Aristotle's kinesis/energeia-test and the semantics of the Greek perfect (1).
The purpose of this article is twofold. First, I aim to arrive at a better understanding of Aristotle's distinction between kineseis and energeiai in the Metaphysics 1048b. I argue that in order to understand Aristotle's test, we must understand the workings of the Greek perfect better. The main part of the article is then concerned with the semantics of the Greek perfect from Homer to classical times, discussing Gero and von Stechow's recent contribution (2002). Finally, an interpretation of Aristotle's test in the light of this discussion is proposed.
Linguists presenting the modern taxonomy of aktionsarten (2) often refer to Aristotle's Metaphysics 1048b, lines 18-35, as the earliest discussion of this problem. In this passage, Aristotle divides actions--praxeis--in two groups, which he calls kineseis and energeiai. Although there seems to be agreement that Aristotle's distinction is related to modern distinctions of aktionsarten--which go back to the works of Vendler (1957) (reprinted in Vendler  from which I henceforth cite) and Kenny (1963)--there is uncertainty as to the exact nature of the relation. Kenny (1963) was explicitly influenced by Aristotle as he proposed his classification, and he affirms that Aristotle's energeiai are his (and our, as the modern classification is based on those of Kenny and Vendler) activities and states. In Verkuyl (1993: 50), however, we read that "Kenny is much more faithful than Vendler to Aristotle's division in Metaphysics IX 1048, 18-35, who distinguishes Actualities (States) from Incomplete and Complete Movements." And Graham (1980) has argued that energeiai are our states, whereas kineseis are non-states, and this opinion is adopted, for example, in Smith (1997: 17). So there is still need to clear up the confusion, and this is only possible, I shall argue, through a better understanding of the Greek perfect, which plays a crucial role in Aristotle's test.
However, before we go on to a closer scrutiny of Aristotle's test, some brief philological comments should be made. The passage where Aristotle discusses the kinesis/energeia distinction does in some ways look suspicious: the sentence following immediately upon our lines is a recapitulation of what precedes the discussion of the kinesis/energeia distinction. Our lines are also omitted in some manuscripts. This led Werner Jaeger (1957:184) to the conclusion that the passage is a later addition to the text, although, as he says, it was probably written by Aristotle himself. The kinesis/energeia distinction is referred to in other works of Aristotle, and there can be no doubt that "it contains sound Aristotelian doctrine and terminology" (Ross 1924: 253). We will therefore follow current usage and treat the passage as genuine. But it does not really fit in its context, and this is no doubt part of the reason for the difficulties in interpreting it. With regard to the text itself, it is corrupt at some points, but it was vastly improved by Bonitz, whose emendations seem to have carried conviction. At any rate, the textual questions do not have implications for the core of the matter.
2. Aristotle's and Kenny's tests
We will now compare the tests of Aristotle and Kenny, starting by citing Ross' (1928) English translation of Aristotle (I have added the Greek terms in square brackets):
Since of the actions which have a limit none is an end but all are relative to the end, e.g. the removing of fat, or fatremoval, and the bodily parts themselves when one is making them thin are in movement in this way (i.e. without being already that at which the movement aims), this is not an action or at least not a complete one (for it is not an end); but that movement in which the end is present is an action. E.g. at the same time we are seeing and have seen, are understanding and have understood, are thinking and have thought (while it is not true that at the same time we are learning and have learnt, or are being cured and have been cured). At the same time we are living well and have lived well, and are happy and have been happy. If not, the process would have had sometime to cease, as the process of making thin ceases: but, as things are, it does not cease; we are living and have lived. Of these processes, then, we must call the one set movements [kineseis] and the other actualities [energeiai]. For every movement is incomplete --making thin, learning, walking, building; these are movements, and incomplete at that. For it is not true that at the same time a thing is walking and has walked, or is building and has built, or is coming to be and has come to be, or is being moved and has been moved, but what is being moved is different from what has been moved, and what is moving from what has moved. But it is the same thing that at the same time has seen and is seeing, or is thinking and has thought. The latter sort of process, then, I call an actuality [energeia], and the former a movement [kinesis] (Ross 1928: 1048b, lines 18-35)
Aristotle here seems to propose a linguistic test that distinguishes between kineseis and energeiai. One part of it, the kinesis test, reads in Greek: (3)
(1) oud' hugiazetai kai hugiastai not [cure.sub.3.SG.MED.PRES.] and [cure.sub.3.SG.MED.PFCT.]
Ross translates: "For it is not true that at the same time we (...) are being cured and have been cured." This seems to say that for kineseis, the present tense verb entails the negation of the perfect tense verb. (4)
Similarly, from his wording
(2) eu zei kai eu ezeken well [live.sub.3.SG.PRES.] and well [live.sub.3.SG.PFCT.] hama at the same time 'At the same time we are living well and have lived well'
it seems clear that he is saying that for energeiai, the present tense verb entails the perfect tense verb.
It is obvious that Aristotle is making an ontological distinction here, and not a linguistic one. If nothing else, this is clear from the fact that he has probably made up at least one of the linguistic forms occurring in his text, namely ezeken, which Ross translates "has lived." Morphologically, ezeken is a regular perfect of the verb zo. However, the form does not occur before Aristotle and only very rarely after, the normal perfect being the synchronically suppletive (but diachronically related) bebioke. bebioke, however, often has the meaning 'having lived to the end.' Thus, Aristotle's statement "we are living, and have lived" would simply be false if he had used the normal Greek perfect of the verb 'live.' Therefore, we may surmise, he made up a new perfect. Now this, of course, means that Aristotle is speaking about living and not about the Greek verb zo.
But even though Aristotle's distinction is essentially a philosophical and ontological one, it has proven interesting for the modern linguistic classification of verbs, whose beginnings are, significantly, to be found in the works of the philosophers Vendler and Kenny. Scholars have, however, not always been aware of the differences between the Greek and the English perfect. We can see this quite clearly in Ross' translation. One of the examples for the kinesis-test reads: "For it is not true that at the same time a thing is walking and has walked." This is false: obviously, the same thing at the same time can be walking and have walked, even if we allow that both VPs refer to the same event. Of course, we should not ascribe such a fallacy to Aristotle. Rather, something went wrong in the translation of the Greek perfect into English. Now Ross was clearly motivated by the desire for a literal translation and thus translated the Greek perfect using an English one. But 'has walked' is naturally understood as an experiential perfect (or equivalently as an existential perfect 'there has been walking,' see Comrie 1976: 58ff.), and this is a reading that the Greek perfect cannot have, as we shall see. But first, we turn to Kenny's test.
Vendler's (1967) and Kenny's (1963) classification of aktionsarten are founded on Aristotle's distinctions, and in Kenny's case, it is explicitly so. Kenny first distinguished states from events; he then went on to divide events into activities and performances. In doing this, he proposed a test to determine whether a given verb is an activity verb or not: for activity verbs, he claimed, a sentence of the form 'A is [??]-ing' entails 'A has [??]-ed.' For example, 'John is running' entails 'John has run.' For performance verbs, however, the situation is the reverse: 'A is [??]-ing' entails 'A has not [??]-ed.' For instance, if a man is building a house, he has not yet built it; if John is deciding whether to join the army, he has not yet decided to. It is immediately clear that Kenny's test bears a resemblance to Aristotle's, and Kenny claims that the activities brought out by the test are part of Aristotle's energeiai (which also, according to Kenny (1963: 173, fn.2), include states). Aristotle's kineseis, Kenny further claims, are his performances.
It has been argued that Kenny's tests are not contradictory and that they therefore do not determine two jointly exhaustive classes (Potts 1965: 66-67). The counterpart to the activity test 'A is [??]-ing entails A has [??]-ed,' should not be that for performances 'A is [??]-ing entails A has not [??]-ed,' but rather the weaker formulation 'A is [??]-ing does not entail A has not [??]-ed.' As Graham says, Kenny's motivation for setting up the tests as he did was clearly his Aristotelian model. He goes on to defend Kenny:
For performances, 'A is [??]-ing' does not entail 'A has [??]-ed'; but furthermore, because performances are complete only over the whole interval (N.E. 1174a 27-29), in all cases 'A is [??]-ing' entails 'A has not [??]-ed' for kineseis. Thus, for metaphysical reasons, the classes picked out by the criteria are jointly exhaustive (Graham 1980: 118, fn. 2).
Graham's use of both Aristotelian and Kennyan terms is confusing, but the point made seems valid enough: if [??] is a performance and A is [??]-ing NOW, it follows from the semantics of the progressive that A has not yet completed [??]-ing, although he may very well do so in the future. This also works in the past: 'A was [??]-ing' entails 'A had not [??]-ed,' because the Reichenbachian reference time of the pluperfect is naturally understood to be identical to the reference time and event time of the past progressive; and it once again follows from the semantics of the progressive that A had not yet completed [??]-ing at that time. However, the test breaks down if we use the simple past to give 'A was [??]-ing' entails 'A did not [??],' because the simple past introduces a new reference time identical with its event time, which may very well be later than that of the past progressive. Hence, the entailment 'John was writing a letter entails John did not write a letter' does not go through--perhaps John finished the letter, we do not know. The weaker formulation 'A was [??]-ing does not entail A [??]-ed' is clearly correct, and it often appears as a test for performances in the literature.
