction, particle and resumptive adverb. þonne developed into 'then', but it lost its use as a conjunction (this was taken over by when), and as a resumptive adverb (except sporadically in conditionals). It survived in the language robustly as temporal adverb. Likewise, nu survives as a temporal adverb, especially in imperatives. We suggest that these losses were primarily due to the massive population changes resulting from the Scandinavian and French invasions, which had a strong phonological and (in the case of Scandinavian) also grammatical impact on the language. It is to be expected that discourse-cohesive devices such as particles, which are unstressed elements and have subtle pragmatic meanings, would be vulnerable under language contact, particularly in the case of substrate influence.
There are indications, however, that particle use was continued over the Middle English period and beyond, particularly in questions and imperatives, the two illocution types in main clauses in which particles were identified here for the Old English period. We  (27b)  The examples of imperatives and questions in PPCME (65 questions, 332 imperatives) very prominently feature the particle in the clause-internal position that we saw for Old English as well, that is in the position PRT in (26), in the case of questions clearly separating discourse-given from discourse-new subjects. In the later Middle English and early Modern periods, we also see instances of clause-final particles with a meaning that is very similar to that of þonne in Old and Middle English. Let us first consider some background to this.
Recent work by Haselow (2011) gives an interesting analysis of clause-final then in present-day spoken English, arguing for a discourse marker then, which is used to link the utterance it accompanies to a preceding utterance, much in the way discussed here for Old and Middle English. An example is given in (28)  According to Haselow, final then converts a proposition p1 expressed in a preceding discourse segment into a conditional protasis and thus marks the proposition p2 it accompanies as motivated by and directly linked to a preceding segment. This relation can be paraphrased as 'if p1, as is the case, p2 then' (2012: 159). Clearly, clause-final then is in a very different position from that of þonne in Old English and Middle English. Haselow (2012) makes a case that the origin of final then is in Middle English, and originates from the optional conjunct then in if…then… constructions in which the conditional protasis is not expressed in a subordinated if-clause, but is implied in a preceding utterance. Two examples that make the condition explicit are given in (29), from The Parsed Corpus of early Modern English (PPCEME, Kroch et al. 2004): (29) a. (jetaylor-e3-h) It is certain that the man or woman are in a state of weakness and folly then, when they can be troubled with a trifling accident; b. (deloney-e2-h) but if I finde meanes to make you a Lady, what wilt thou say then?
Haselow (2012) makes a case that final then originated over the ME period in the spoken language. It may be noted that it is hardly attested at all in PPCME2, which covers the period from 1150 to 1500, and sparsely in PPCEME, from 1500 to 1720. Haselow com- van Kemenade and Links: Discourse particles in early English Art. 3, page 19 of 23 piled a small corpus containing more spoken genres, and arrived at the following timeline for final then, as part of a total of different attested uses (2012: 161): (26) OE early ME late ME eModE final connector then 0 3 (3%) 13 (8%) 51 (34%) Space does not allow further consideration of this development, but we do wish to observe that there is thus a shift from clause-internal then as discussed here for OE and ME, to the use of final then from ME onward, maintaining the same discourse-linking and pragmatic function. This development presumably persisted, as the use of clause-internal then is quite marginal in present-day English. We hazard a speculation that this reflects a syntactic development: the rise of SVO word order over the late OE and ME period; the loss of V2 word order from late ME onward, and the rise of an extensive verbal periphrastic system including auxiliation of the modal verbs and the rise of do-support (e.g. Warner 1993; and references cited there) restricted the word order flexibility in the middle field, ousting particles from the syntactic niche that they had occupied for many centuries. It is striking, however, that they maintained their discourse-cohesive and pragmatic functions in clause-final position. Further refinement of this speculation must await further research. The development of clause-final then may be viewed as part of a larger development of what Lenker (2010) dubs the rise of final connectives over the late ME and later periods.

Conclusion
This article has made a detailed, corpus-based and explicit case for treating OE þa, þonne and nu as discourse particles functioning at the interface between syntax (in a fixed position as a grammaticalised functional head, information structure (marking the boundary between discourse-given and discourse-new information), pragmatics (expressing pragmatic meanings) and discourse management (linking the particle to the common ground between speaker and hearer). We have embedded this in a first sketch of an account of the relation between syntax on the one hand, and information structure, pragmatics and discourse on the other hand. Syntax is a bare structural template in which grammaticalised heads such as particles come to mark the boundary between discourse-given and discourse new information. In the syntax, discourse-given information is equipped with a pronominal marker which may be probed by an edge feature in the left periphery. The relation between this material and the extra-clausal context is established outside the syntax. This is true as well for the pragmatic meanings and the discourse management functions of particles: since the meaning of particles is underspecified, and contextdependent, pragmatic meanings arise that are sensitive to the illocutionary force of the clause. This approach, we think, is supported by the fact that particles may maintain their pragmatic and discourse-structuring properties over time, in spite of profound syntactic changes. We have discussed this here in connection with the development of then over the Middle English and later periods.