A unified account of the semantics of discourse particles Markus Egg Humboldt-Universitat¨ Berlin [email protected] Abstract This relation is described by the features KNOWN and CORRECTION in Thurmair (1989), The paper investigates discourse particles i.e., doch-utterances correct a previous utterance on the example of German doch, assigning by introducing old information. Karagjosova to them very specific semantic interpreta- (2004) regards doch-utterances act as reminders, tions that still cover a wide range of their which present old information to hearers. Ac- uses. cording to Konig¨ and Requardt (1997), doch- The analysis highlights the role of dis- utterances point out inconsistencies between old course particles in managing the common information and a new piece of information or ac- ground and crucially takes into account tion. that discourse particles can refer not only Such general descriptions of doch apply to cases to utterances they are a part of and to previ- like Karagjosova’s (1): B reminds A of Peter’s ously uttered utterances, but also to felicity illness, which seems inconsistent with A’s an- conditions of these utterances. nouncement and therefore can act as a correction of A: 1 Introduction (1)A: Peter wird auch mitkommen. B: Er ist This paper advocates very specific semantic in- doch krank. terpretations for discourse particles, concentrating ‘A: Peter will come along, too. B: But he is on German doch. There is a very wide range of ill.’ concrete usages of discourse particles in context While these general descriptions (excepting (which has motivated analysing them as polyse- Karagjosova (2004)) do not spell out in detail the mous, e.g., in Helbig (1988)). way in which doch contributes to the meaning of Assigning them a uniform semantic interpreta- larger discourses, they can capture a wide range of tion seems to be subject to two conflicting require- uses of these particles. ments: There remain a number of problematic cases, including discourse-initial uses of doch-utterances • the interpretation must be sufficiently specific like Konig¨ and Requardt’s (2), which functions as to allow deriving the interpretation of con- an opening line in a conversation, it neither cor- crete uses rects nor reminds the hearer, nor is there an incon- • it must be sufficiently general to cover a wide sistency between the utterance and the context: range of concrete uses (2) Sie sind doch Paul Meier. ‘You must be Paul Meier.’ So far, research on the interpretation of doch focusses on the second requirement (e.g., Thur- The proposed analysis of the particle doch is mair (1989), Konig¨ (1997), Karagjosova (2004), sufficiently general to account for a wide range of or Konig¨ and Requardt (1997)). uses yet being specific enough to specify the se- The meaning of doch emerges as a two-place mantic construction for discourses that comprise relation between the utterance doch is a part of and doch. a previous utterance to which the doch-utterance is I follow much previous work in developing my a reaction. analysis on the basis of simple examples like (1), Proceedings of SIGDIAL 2010: the 11th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 132–138, The University of Tokyo, September 24-25, 2010. c 2010 Association for Computational Linguistics 132 and then extending it to cases like (2). Most ex- problems of defining the semantics of the particles amples consist of two utterances, the second utter- in the literature, which is characteristically based ance comprises a discourse particle and is a reac- on the meanings of a- and p-utterance. tion to the first one. In the remainder of this pa- My claim is that there is a link between a- and p- per, these two utterances are called ‘involved ut- proposition and a- and p-utterance, respectively, in terances’. that the propositions can either be the meanings of In (1), the (propositional) semantic arguments the utterances or emerge through the felicity con- of the particle are the meanings of these two utter- ditions of the utterances. ances. But the semantic arguments of a discourse E.g., in (3) the doch-proposition reminds A of particle may differ from the meanings of the in- the fact that the first preparatory condition for a volved utterances, as illustrated by (3) (from Thur- question (that the speaker does not know the an- mair 1989): swer) does not hold, since A (as the one who gave the book to B) should know since when the book (3)A: Seit wann hast du denn den has been in B’s possession. ,,Zauberberg“? B: Den hast du mir doch vor The plan of the paper is to introduce back- zwei Jahren geschenkt. ground assumptions on discourse particles in sec- A: ‘Since when have you owned the tion 2, to