It seems then that it makes no difference whether we use Kenny's formulation or the weaker one, as long as we study the relationship between the present progressive and the perfect, or between the past progressive and the pluperfect (i.e. in cases where the reference time of the second sentence is hooked up to the event time of the first)--but the weaker formulation can also be applied to the relationship between the past progressive and the simple past. However, this is not necessary for our present purposes, so we will stick to Kenny's original formulation, which is close to Aristotle's and therefore particularly useful in comparing Aristotle's classification with the modern ones.
There seem to be no major differences between Kenny's and Vendler's tests. In the following, we shall use Kenny's since it is explicitly modeled on Aristotle's. However, we shall use much of Vendler's terminology, since it is more widespread in the modern literature. Since there are many terminological confusions in this field, it may be useful to state our usage here: in Vendler's system, Kenny's performances are divided into accomplishments and achievements, where accomplishments are durative ('write a letter') and achievements punctual ('win the race'). We will refer to them as "accomplishments and achievements" (as the difference is mostly not crucial for our present purposes), or as "events falling under a telic predicate." This last formulation points to two other important issues, one terminological and the other ontological: by "event" we shall mean loosely "something that happens" and not--as is often seen--a cover term for accomplishments and achievements (i.e. a synonym for Kenny's performances). Further, and this will turn out to be crucial, we shall follow Krifka (1992: 34-35) and others in understanding telicity as a property of predicates and not of events. We will return to these issues, which are important for an understanding of Aristotle's test, in the last section.
Vendler and Kenny both seem to apply their tests to verbs. As Verkuyl (1993), discussing Vendler, notes
His inclination to stay at the lexical level enforces the idea that what he really did was to propose ontological categories: if knowledge of the world and knowledge of a language tie up intimately at some place, they do that at the lexical level and not so much at the structural level (Verkuyl 1993: 33).
Vendler and Kenny did what we would expect in an Aristotelian tradition: as we have said, Aristotle's distinction is essentially ontological and not linguistic. He does use a linguistic test, but then it is not surprising to find an ancient Greek holding that the Greek language reflects reality in a rather direct way.
However, modern linguists have stressed that the categorization of situation types must be applied to VPs and not to verbs, because the nature of the object may influence the situation type: draw circles is an activity, whereas draw three circles is an accomplishment. This can be seen from Kenny's test: John is drawing circles entails John has drawn circles (at least after some time), but John is drawing three circles entails that he has not yet done so. In the same way, walk in the park is an activity, whereas walk to the store is an accomplishment.
This phenomenon has been used to "save" Aristotle's test. As we remarked, Ross' translation says, "it is not true that at the same time a thing is walking and has walked." We can make this sentence true in English by supplying a terminus: "it is not true that at the same time a thing is walking from Oxford to Reading and has walked from Oxford to Reading." In doing this, we have changed the activity 'to walk' into a goal-oriented, durative action, that is, an accomplishment. It can be argued that Aristotle implies the termini, and such a conception clearly lies behind Ross' translation. But if it is possible, we should prefer to interpret the text as it stands, without making any suppositions as to implicit objects or termini. The reason for this is that if we start supplying underlying objects and prepositional phrases, we soon run into problems. Where are we allowed to understand such complements and where are we not? For example, the object of one's enjoyment could limit the enjoying in just the same way as the finished house limits the house building. Following this line of thought, Ackrill (1965) argued that one cannot be said 'to have enjoyed a symphony' before one has listened through it. That is, while I am enjoying the symphony, I cannot claim that I have yet enjoyed it. This object NP would make enjoying a kinesis, in plain contradiction with what Aristotle tells us in the Nicomachean Ethics. And if we are allowed to introduce termini in some cases, why not here? Ackrill concluded that Aristotle's grammatical test is flawed. However, his arguments presuppose that 'enjoy' is an adequate translation of hedesthai, which is not self-evident. If hedesthai is a stative verb, an object will not make the VP telic. But anyway, we can avoid such problems by not supplying objects or prepositional phrases for Aristotle's verbs. After all, they are not in the text.
We summarize Kenny's and Aristotle's tests in the following:
Kenny's test: Performance verbs: A is [??]-ing entails A has not [??]-ed. e.g. John is building his house entails John has not built his house. PRES.PROG. entails not-PFCT. Activity verbs: A is [??]-ing entails A has [??]-ed. e.g. Mary is walking entails Mary has walked. PRES.PROG. entails PFCT. Aristotle's test: Kineseis: oud' hugiazetai kai hugiastai For it is not true that at the same time we are being cured and have been cured. PRES. entails not-PFCT. Energeiai: eu zei kai eu ezeken hama At the same time we are living well and have lived well. PRES. entails PFCT.
One difference between Kenny and Aristotle springs to our eyes at once. Kenny's test uses the present progressive, and so it cannot be applied to state predications. Greek, of course, has no progressive and we cannot therefore exclude that Aristotle's test applies to states. It will be important to have this in mind, although I shall argue that the main difference between the two tests lies in the semantics of the perfect in English and Greek.
The semantics of the Greek perfect was quite recently examined by Gero and von Stechow (2002), from whom we have drawn many insights. However, our need for comparing the English and the Greek perfect has led us to focus on other aspects, inspired by Parsons' (1990) account of the English perfect on the basis of his notion of resultant state. Especially with regard to the Greek perfect in classical times, we have deviated from Gero and von Stechow.
3. The early Greek perfect--general considerations
There seems to be general agreement that the Greek perfect expresses the present existence of a state resulting from a past event. Unfortunately, this is not very precise. We would like to know exactly what kind of state is meant. We will discuss this question later; for the moment, we would only note that such resultant states are not necessarily to be identified with states in the Kenny-Vendler taxonomy. This must be kept in mind during our discussion of the other question: does the Greek perfect present a state resulting from some past event or not?
In some cases it is clear that the state results from a past event. The perfect tethneke 'he is dead' is clearly a result of a past event 'he died,' which can be expressed by the aorist (perfective past) ethane. This event can again be seen as including an activity subevent 'he was dying' expressed by the imperfect ethneiske. The verbs which have such a relationship between imperfect--aorist--perfect were called action-event-state verbs by Lyons (1967: 117ff.). Lyon's insight is important, but from a terminological point of view, this use of the term "event" does not fit into our scheme, and in the following we shall refer to the coming about of a new state through an event as the culmination of that event. In this, we are following Parsons' (1990) terminology, but as we shall see in the last section, this is not entirely unproblematic.
There are some verbs for which this relationship never holds. Such verbs have a so-called intensive or abnormal perfect. We can give as an example the verb gethein 'be happy.' The present gethei means 'he is happy,' the aorist egethese 'he became happy,' and the perfect gegethe means 'he is happy.' We see the reasons for the somewhat unhelpful labels "abnormal" or "intensive" perfect. The perfect tense is abnormal in that it seems to have the same meaning as the present. Some scholars have felt that this cannot be the ease, and so they have postulated that the perfect is somehow more intensive than the present. But this cannot be safely read out of the texts. As we do not have ancient Greek informants, we shall probably never know the exact nature of the semantic difference between the perfect and the present in such cases, although it seems probable that there is one.
Our problem here, however, is whether the perfect gegethe 'he is happy' can be seen as the result of a former action. Clearly, it would be possible to say that it means 'I have become happy and now I am happy.' Ruiperez (1982) denies this, however, on the grounds that gethein means 'be happy' and not 'become happy.' To this, one would reply that the aorist egethese does mean 'become happy.' Ruiperez denies such an analysis, but to my mind his arguments are not convincing. They are mainly morphological: he says that many verbs with an abnormal perfect do not have an aorist, and that the morphological markers of the aorist and the perfect are in any case quite different. But these morphological arguments cannot disprove that there is a semantic relationship between the aorist and the perfect. In fact, I will argue that there is a fundamental relationship between the perfect and the perfective (aorist) aspect. The obvious advantage of such an analysis is that it makes it possible to give a unified account of the semantics of the Greek perfect, and this is what we would expect in such a highly marked form. The "abnormal" and the "normal" perfect turn out to be derived semantically in the same way; for some verbs, however, this gives rise to a near-synonomy of the present and the perfect forms. These doublets could be pragmatically exploited to express differences whose exact nature will probably remain obscure to us.
A priori, then, it seems plausible to see a relationship between the semantics of the aorist and of the perfect. The texts also seem to support such a semantic relationship between the aorist and the perfect. This can be seen readily with the verb thneskein: in principle, the imperfective could be used of a dying person who nevertheless survived. If the aorist is used, however, the person died irrevocably. And the perfect, of course, refers to the state resulting from this culminated event expressed by the aorist. It does not mean 'having had a near-death experience.'
