What can you do? Imperative mood in Semantic Theory Rosja Mastop Abstract: A common assumption in linguistic semantics is, that sentences of all types share a common, truth conditional core. For non-declaratives this means that we must distinguish between a mood-operator and a propositional (truth-conditional) content. So, for instance, an imperative sentence is treated as commanding that some proposition be made true. In this dissertation it is argued that such an analysis is not reasonable. First of all, it makes declarative sentences the prime carriers of meaning. This is a way to maintain compositionality (word meaning is determined by the contribution to truth conditions alone), but at the cost of creating an artificial divide between a semantics that accounts for the meaning-relations between declaratives (such as consistency, entailment, etc.) and a pragmatics that has to account for the meaning-relations between non-declaratives. Second of all, the mood/content analysis of imperatives fails to explain some of the most striking features of such sentences: (i) the properties of their (optional) subjects and (ii) their temporal and aspectual orientation. Ad (i), by default imperatives do not have a subject, but when they do, it functions as a means to pointing out who is to comply with the given instructions. Thus the quantified subjects in the examples below are not part of the truth conditional contents. (1) Somebody get a doctor. (is not: see to it that somebody gets a doctor.) (2) Nobody make a move. (is not: see to it that nobody makes a move.) On the mood/content account, we cannot give a semantic analysis of the meanings of these sentences. This shortcoming is all the more striking when it is noted that many languages have first and third person imperatives. (One example, the Dutch ``laten wij/zij VP-inf'', is discussed in some detail.) Ad (ii), imperatives are future oriented. They do not contain stative VPs and, in general, they do not have a past tense form. However, Dutch has a ``plusquamperfect imperative'', expressing a reproach (a counterfactual instigation) and always directed at the addressee. (3) Had (*jij/*hij) dan ook gewoon de trein genomen! (4) Was (*jij/*hij) toch lekker gaan fietsen! Again, these phenomena cannot be dealt with if the imperative is merely a mood-operator with pragmatic rules of use. In contrast to the mood/content analysis, the dissertation defends a semantic account that is characterized by the following traits. (1) It is an update semantics. Declarative sentences are viewed as cues to change one's state of information and imperative sentences are viewed as cues to change one's to do list. This allows us to adopt a one-level approach in which imperative meaning is not `derivative' or pragmatic. Furthermore, by using a constructive update semantic model, we can give a straightforward solution to the problems of disjunctive commands and permissions, much discussed in the literature. (2) It treats those ``cues'' as perspective-dependent. Imperative sentences are interpreted from a subjective (second person) perspective, which explains why in general they do not need an overt subject. The quantified and non-second person subjects are analyzed as perspective *shifters*, which explains why in English and Dutch they are typically used when it is not clear from the context who are the ones that are being instructed. The speech time is the default temporal perspective, but the plusquamperfective operator is analyzed as a shifting this perspective into the past, resulting in an irrealis context of interpretation. This explains why it is meaningful to issue a `command' from a past-tense point of view. Taken together, these two traits---dynamic and perspectival---make for a semantic approach in which declaratives do no longer form a privileged sentence type. Instead, the variety of sentence types can be accounted for by characterizing the different aspects of a person's cognitive state and the different ways they can be altered.