apply the proposed approach to the (un- ‘Zauberberg’? B: But you gave it to me two stressed) particle doch in section 3, and to con- years ago.’ clude with an outlook on further research. B reacts to the implicit statement that A does not 2 Formal background know the answer to his question. This statement is an argument of doch in (3), even though it is not This paper follows much previous work in assum- the meaning of A’s utterance. This shows that the ing that discourse particles refer to the common semantic arguments of discourse particles must be ground (CG), e.g., Konig¨ (1997), Karagjosova distinguished from the meanings of their involved (2004), or Zimmermann (2009). utterances. Common ground and the interlocutors’ individ- Utterances with a discourse particle and pre- ual backgrounds are modelled as common or in- ceding utterances to which they react are called dividual belief (Stalnaker, 2002). Individual be- ‘p(article)-utterances’ (or ‘doch-utterances’) and lief is equated with the set of propositions that are ‘a(ntecedent)-utterances’. They are distinguished true in all possible worlds compatible with the in- from the semantic arguments of the particle, dividual’s beliefs; common belief, with the set of which are referred to as ‘p-proposition’ and ‘a- propositions believed by all members of the re- proposition’, respectively. spective group of believers. This is not just a question of nomenclature but Stalnaker notes that this is an idealisation in that reflects a fundamentally different view on the role the CG might comprise propositions not shared by of discourse particles. Instead of indicating the re- the background of every member of the group. But lation between two already identified propositions, this idealisation is not a problem for the analysis the meaning of the particle applied to its first ar- presented in this paper. gument (very often but not always the interpreta- Reasoning on CG and individual backgrounds tion of the p-utterance) determines the range of po- often uses defeasible deduction (Asher and Las- tential a-propositions in the context of utterance. carides, 2003). I.e., from statements of the form From this range, the hearer selects the appropriate ‘p defeasibly entails q’(p > q) together with p proposition. one can defeasibly deduce q. This resembles the intuition of Konig¨ and This defeasible Modus Ponens applies if ¬q Requardt (1997) that discourse particles are does not hold and ¬q cannot be deduced simulta- ‘metapragmatic instructions’ which tell hearers neously (Asher and Lascarides, 2003). Defeasible how to deal with the p-utterance in a communica- deducability of p from a set of propositions C is tive situation. written as ‘C |∼ p’. Consequently, a- and p-utterances do not deter- Reference to the common ground makes the se- mine the semantic arguments for all uses of dis- mantics of discourse particles context-dependent, course particles, which might account for some because the CG is relative to the interlocutor(s) of 133 a- and p-utterances. This shows up in the shifting Formally, doch states that the common ground effects observed in Doring¨ (2010). Consider e.g. C defeasibly entails q and the fact that q defeasi- what happens if one embeds (1) in a quotation like bly entails ¬p (which by defeasible Modus Ponens in (4): would allow one to infer ¬p, if the conditions for defeasible Modus Ponens are met): (4) A sagte, Peter komme auch mit. B entgegnete, er sei doch krank. (7) [[doch]](q)(p) iff C |∼ q ∧ C |∼ q > ¬p ‘A said Peter would come along, too. B ¨ retorted that he was ill.’ This analysis differs from the one of Konig (1997), who assumes that doch q points out a con- The shift in (4) arises because doch presents a tradiction in the CG, in that p is incompatible with proposition (here, that Peter is ill) as part of the a consequence of q. In contrast, I regard this common ground, and the relevant CG is calcu- incompatibility as a default only. The status of lated with respect to A and B, not with respect to q as derivable from the CG is also expressed in the interlocutors of (4). I.e., (4) does not express Karagjosova (2004) claim that doch introduces q that Peter’s illness is in the common ground of the as a reminder and in Thurmair’s (1989) feature speaker and hearer of (4). KNOWN. In (5) and (6), p and q are the semantics of the a- 3 The analysis and the doch-utterance. Being ill is a potential im- pediment for going out, so, by pointing out Peter’s The proposed approach to discourse particle is illness in (5), B expresses surprise or disbelief at now applied to doch, which introduces a notion A’s announcement but does not necessarily correct of tension between the a- and the p-proposition.