Another example of this semantic relationship can be found in the verb bouleuo in its standard sense, that is, excluding the special meaning 'to be a member of the boule (council).' According to the Liddell and Scott (1996) lexicon, it means "'take counsel,' 'deliberate,' in past tenses 'determine' or 'resolve after deliberation'." But saying that the meaning 'determine' or 'resolve after deliberation' belongs to the past tenses is plainly wrong and shows a disregard for aspectual issues in the Greek lexicographical tradition. Naturally, it is the aorist (the perfective past), and--in most cases at least--not the imperfect, that has the meaning 'determine,' 'resolve after deliberation' as is noted in the Diccionario Griego-Espanol (Rodriguez Adrados 1980). In other words, the aorist presents an event that culminated. But what about the perfect?
We have a particularly nice example of its use in Plato's Crito 46A. Crito tells Socrates about the plan for his escape and concludes:
(3) oude bouleuesthai eti hora alla not [deliberate.sub.PRES.INF.] still time but bebouleusthai [deliberate.sub.PFCT.INF.]
The meaning cannot be: this is no longer the time for deliberating, but for having deliberated. The perfect bebouleusthai must have the meaning 'having made a decision,' or more precisely 'be in the state of having made a decision.' And this seems to be the meaning of bebouleusthai in all its classical occurrences: it refers to the consequent state of the culminated event ebouleusa and not the past activity ebouleuon 'was discussing.'
We therefore espouse the view that the perfect denotes a present state resulting from a former event that can be expressed by the VP in the aorist. The perfect, therefore, has a double reference: a present state and a past event that culminated. But, as we would expect in such situations, pragmatic factors can put emphasis on the state or on the event. This is a question that we will not pursue further here. Suffice it to say that in early Greek, the notion of state seems to be much more important than the preceding event.
Furthermore, as Wackernagel (1904) recognized in his classical study, the perfect is often intransitive in Homer: this is so because, in most cases, it refers to a state of the grammatical subject, resulting from some former event. For example, the present histemi is transitive and means 'I make to stand,' whereas the perfect hestemi intransitive and has the meaning 'I stand,' that is 'I have made myself to stand and stand now,' although in this case there seems to be almost no emphasis on the past action. However, the Greek perfect is not necessarily intransitive: many verbs can take an object and nevertheless refer to an event which gives rise to a state of the grammatical subject. Despite claims to the contrary in the post-Wackernagelian tradition, this was obviously so even in the oldest reconstructable stage of Indo-European: the obviously archaic formation *woyd-[H.sub.2]e (perfect from the root *weyd- 'to see') means 'I know' and is of course transitive, but the state expressed is nevertheless a state of the subject. Another example attested in Greek is the verb 'to learn,' Greek manthano. The perfect mematheka is of course transitive, but it nevertheless expresses a state of the subject, namely the state of having learned something.
Furthermore, it is essential to the truth-conditions of the Greek perfect that the state holds at utterance time. This is quite different from the English perfect. The notion of resultant state is relevant to the English perfect also (at least in theories other than the extended now-theory, to which we will return later), in that it asserts of the subject that it has what Carlota Smith (1997: 107ff.) calls the participant property. In Smith's analysis, the sentence 'Elaine has danced with Bill' asserts of Elaine that she has the property of having danced with Bill. This is in some way a state resulting from her dancing with Bill. There is therefore a felicity condition on the use of the perfect in English. The classical example is 'Einstein has lived in Princeton,' which is odd when uttered after Einstein's death; he can no longer be ascribed the participant property. But this felicity condition is utterly different from the truth conditions imposed by the Greek perfect. Consider the corresponding Greek sentence:
(4) Aristoteles Athenas [Aristotle.sub.NOM.SG.] [Athen.sub.ACC.PL.] oikeke [inhabit.sub.PFCT.3.SG.]
This sentence is simply false if uttered at a time when Aristotle no longer lived in Athens. The difference is, of course, that the notion of state inherent in the English perfect in its experiential reading is different from the state to which a Greek perfect refers. This difference seems in fact to be crucial, and we will return to it below. But we can at once note that this use of the perfect is explainable in terms of the relation between the perfect and the aorist: the aorist from this verb, oikesa, does in fact mean 'to settle.' Clearly, we can paraphrase the meaning of (4) as 'Aristotle has settled in Athens and the consequent state of this past event still holds.' This is not different from the meaning of the present, except that there is a reference to the beginning of the state. The sentence (4) was made up to parallel the English example, but the fact that the perfect of oikeo refers to the beginning of the state can be seen in the texts. In Sophocles' play Electra (1101), Orestes asks where Aigisthus lives using the perfect of this verb. Anyone who knows the play will see that it is central to Orestes' concerns that he refers to the beginning of the state; Orestes is seeking Aigisthus because Aigisthus has taken his father's place and now lives in his palace. Here, the reference to the beginning of the event is clear, but the emphasis is naturally on the present state.
Given the fact that the early Greek perfect is essentially intransitive and that the notion of state is all-important, it is tempting to conclude that it has less in common with the category perfect than with the category which Osten Dahl (1985: 133ff.), following Nedjalkov and others, labels "resultative." Nedjalkov in fact uses two terms which seem relevant for the understanding of the Greek perfect, namely stative and resultative. The difference is defined as follows (Nedjalkov 1988: 6): "the stative expresses a state of a thing without any implication of its origin, while the resultative expresses both a state and the preceding action it has resulted from." It is notable that in many languages there is wavering between the stative and the resultative meaning: that is, the same form can be used to denote a state, with or without reference to the event that brought it about. This is reminiscent of the situation in Greek: the reference to a past event can often be rather weak or even nonexistent.
Both the stative and the resultative, then, refer to a state, and in this they differ from the perfect, which primarily refers to a past event. For the time being, we uphold that the notion of state is relevant to the use of the perfect also--cf. the participant property--but it is much weaker than the state to which statives and resultatives refer. This can be seen from the test proposed by Dahl (1985: 134): an adverb like 'still' like in 'I am still reading Proust' implies a state-of-affairs and indicates the lack of a difference between the state of affairs expressed in the proposition, and the actual one. This analysis will explain why 'still' is incompatible with perfects (which primarily refer to a past event) like in the ungrammatical '*I have still been frightened,' but good with resultatives, like the Greek perfect in the Iliad 21.206:
(5) hoi r'eti par potamon pephobeato they still along-the-river [put-to-flight.sub.3.PL.MED.PFCT.] dineenta whirling 'They were still fleeing along the whirling river.'
The perfect expresses the resultant state of having been put to flight, and the adverb eti 'still' can access this state.
Dahl (1985: 134ff.) further claims that resultatives are characterized by the fact that they are used of results in a narrow sense, whereas the perfect of result allows for a less strict conception of resulting state--which is why its use shades off into the experiential perfect. It follows from this that resultative constructions can only be formed from verbs whose semantics involves change of some kind. In general, resultative constructions tend to be highly lexically restricted. There also seems to be a high correlation between passive voice and resultative constructions.
Now this apparently fits very well with the Homeric perfect. We have already noted that Homer's perfect is generally intransitive, and we may add that fewer verbs have attested perfects in Homer than in later texts. Its use is restricted to certain types of verbs: besides a peculiar use of the perfect with noise verbs, we mostly find verbs of change, whether mental or physical, and motion verbs. It is therefore tempting to analyze the Greek perfect as a resultative, as Geto and von Stechow do. And although we shall do so in the following, there is really no point in changing a well-established terminology. For this reason, we shall continue to speak of the "Greek perfect," although this may give rise to apparent absurdities like 'the Greek perfect is a resultative, and not a perfect.'
4. The English and the Homeric Greek perfect
In order to bring out the difference between the semantics of the English and the Greek perfect, we will focus on Parsons' (1990) distinction between resultant state and target state, which is often exploited in formal semantic work on resultatives (e.g. Kratzer 2000). The resultant state and the target state are defined as follows:
For every event e that culminates, there is a corresponding state that holds forever after. This is "the state of e's having culminated," which I call the "Resultant state of e," or "e's R-state." If Mary eats lunch, then there is a state that holds forever after: The state of Mary's having eaten lunch. The notion of resultant state is clearly subject to the defining principle e's R-state holds at t [??] e culminates at some time at or before t
It is important not to identify the Resultant-state of an event with its 'target state.' If I throw a ball onto the roof, the target state of this event is the ball's being on the roof, a state that may or may not last for a long time. What I am calling the Resultant-state is different; it is the state of my having thrown the ball onto the roof, and it is a state that cannot cease holding at some later time (Parsons 1990: 234f.).
Using his notion of resultant state, Parsons (1990: 236) is able to formalize the semantics of the English perfect as in the following example (although he does not spell it out in predicate calculus notation):
(6) Mary has eaten the apple [??]e (eat (e) [??] Agent (e, Mary) [??] Theme (e, the apple) [??] Hold (R(e), now)) 'There is an event e such that e is an eating-event and Mary is the agent of e and the apple is the theme of e and the resultant state of e holds now (at utterance time).'
R is to be interpreted as a function from eventualities to eventualities which assigns each event e that culminates its resultant state. Parsons restricts his notion of R-state to events that culminate because he analyzes processes (Vendler's activities) as homogenous compositions of events that do culminate. We will see in the final section that this perhaps counterintuitive analysis does have some justification, at least in a modified form.
Parsons extends his analysis to states. He affirms that "s's R-state holds at t [??] the period of time for which s holds terminates at or before t." In other words, if Mary is sick, then there is an R-state, the state of Mary's being sick having terminated, expressed by 'Mary has been sick.' However, this definition of R-states of states does not seem to be correct. The natural interpretation of 'Mary has been sick' is of course that she no longer is, but the possibility of extensions like 'for two weeks now' suggests that this is only an implicature. The perfect seems also to allow an open reading. This can be accounted for if we allow that the consequent state of a state holds from the moment the state holds (inclusive or exclusive); that is,
s's R-state holds at t [??] there is a time t' < t (or [less than or equal to] t) at which s holds
Thus, s's R-state continues to hold even if s itself no longer does. In this way 'Mary is sick' entails 'Mary has been sick,' but there is no entailment from 'Mary has been sick' to either 'Mary is sick' or its negation.
Parsons's analysis is clearly simplified and it leaves important questions unanswered, especially from a strictly compositional point of view. What remains unanswered, for example, is how the different R-states to which 'Mary has eaten apples' and 'Mary has eaten the apples' refer can be traced to the difference between the bare plural and the plural with the definite article. To answer this question, a much more refined analysis along the lines of Krifka (1998, and other related work) or Verkuyl (1993) is needed. For our present purposes, however, we may abstract from this problem. Parsons' analysis gives us all we need for explaining Kenny's test. For if we accept that the English perfect states that a resultant state as defined above holds at utterance time, we can clearly predict the entailments from Kenny's test.
We may now attempt an analysis of the Homeric perfect along the same lines. As a resultative, we expect it to occur mainly with telic VPs and this seems to be the case, as noted above. It is convenient to start our analysis with the motion verbs. Consider the following example (Iliad 21.80ff.; Gero and von Stechow 2002: 16):
(7) eos de moi estin hede [morning.sub.NOM.] and [I.sub.DAT.] [be.sub.PRES.3.SG.] this duodekate hot' es Ilion eileloutha [twelfth.sub.NOM.] that to [Troy.sub.ACC.] [go.sub.PRCT.1.SG.] 'This is for me the tweltfh morning that (5) I am in Troy.'
The relative sentence es Ilion eileloutha can plausibly be analyzed in the following way:
[??]e (elthein (e) [??] Agent (e, 1.sg.) [??] To (e, Ilion) [??] Hold (T(e), now)) 'There is an event e such that e is an event of going, and the speaker is the agent of e and Ilion is the goal of e and the T-state of e holds now (at utterance time).'
This is identical with Parsons' analysis of the English perfect, except that it uses the notion of target state instead of resultant state. This reflects an important difference between the truth-conditions of the Greek and the English perfect: in this particular case, as pointed out by Comrie (1976: 59), English can make a difference between 'I have gone to Troy' (resultative) and 'I have been to Troy' (experiential), but in general, as Carlota Smith (1997: 108) says: "if the situation involves change of state, the resultant state (i.e. Parsons' target state) need not obtain at reference time." So it seems that the main difference between the English and the Greek perfect is that the former involves a resultant state and the latter a target state.
Such an analysis is a bit too simple, since it ignores the fact that in Greek, the target state is open for adverbial modification. As ancient Greek clearly distinguishes between illatival, allatival, and locatival prepositional complements, we can find evidence of this in the use of the perfect with motion verbs. In the following example, the use of the dative with the preposition amphi presupposes a locative meaning for the preposition phrase:
(8) amph' autoi bebamen around [him.sub.DAT.SG.] [walked.sub.PFCT.1.PL.] 'We stand around him', that is, 'We have walked and now stand in positions around him.'
This can be analyzed as:
[??]e (walk (e) [??] Agent (e, 1.pl.) [??] Hold (T(e), now) [??] Around (T(e), him)
This is quite different from the English perfect where an adverb can only modify the event. 'We have walked around him' can only be interpreted as to mean that there was a walking event by us around him and that the resultant state of this event holds now. It clearly does not mean that there was a walking event by us and that the resultant state of this event holds around him. We will not go further into this difference; but what seems clear is that Parsons' R-state, whatever it really is, does not allow for adverbial modification, whereas T-states do. Of course, it is tempting to conclude from this that the semantics of the English perfect involves no state at all and that an extended now-analysis (henceforth: XN-analysis) should be preferred. On this analysis, "the perfect serves to locate an event within a period of time that began in the past and extends up to the present moment" (Dowty 1979: 341). For the early Greek perfect, however, the XN-analysis is excluded, since it involves no notion of state to which the prepositional phrase in (8) could apply.
5. Target state in Greek
5.1. "Default aktionsart"
It is well known that in languages with no overt aspectual morphology, telic predicates tend to be understood perfectively and atelic predicates imperfectively. Consider the following German examples from Bohnemeyer and Swift (2001), where a formal analysis of the phenomenon in terms of a default aspect-operator is offered:
(9) Als ich Marys Buro betrat, schrieb sie an einem Brief. When I Mary's office entered wrote she at a letter. (10) Als ich Marys Buro betrat, schrieb sie einen Brief. When I Mary's office entered wrote she a letter.
In (9) one understands that the writing event was going on during the entering event, that is, the main clause is understood imperfectively. Such an interpretation is possible in (10) also. However, in (10), it is more natural to assume that the onset of the writing event coincided with the entering, that is, the main clause is understood perfectively. These interpretations can be shown to be no more than implicatures, but there is clearly a connection between telicity and aspect. This connection also holds in the opposite direction, in what may be labelled a "default aktionsart" phenomenon. This is true, for example, in Russian where aspect is morphologically marked, but where there are no articles, which means that some noun phrases can be ambiguous between count and mass interpretation. In such cases, the perfective aspect selects for the interpretation of the object NP as count noun:
(11) Ivan pil pivo Ivan [drink.sub.IMAF.PAST] [beer.sub.ACC.] (12) Ivan vypil pivo Ivan [drink.sub.PFV.PAST] [beer.sub.ACC.]
In (12), pivo is most naturally understood as a count noun (6) so that the predicate as a whole is telic, whereas both interpretations are possible in (11).
A similar phenomenon no doubt lies behind the Greek opposition between ebouleuon 'they deliberated' and ebouleusan 'they decided' as discussed above. As we concluded in that section, such a "coercion" effect applies to the perfect as well, and it clearly must be taken into account in an analysis of the target state to which it refers.
5.2. Target state from telic VPs
With telic VPs, adopting Dowty's (1979) decompositional analysis, it is easy to define the target state of a telic VP (BECOME [??]) as [??]. Adopting Dowty's semantics for BECOME [??] (Dowty 1979: 140ff.), we see at once that the target state of an event holds once the event has culminated. There is one further complication, however, namely that the Greek perfect expresses a state of the subject and thus cannot readily combine with externally caused change-of-state verbs (i.e. CAUSE BECOME verbs). For example, there is little evidence that the root *[g.sup.wh]en- 'kill'--although obviously telic--could form a perfect in Indo-European. But in Greek, there is a tendency to "build a conjugation" that will be discussed below. This means that in the end, all verbal roots are given a perfect. We find three strategies to form perfects from externally caused change-of-state verbs. We will discuss the two first here, and the third in connection with the perfect of activities.
First, the perfect can interact with the argument structure of the verb: the verb (ap-)ollumi is transitive and regularly means 'ruin, destroy.' Following McKoon and Macfarland (2000: 834), we can give its meaning as (([??]) CAUSE (BECOME (x (be ruined)))). The perfect olola, however, is intransitive and regularly means 'I have perished, I am dead, ruined, etc.' Note that there is no implication that the subject brought about its own ruin; we cannot assume that the meaning is (([??]) CAUSE (BECOME ([??] (be ruined)))).
The second strategy is essentially the same, but with added medio-passive morphology on the verb. It seems certain that the IE perfect did not have medio-passive forms, and they are still rare in Homer. However, examples can be found like outao 'wound' (([??]) CAUSE (BECOME (x (be wounded)))), which has a medial perfect outasmai 'I am wounded.' As noted, the effect is about the same, but in this case there is explicit medio-passive morphology on the verb.
These two phenomena are in many ways reminiscent of the anticausative alternation, and can probably be analyzed in the same way (see, e.g., Saebo 2001 with references). We will not pursue the problem here, since it is not directly relevant for the analysis of Aristotle's test. We may safely assume that Aristotle did not intend his test to apply to cases where the perfect interacts with the argument structure, since an entailment like ou gar hama ollusi kai olole ('it is not true that at the same time a thing is destroying and is destroyed') is obviously and trivially false. We will therefore ignore this problem and keep to the simple assumption that the perfect of a telic VP BECOME [??] asserts that the target state [??] of this event holds.
5.3. Target state from stative VPs
This analysis can furthermore be directly extended to states, if we admit that the perfect of state-VPs first demands a perfective reading and then selects for the ingressive one. (7) This is essentially the analysis adapted above; oikeka, the perfect from oikeo 'live' affirms that the target state of a past event of settling (BECOME live at) holds at present. In this case, it is utterly implausible that the perfect should have an intensive force. In other cases too, where such an interpretation is a priori possible (as with getheo 'be happy'), no such difference can be safely read out of the texts. The category "intensive perfect" seems to be a creation of philologists startled by the apparent synonymy of the perfect and the present in such cases. The "intensive perfect" analysis is also problematic on diachronic grounds, as Gero and von Stechow (2002)--themselves partisans of such an analysis--admit: "one wonders by which historical coincidence two very different meanings, intensity and resultativity, were associated to the same morphology."
5.4. Target state from activity VPs
At first, we should note that the so-called "intensive perfect" is not limited to stative VPs; examples can also be found from dynamic VPs. This is particularly common with noise-verbs, such as kekraga 'I shout,' which is very hard to distinguish from the present krazo. These can probably be analyzed along the same lines as the corresponding stative VPs.
In principle, events that fall under activity VPs do not give rise to a target state in any obvious way, and more often than not, they do not have perfect forms at all in Homer. There is one notable exception, namely the motion verbs. We have already seen that such verbs can appear in the perfect with a locatival prepositional phrase. Moreover, such perfects can appear without ANY prepositional phrase, but they still imply a preceding, CULMINATED--and not terminated--motion event. Consider the following example (Iliad 15.90):
(13) Here, tipte bebekas; [Hera.sub.VOC] why [walk.sub.PFCT.2.SG].
Translators give 'Hera, why are you here?' or 'Hera, why have you come?' and this is obviously the right translation: tipte bebekas cannot mean 'Why have you been walking?' We clearly have a case of "default aktionsart." Default aspect, as discussed briefly above, is an implicature that may be cancelled by other factors; it would be interesting to know whether this is so for default aktionsart too. However, the materials allow for no such interpretation--there seem to be no cases of the perfect from a motion verb referring to the target state of a past event which did not culminate. (8) Indeed such an event would not at all give rise to a target state, but only a resultant state. While, as we will see, there are indeed Greek perfects referring to something like a resultant state, this never seems to happen with motion verbs, where it was too easy to impose a telic reading.
With other activity verbs, however, a telic reading could not so easily be imposed, and such perfects are exceedingly rare in Homer. There are, however, a couple of interesting examples: dakruo 'to cry' and kamno 'to work' do form perfects in Homer, and these have the meanings 'to be tear-stained' and 'to be tired.' These examples are interesting because they are exactly paralleled in other languages: Nedjalkov (1988: 35f.) cites Nivkh and Evenki resultatives from the verb 'to cry' with the meaning 'to be tear-stained,' and from Hausa, a resultative from the verb 'to work' which means 'to be tired from work.' It seems that when a state occurs frequently enough as a consequence of some action described by an atelic predicate, this state can be expressed through what Nedjalkov labels a quasi-resultative. We will see more Greek examples when we turn to the post-Homeric period.
6. The post-homeric development
The development of the verbal system from Homer and down to classical times has been described by Chantraine as one of "building a conjugation," and the term seems appropriate. Greek had inherited from Indo-European a system with quite independent verbal stems, in many cases formed from different roots. A mere glimpse at the Lexikon der Indo-germanischen Verben (Rix 2001) gives an impression of the extent of suppletivism and the independence of the aspect stems in Indo-European. But within the history of the Greek language, the verbal system tends to evolve into a system where verbs are lexical unities that can be conjugated through the different categories. We will now look at the consequences this has for the meaning of the perfect.
The first and obvious conclusion to draw is that when all verbs are given a perfect, the so-called quasi-resultatives multiply. We have already seen that such quasi-resultatives denote states which are hard to formalize. In fact, they often seem to come close to Parsons' resultant state. We will consider a few such examples where the Greek perfect seems to take on a meaning quite close to the English perfect. In fact, such a development from a strict resultative to a perfect like the English one seems to be quite a common feature in the semantic evolution of languages. In Greek, with its extraordinary diachronic depth, we can observe this evolution as it proceeds, and this gives us a unique opportunity to see how such a change comes about.
After the study of the examples which seemingly must be analyzed in terms of Parsons' resultant state, we will point to some oft-quoted difficulties with this notion, leading up to a discussion of Gero and von Stechow's analysis of the classical Greek perfect within an XN-approach. We will argue that the similarity of the resultant state analysis and the XN-theory in the case of atelic and stative predications constitutes exactly the link which makes the semantic transition from resultative to perfect possible.
6.1. The perfect with atelic VPs
We have already seen some such cases in the Homeric perfects kekmeka and dedakrumai. These perfects clearly denote states, although these states are idiosyncratically derived and not always predictable from the meaning of the base verb. Such verbs can still be found in classical Greek. Consider first the verb tikto 'to beget.' In the present and the aorist, this verb is of course transitive, taking the child as an object. The perfect tetoka, however, is intransitive and carries the meaning 'be a mother' or 'be in the puerperal state,' as in the following example from Xenophon (Cyn. 5.13):
(14) ta men tetoke, ta de tiktei, ta de either [beget.sub.PFCT.3.SG.] or [beget.sub.PRES.3.SG.] or kuei be-[pregnant.sub.PRES.3.SG.] '(The hare is so prolific, that) she is either in the puerperal state or giving birth or pregnant.'
Although, linguistically speaking, being in the puerperal state is not a result of begetting, world knowledge tells us that this is the way things normally are, and this is enough to justify a quasi-resultative. It is hard to generalize on (not to mention formalize) the semantic derivation of such states; they must be specified in the lexicon. However, some generalizations are possible. Quite often, for example, they refer to the realm of responsibility. A particularly nice example (Sophocles, Antigone 442f.) is analyzed by Rijksbaron (1994: 35):
(15) phes e katarnei me [admit.sub.2.SG.PRES.] or [deny.sub.2.SG.PRES.] (not) dedrakenai tade? [do.sub.INF.PFCT.] [this.sub.ACC.PL.] kai phemi drasai kouk and [admit.sub.1.SG.PRES.] [do.sub.INF.AOR.] and not aparnoumai to me. [deny.sub.1.SG.PRES.] (not)
Creon asks Antigone phes e katarnei me dedrakenai tade with drao 'to do' in the perfect: 'do you admit or do you deny that you are responsible for these actions?' Antigone aptly answers in the affirmative, but with the verb in the aorist: she admits to having done what Creon says, but implicitly, by using the aorist and not the perfect, she denies any consequent state of guilt.
(14) quite clearly involves reference to a state which goes beyond a mere current relevance or a participant property. In (15), this is already less clear, although the opposition between the perfect and the aorist does seem to involve more than current relevance. There are, however, cases where the perfect refers to no more than Parsons' resultant state. Many such examples are discussed in Gero and von Stechow (2002), from whom we cite the following (Plato Apology 17c):
(16) di' honper eiotha legein kai en agorai epi ton trapezon, (the words) by which I am accustomed to speak also in the market at the banker's tables hina humon polloi akekoasi where [you.sub.GEN.PL.] [many.sub.NOM.PL.] [hear.sub.PFCT.3.PL.]
The last relative clause must be translated "where many of you have heard me," and this seems to come quite close to the English experiential perfect, which can be analyzed in terms of Parsons' resultant state or within an XN-approach. We will return to the question as to which of these should be preferred, but first we will have a look at the use of the perfect with telic VPs. It seems that the difficulty of constructing any reasonable target state for many atelic VPs, combined with the general tendency to build a conjugation and conjugate all verbs through all tenses, led to the perfect of atelic verbs referring to resultant states. But for telic VPs, there is in general no problem in defining a target state, and this made it possible for the perfect to retain its old resultative force in such cases, as we shall see.
6.2. The perfect with telic VPs
Telicity in this connection must be somewhat delimited, since the Greek perfect orginally could only denote a state of the subject. It is controversial whether this is still so in classical Greek; Wackernagel (1904) and Chantraine (1927: 122) claimed to have found object resultatives already in Pindar (fifth century BC), whereas McKay (1965, 1980) and Ringe (1984) have doubted this and claim that the first examples appear much later, in the Christian era. The nuances involved are so fine that we shall probably never get an exact answer. Consider the following example from Thucydides (5.26.1):
(17) gegraphe de kai tauta [write.sub.PFCT.3.SG.] and [this.sub.ACC.PL.N.] ho autos Thoukudides Athenaios the-same-Thucydides-[Athenian.sub.NOM.SG.]
It is hard to know whether the perfect here expresses the state of this being written (object resultative) or the state of Thucydides being the author of this (subject quasi-resultative), although the highly marked sentence-final position of the subject speaks for the last analysis, which implies that the predicate is treated as atelic (since it does not give rise to a state of the subject) and thus given a quasi-resultative interpretation: writing something is often followed by the state of being an author. We will not pursue this problem further, but rather concentrate on those telic predicates that clearly give rise to a state of the subject.
That the perfect in such cases still in classical times refers to a resultant state and not a target state is shown quite clearly by the verb baino. Consider the following example from Euripides (Heraclidae 62, similarly Sophocles, Oed. Col. 52):
(18) gai' en hei bebekamen [land.sub.NOM.] in [which.sub.DAT.SG.] [walked.sub.PFCT.1.PL.]
The use of the dative with the preposition en implies a locatival force without any implication of motion. If we translate word for word, we get the English sentence 'the land in which we have walked,' which is naturally understood as an experiential perfect of an activity VP. This is, however, not the right meaning, as the context makes clear. The intended meaning is clearly 'the land (to which we have come and) in which we now are.' This means that there must be some state present which can be modified by the prepositional phrase. We have already seen that this does not happen with resultant states. It also seems clear that the truth conditions of such perfects as bebeka do indeed demand that the subject is still at the location towards which the preceding walking event was directed. Only in this way can we explain how it happened that bebeka, by losing its implication of a preceding event, came to approach the copula in meaning (see Liddell and Scott 1996: baino A.I.2).
Further confirmation that the Greek perfect, even in classical times, cannot be analyzed within a simple resultant state-approach comes from the fact that "default aktionsart" phenomena are still alive in classical Greek, as illustrated by examples (3) and (4) above, and also by (18), where there is no illatival prepositional phrase (although it could arguably be supplied from the locatival phrase). It is furthermore clear that with telic VPs, it belongs to the truth conditions of the Greek perfect that the target state holds at present, whereas this is merely an implicature with the English perfect. We have already seen this with motion verbs, but it also holds for other telic VPs. For example, on hearing 'the patient has woken up,' we naturally infer that the patient is awake, but this implicature can be cancelled, for example, 'the patient has woken up occasionally, but always slips back into a coma.' In classical Greek, however, as far as one can judge, the perfect egregora from egeiromai 'wake up' simply means 'is awake'--that the target state holds at present is no mere implicature, as an XN-analysis would predict.
In claiming that the XN-analysis can be applied to the classical Greek perfect, Gero and von Stechow seem not to have realized that their examples are mostly from atelic verbs, whereas the examples they give for the resultative meaning are from telic verbs. The only possible example they produce for a telic VP in an XN-use is the following:
(19) ei tis emoi kai Eratosthenei ekhthra popote if some [I.sub.DAT.]-and-[Eratosthenes.sub.DAT.] enmity ever gegenetai [arise.sub.PFCT.3.SG.]
which they translate "(ask yourselves) if any enmity ever arose between me and Eratosthenes (beside this one)." It is probably the adverb popote 'ever' which led them to render gegenetai by a simple past 'arose.' While it is true that this adverb is unusual with the perfect, we have already seen that the perfect does allow for event-modifying adverbs; we might thus translate 'ever arose and now exists,' maintaining a strict target-state analysis.
6.3. The diachronics of the Greek perfect from Homer to classical times
The picture I would like to suggest is the following: in Homer, the perfect always referred to a target state. This target state could in most cases be derived directly from the semantics of the base verb (in resultatives from telic VPs); in other cases, the connection with the base verb was looser (as in the quasi-resultatives). However, as the perfect category expanded to be used with almost all verbs, it came to denote a resultant state with atelic verbs, although its use with telic verbs still denoted a target state. At this point, it may seem that the perfect has no unitary semantics, as it denotes a target state with telic verbs and a resultant state with atelic verbs. However, it seems possible to unify these two by exploiting the notion of event realization (see Bohnemeyer and Swift 2001):
An event e denoted by a predicate P is realized at an interval t if and only if there is a subevent e' of e that falls under the predicate P and whose runtime is included in t. In other words, events that fall under atelic predicates are realized as soon as they have begun, as long as the interval considered is large enough to for the event to count as an instance of the predicate: some strokes of the pen are enough to realize an event failing under the predicate 'write.' Events that fall under telic predicates, however, are only realized when they have culminated: a letter must be finished for the event denoted by 'write a letter' to be realized.
We could claim, then, that the perfect of a predicate P at this point refers to the state resulting from the event denoted by P (in the perfective aspect) being realized. As we shall see, this is a bit vague concerning atelic predicates, since no such state can be clearly defined, but this vagueness was probably exactly what triggered the further semantic evolution. Perfects from stative VPs are ambiguous just as their aorists are. A perfect based on the ingressive reading of the perfective aspect will be something like the old 'intensive perfect' (and these are rare in classical times), whereas perfects based on the complexive reading (see note 7) will have semantics very close to the English perfect with states. The ambiguity here lies in the aorist: ebasileuse can mean 'became king' and 'was king' (for a certain period), and the corresponding perfect (which in this case only appears late) could be based on either of these readings. However, the ambiguity seems to be lexically resolved: for example, sesigeka (from sigao 'to keep silent') means 'the event of my falling silent was realized in the past and the result still holds,' whereas akekoa (from akouo 'to hear') means 'my hearing was realized in the past and the resultant state still obtains.'
As one can see from the last example, it is quite unclear what state a perfect from an atelic predicate refers to. This has often been pointed out as a difficulty in discussions of resultant state theories of the English perfect: it seems difficult to pick out the right state. For telic predicates there is one especially salient state, since all events that fall under such predicates are bordered at the right by a state, the target state. Even in English, where the perfect does not explicitly denote the target state, the present existence of this state is often an implicature, as we have seen. For atelic predicates, there is no easy way to identify a resultant state: we have cited Parsons' definition above and used it throughout this article, but it is clear that with such a broad notion of resultant state, we come dangerously close to an indefinite past theory, which means that we cannot explain the unacceptability of the following sentence (Portner 2000: 12):
(20) *Gutenberg has discovered the art of printing.
And if we, following Carlota Smith, identify the resultant state with the participant property, it seems hard to account for:
(21) Frege has contributed a lot to my thinking. (Portner 2000: 13)
We will not discuss these problems further, but simply note that the difficulties of indentifying the right resultant state have led many linguists to reject the resultant state approach in favor of an XN-analysis. It is clear, however, that both approaches yield about the same truth conditions for the English perfect. This is easily seen in discussions of the relative merits of such theories, which always focus on theoretical simplicity and predictions for rather strained cases. This is by no means unjustified, but it does show that both theories make the same empirical predictions about the core meaning of the perfect. Analogously, the "state of event realized" analysis presented here gives at least approximately the same truth conditions as the XN-analysis FOR ATELIC PREDICATES, whereas they differ significantly for telic predicates.
If we now assume that Greek speakers of the classical period interpreted the perfect as denoting the state resulting from an event being realized, we may conjecture that the vagueness of resultant states from events falling under atelic predicates caused a new XN-interpretation to evolve in this domain, where the truth conditions remained unaffected. In the same way, the perfect with state verbs based on the complexive use of the aorist (as akekoa above) can be analyzed in an XN-interpretation. Both of these changes can be understood as cases of Andersen's (1973) abductive change: the speaker abduces a new grammar which yields the same output. However, when this semantic interpretation, by deductive innovation, was transferred to telic VPs, the truth conditions changed radically and the Greek perfect became a real perfect and no longer a resultative. This was clearly the case in post-classical Greek, and it is possibly underway in classical times, if Gero and von Stechow's interpretation of (19) is correct. Later, and outside the scope of this article (but see Gero and von Stechow 2002), the perfect fell together with the aorist, as also happened with the French passe compose. In this way, a combination of the resultant state theory and the XN-theory allows us to explain the typologically very common evolution of resultatives into perfects. This evolution runs counter to the principle of "strengthening of implicatures" that is often observed in semantic evolution, since in the use with telic VPs, the present existence of the target state is weakened from a truth condition to an implicature. The driving force behind this evolution seems to be the desire to avoid lexical restriction of a grammatical morpheme.
7. Aristotle once again
Let us now return to Aristotle's test: it is clear that accomplishments and achievements pass his kinesis-test. For the perfect of such a VP to be true, the event must have been realized in the past, and for events which fall under telic predicates, this means that they must have culminated. In this case, the present tense VP is no longer true. So, telic situation types clearly belong to the category of kineseis, and this also seems to be the common opinion of scholars.
States are also no problem; we have seen that the perfect of state verbs in early Greek is often identical with the meaning of the present; thus the entailments of Aristotle's energeia-test clearly go through. In classical Greek, there is possibly a tendency to form experiential perfects (i.e. perfects based on the complexive use of the perfective aspect), but this does not affect the entailments. Once the present tense state VP can be affirmed, the state must also have held in the past.
However, it seems that activities have much as unsure a place within the Aristotelian classification as within a modern ontology of states and events (see, e.g., Smith 1999), a problem that we will return to. Nevertheless, a more complete understanding of the semantics of the Greek perfect enables us to understand his test better. It is quite clear that Aristotle, in many cases, views what WE would classify as an activity-VP as an incomplete description of an accomplishment. It is obvious that baino eis Athenas excludes bebeka eis Athenas, because if I am walking to Athens, I have not yet reached the state of having walked to Athens and being there. But we can also see why baino precludes bebeka, even when there is no goal expressed, because--by the phenomenon of default aktionsart --the perfect imposes a telic reading by supplying such a goal. bebeka, as opposed to the English perfect of the verb 'to walk' simply cannot mean 'I have done a bit of walking'--one has to understand a goal, as in our Iliad example (13). Such an implication of a goal must also lie behind the classical use of bebeka as almost a copula. Default aktionsart also applies to verbs like 'to thin' (iskhnaino) and 'to learn' (manthano). To be telic, such verbs must presuppose a telos in the form of a definite state of thinness and something being learned, respectively.
But are we not then faced with the difficulty noted by Ackrill, namely that we do not know where we should supply such complements and where we should not? No, because it is not we who are supplying the complement, the telic reading is imposed by the use of the perfect with such activities. With states, however, the situation is different, because they cannot be made telic by supplying an object. Indeed a verbal object does not limit a state in the way that it limits events. Obviously, the object of my seeing does not limit my seeing in the same way that the object of my walking limits my walking. As Aristotle himself says, in the Nicomachean Ethics, 1174b 5-6, the whither and whence--the termini --constitute the form of kineseis, whereas enjoying and other energeiai seems to be complete at any moment. In interpreting Aristotle's test, nothing is to be supplied BY US; the use of the perfect by itself supplies what there is, so there are indeed no presupposed objects or other complements to be supplied in Aristotle's text.
Following this principle, it is clear that we must interpret hedesthai, claimed by Aristotle to be an energeia, as a state which cannot be limited by an object. Ackrill may be right to argue that while we are enjoying the symphony, we cannot yet be said to have enjoyed it, but this depends on viewing the English verb 'enjoying' as an activity, and is no objection to Aristotle's test. If we conceive of hedesthai qua energeia as an activity, this would make Aristotle's test fail; but we make a presupposition which is not necessary, and it is clearly preferable to have an interpretation which makes Aristotle's grammatical test a viable one. If we take the semantics of the Greek perfect into account, it is possible--without unnecessary presuppositions--to interpret the test as a grammatically correct one, bringing out interesting properties of the Greek perfect.
Is the test also ontologically correct? Strictly speaking, it is not. There are indeed some activities which would pass Aristotle's energeia-tests, namely those that cannot easily be given a telic interpretation, that is, such inergatives as 'scream,' 'cry,' etc. The unsure position of activities in Aristotle's taxonomy and the fact that it cuts through what we see as a homogenous class 'activities' seem to derive from a supposition which is also commonly found in modern literature on aspect and aktionsart, namely that differences brought out by linguistic tests necessarily define ontological categories. This is clearly a presupposition of Aristotle's, and the corresponding view that the Vendler classes are ontological categories which divide events into mutually exclusive classes is found in modern semantics (see, e.g., Mourelatos 1978; Bach 1986 and Smith 1999--generally all talk of "telic events" presupposes such a view). But this seems to be contradicted by the fact that the telic predicate 'Mary walked to school' and the atelic predicate 'Mary walked' can be made true by one and the same event (a manner of speaking that clearly presupposes that we have sufficiently clear identity conditions for events). The natural boundedness criterion, which is often invoked in making the distinction between activities and accomplishments/achievements clear, is also hard to apply to an ontological distinction. In contrasting the predicates 'write a letter' and 'play soccer,' we easily see that the first includes a bound which is not present in the second. This is not at all clear when we consider the corresponding events. It is hard to see that the finished letter is a natural (as opposed to arbitrary) endpoint to my writing, since I can finish it whenever I like by adding a polite formula and sign. And if the ontological distinction between arbitrary end and natural endpoint cannot be made good, the distinction between activities and accomplishments also cannot be upheld in the ontology, since there is, quite trivially, an end to every event. (9)
This also poses problems for the notion of culmination as a two-place predicate of events and times (as in, e.g., Parsons 1990); if I have eaten an apple, then an event has culminated, namely the event of my eating an apple. But, along the way, a lot of other events have culminated, for example, the event of my eating half an apple. Eating an apple seems to have no more (and no less) a natural endpoint than eating half an apple or eating two apples. From this, we must conclude that culmination is a three-place predicate of events, telic VPs, and times--or we may dispose of culmination altogether and adopt the notion of event realization, which amounts to the same for events that fall under telic predicates but is also defined for events that fall under atelic predicates, which are realized on any subinterval down to intervals sufficiently large for the subinterval property to hold. This notion of event realization is reminiscent of Parsons' analysis of activities as being composed of homogenous events that do culminate, but with the important difference that event realization is a three-place predicate of events, predications, and times.
But whereas events cannot be further subdivided in the ontology, there is obviously an ontological distinction between events and states, whether the latter are taken as individuals or as properties of times. And, although Aristotle's test does not strictly speaking bring out this distinction, it is still possible that he had the distinction between states and events in mind. It is notable that all stative verbs pass his energeia-test; he may have conceived of the kinesis-test as an ancillary test that would not necessarily apply to all kineseis. As Graham (1980) argues, all of Aristotle's examples of energeiai should be interpreted as stative verbs. It is no less clear that all the examples of kineseis are verbs which--at least in their perfective aspect--are telic predicates. The fact that Aristotle does not at all mention such verbs as dakruo 'cry,' kamno 'work,' or other dynamic predicates which would pass his energeia-test might suggest a genuine hesitation as to where they should belong in his ontology. On the other hand, he must have been sure that 'living' should be included among the energeiai, since he probably made up the form ezeke in order to make zo 'to live' pass his energeia-test.
There is another point which suggests that Aristotle is really after the event/state distinction. In the section of the Nicomachean Ethics (1173-1174) on pleasure, (10) Aristotle tells us that pleasure is not a kinesis since it cannot be said to be quick or slow. Interestingly, he exploits the ingressive aorist in exposing this difference: "we can get pleased (hesthenai, aorist infinitive) quickly like we get angry, but we cannot be pleased (hedesthai, present infinitive) quickly." (1173a, lines 34-35) If this incompatibility with such adverbs is thought of as a criterion for energeia, it suggests states rather than activities: kamno 'to work' could probably combine with such adverbs. Thus, if we are correct in thinking that Aristotle's kinesis/energeia-distinction corresponds to the modern distinction between state and event--even though this is not clearly brought out by his grammatical test--we must conclude that Aristotle discovered a distinction which is still among the most important ones in modern semantics.
(1.) This paper originated as one of my trial lectures for the doctor artium degree at the University of Oslo, and was finished during my stay at the Sprachwissenschaftliches Seminar, Universitat Freiburg, as a research fellow of the Alexander von Humboldt foundation, to whom I would like to express my gratitude. I would also like to thank Atle Gronn, Eirik Welo, and Halvard Fossheim (all Oslo) for comments on earlier versions of this paper, Christina Leluda-Voss (Freiburg) for interesting discussions, and Prof. James Cathey (Amherst) for help with some of the English examples. Correspondence address: Klassisk og romansk institutt, Postboks 1007 Blindern, 0315 Oslo, Norway. E-mail: email@example.com.
(2.) The German term "aktionsarten" is kept here (though without capitalization), since no universally accepted English equivalent exists. The term seems to have been coined by Karl Brugmann (1885) in his Greek grammar and thus belongs to a period when much German linguistic terminology was borrowed into English; ablaut and umlaut are still main entries in Trask (2000), although the calques apophony and metaphony have been created.
(3.) In the transcriptions, I ignore accents, since they are not imporant here. The digrams kh, th, ph stand for unvoiced, aspirated stops that are noted by single letters of the Greek alphabet. Accord to tradition, vowel length is marked with makron, and not, as in the IPA system, with [??]. I transcribe vowel length only when it is indicated in the Greek alphabet, that is, only by e and o, and when a subscribed iota follows.
(4.) As has been pointed out in the literature on Kenny's test (e.g. Parsons 1990: 37ff.), the test is only valid if we understand that both verbs refer to the same event, in this case a curing-event.
(5.) It is important to realize that hote does not mean 'since' but is a relative adverb of time.
(6.) However, as the editor points out, a mass noun interpretation can still be forced in perfective sentences by means of the partitive genitive. Thus, Ivan [vypil.sub.PFV.PAST] [piva.sub.GEN.] would mean 'Ivan drank (some) beer (at one occasion).' It is arguable whether this sentence should be considered telic.
(7.) Note that the Greek aorist can have two readings with states: ebasileuse (the aorist from basileuo 'to be king') can have the meanings 'became king' (ingressive) and 'was king' (traditionally called 'complexive use'). In Homer, perfects from such verbs are always based on the ingressive reading, whereas in classical times they are often based on the complexive use.
(8.) Perfects of motion verbs with allatival prepositional phrases are instructive in this regard: consider Iliad 6.495 oikonde bebekei (homewards [walk.sub.3.SC.PLUPFCT.]) which means 'she was walking homewards.' It seems that the allatival oikonde (as opposed to an illatival eis oikon 'to the house') excludes an endterminative reading. Instead, we get an ingressive reading 'BECOME be walking homewards.' It is interesting that some translators (e.g. the Norwegian Eirik Vandvik) have seen a reference to the beginning of the event in this passage.
(9.) This is a most intricate problem for semantics, as can be seen from the treatment in Krifka (1998: 207), where initial and final parts of an event e are defined so that e' is an initial part of e if it is not preceded by any part of e, and similarly for final parts. This clearly (and in my opinion correctly) does not distinguish natural and arbitrary limits. Telicity is then defined as the property of an event predicate X that applies to events e such that all parts of e that fall under X are initial and final parts of e. This is the mereological heterogeneity criterion for telicity: telic predicates are heterogenous on every interval that does not contain the initial and final part of the event. However, this criterion is too strict, as can be seen from the following example: I build a house, thereby completing an event e' of building a house. However, my wife, refuses to move in before I have built a balcony and I immediately go on to fulfill her wishes, thereby completing an event e of building a house. In any natural interpretation, e' is a subevent of e, and they both fall under the predicate 'build a house,' but e' is not a final part of e. Thus, 'build a house' does not fulfill the telicity criterion of Krika, which it should. This might lead us to reject the heterogeneity criterion for telic predicates, and instead adopt a homogeneity criterion for atelic predicates, but, as is well known (Dowry 1979: 163ff.), atelic predicates are also not homogenous at smaller intervals. Thus, since not all atelic predicates are totally homogenous and not all telic predicates totally heterogenous, we are (from the mereological view) forced to conclude that (a)telicity is not a binary feature, but rather a scale. That this can be relevant for linguistics is shown by the following examples from Smollett (2002). On a scalar interpretation of telicity, it is clear that 'build a lego tower' is less telic than 'build a house' since a lego tower has no set end point, it can be added to indefinitely and the result is still a lego tower, at least from the point where it has reached the height necessary to qualify as a lego tower. And indeed, Smollett claims that 'Thomas built a lego tower for three hours' is fine for her. She also claims that 'Kathleen ate an apple for a couple of minutes while talking on the phone' is fine. We might just speculate that an apple has a less clear end point than other things, since most people do not eat the kernel. In the same way, 'bake the cake' (as discussed by Zucchi 1998) has an unclear end point, which accounts for the acceptability of 'John baked the cake for an hour.' Clearly, a telicity hierarchy should be defined and the theory further developed for such cases; see Zucchi (1998) for an attempt involving precision states.
(10.) Note that in the following, I translate hedesthai as 'be pleased' and not as earlier as 'enjoy,' in order to make clear the state connotations that I think are there.
Ackrill, John Lloyd (1965). Aristotle's distinction between energeia and kinesis. In New Essays on Plato and Aristotle, Renford Bambrough (ed.), 121-141. London: Routledge.
Andersen, Henning (1973). Abductive and deductive change. Language 49, 765-793.
Bach, Emmon (1986). The algebra of events. Linguistics and Philosophy 9, 5-16.
Bohnemeyer, Jurgen; and Swift, Mary (2001). Default aspect: the semantic interaction of aspectual viewpoint and telicity. Paper presented at the Perspectives On Aspect conference, Utrecht 2001. http://www-uilots.let.uu.nl/conferences/Perspectives_on_Aspect/ Proceedings/bohnemeyer.pdf
Brugmann, Karl (1885). Griechische Grammatik, vol. 2, Handbuch der klassischen Altertumswissenschaft. Nordlingen: C. H. Beck.
Chantraine, Pierre (1927). Histoire du parfait grec. Paris: Honore Champion.
Comrie, Bernard (1976). Aspect. Cambridge: Cambridge University Press.
Dahl, Osten (1985). Tense and Aspect Systems. Oxford: Blackwell.
Dowty, David R. (1979). Word Meaning and Montague Grammar. Dordrecht: Reidel.
Gero, Eva; and von Stechow, Arnim (2002). Tense in Time." The Greek Perfect. http://vivaldi.sfs.nphil.uni-tuebingen.de/~arnim10/Aufsaetze /Gero-Stechow.8.2.02.pdf. Now published in Words in Time, Regine Eckard, Klaus von Heusinger, and Christoph Schwarze (eds.). Berlin: Mouton de Gruyter, 2003.
Graham, Daniel W. (1980). States and performances: Aristotle's test. The Philosophical Quarterly 30, 117-130.
Jaeger, Werner (1957). Aristotelis Metaphysica. Oxford: Clarendon Press.
Kenny, Anthony (1963). Action, Emotion and Will. London: Routledge.
Kratzer, Angelika (2000). Building Statives. Berkeley Linguistic Society. http:// semanticsarchive.net/Archive/GI5MmI0M/kratzer.building.statives.pdf
Krifka, Manfred (1992). Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan A. Sag and Anna Szabolsci (eds.), 29-53. Stanford: Center for the Study of Language and Information.
-- (1998). The origins of telicity. In Events and Grammar, Susan Rothstein (ed.), 197-235. Dordrecht: Kluwer.
Liddell, H. G.; and Scott, Robert (1996). A Greek-English Lexicon: With a Revised Supplement. Oxford: Clarendon Press.
Lyons, John (1967). Structural Semantics. An Analysis of Part of the Vocabulary of Plato. Oxford: Blackwell.
McKay, K. L. (1965). The use of the ancient Greek perfect down to the second century A.D. Bulletin of the Institute of Classical Studies 12, 1-21.
-- (1980). On the perfect and other aspects in the Greek non-literary papyri. Bulletin of the Institute of Classical Studies 27, 23-49.
McKoon, Gai; and Macfarland, Talke (2000). Externally and internally caused change of state verbs. Language 76, 833-858.
Mourelatos, Alexander (1978). Events, processes and states. Linguistics and Philosophy 2, 415-434.
Nedjalkov, Vladimir P. (1988). A Typology of Resultative Constructions. Amsterdam: John Benjamins.
Parsons, Terence (1990). Events in the Semantics of English. Cambridge, MA: MIT Press.
Portner, Paul (2000). The (Temporal) Semantics and (Modal) Pragmatics of the Perfect. http://semanticsarchive.net/Archive/2MxZjI3M/
Potts, Timothy (1965). States, activities and performances. Proceedings of the Aristotelian Society, suppl. vol. 39, 65-84.
Rijksbaron, Albert (1994). The Syntax and Semantics of the Verb in Classical Greek. Amsterdam: J. C. Gieben.
Ringe, Donald A. (1984). The perfect tenses in Greek inscriptions. Unpublished doctoral dissertation, Yale University.
Rix, Helmut (ed.) (2001). Lexikon der Indogermanischen Verben, 2nd ed. Wiesbaden: Reichert Verlag.
Rodriguez Adrados, Francisco (ed.) (1980). Diccionario Griego-Espanol. Madrid: Consejo Superior de Investigaciones Cientificas.
Ross, W. David (1924). Aristotle's Metaphysics, vol. II. Oxford: Clarendon Press.
-- (1928). The Works of Aristotle, vol. VIII, Metaphysica. Oxford: Clarendon Press.
Ruiperez, Martin. S. (1982). Structure du systeme des aspects et des temps du verbe en Grec ancien: analyse fontionnelle synchronique. Paris: Belles Lettres.
Saebo, Kjell Johan (2001). An analysis of the anticausative alternation. http://semanticsarchive.net/Archive/WE3ZDA0Y/
Smith, Carlota (1997). The Parameter of Aspect. Dordrecht: Kluwer.
--(1999). Activities: states or events? Linguistics and Philosophy 22, 479-508.
Smollett, Rebecca (2002). Quantized direct objects don't delimit after all. Poster presented at the Perspectives On Aspect conference, Utrecht 2001. http://www-uilots.let.uu.nl/ conferences/Perspectives_on_Aspect/Proceedings/smollett.pdf
Trask, Robert L. (2000). The Dictionary of Historical and Comparative Linguistics. Edinburgh: Edinburgh University Press.
Vendler, Zeno (1957). Verbs and Times. Philosophical Review 66, 143-160.
-- (1967). Linguistics in Philosophy. Ithaca, NY: Cornell University Press.
Verkuyl, Henk (1993). A Theory of Aspectuality. Cambridge: Cambridge University Press.
Wackernagel, Jakob (1904). Studien zum griechischen Perfektum. Unpublished doctoral dissertation, Gottingen University.
Zucchi, Sandro (1998). Aspect Shift. In Events and Grammar, Susan Rothstein (ed.), 349-370. Dordrecht: Kluwer.
Received 2 July 2002
Revised version received
6 January 2003
University of Oslo