A puzzle about until and the Present Perfect
Anastasia Giannakidou
1. Introduction
This paper is concerned with the status of sentences combining a present
perfect and an UNTIL adverbial, like the ones below from Greek and English. I use UNTIL as a generic label for the connective crosslinguistically.
(1) * I
Ariadne exi zisi
sto Parisi mexri tora.
the Ariadne has lived in Paris until now
‘?Ariadne has lived in Paris until now.’
(2) * I
Ariadne exi xasi ta
klidia tis
mexri tora.
the Ariadne has lost the keys hers until now
‘* Ariadne has lost her keys until now.’
Until and its Greek counterpart mexri seem to create an anomaly when they
modify an eventuality in the present perfect.1 The anomaly is weaker in
English than it is in Greek with a stative verb like live and until now, but
notice that any other time prior to now gives a result as bad as in Greek:
(3) * Ariadne has lived in Paris until 1998.
The anomaly is quite puzzling in the context of the quite standard assumptions that (a) perfect eventualities denote result states (McCoard 1978,
Dowty 1979, Vlach 1983, Kamp and Reyle 1993), and (b) UNTIL is a stative modifier. Both assumptions would predict unobstructed compatibility
between UNTIL phrases and the perfect.
The observed incompatibility becomes even more intriguing when we see
that it is removed if we insert negation, or iterative adverbials like tris fores
‘three times’:
(4) a. I
Ariadne dhen exi zisi
sto Parisi mexri tora.
the Ariadne not has lived in Paris until now
‘?Ariadne has not lived in Paris until now.’
b. I
Ariadne dhen exi xasi ta klidiatis
mexri tora.
the Ariadne not has lost the keys
hers until now
‘Ariadne has not lost her keys yet.’
102
Anastasia Giannakidou
(5) a. I
Ariadne exi zisi
sto Parisi tris
fores mexri tora.
the Ariadne has lived in Paris three times until now
‘Ariadne had lived in Paris three times until now.’
b. I
Ariadne exi xasi ta
klidia tis
tris
fores
the Ariadne has lost the keys hers three times
mexri tora.
until now
‘Ariadne has lost her keys three times until now.’
In (4) UNTIL becomes fine if the present perfect is negated (though again,
the status of the English sentence is slightly different from that of the
Greek one); in (5) we witness an improvement with the addition of an iterative adverbial. Why should we have these contrasts?
In this paper, I will try to answer this question by exploring how the
semantics of the perfect in Greek and English interacts with the constraints
imposed by UNTIL adverbials. I will focus primarily on Greek, where the
contrast seems to be the clearest. In the last section I show that the proposed explanations accommodate English easily. The incompatibility of
UNTIL and the present perfect will be explained as a semantic clash between the semantics of durative UNTIL, which requires that a state extend
through all subintervals introduced by it, and the perfect, which contains
both an event and a result state, thereby creating a conflict with durative
UNTIL. The improvements suggest that the result state is suppressed, enabling only event predication to map onto the interval contributed by
UNTIL. Two premises will be crucial to this explanation: (a) that UNTIL
has three meanings, among which a purely temporal one; and (b) that the
aspectual information that comes from the participle is important: in Greek,
where the participle is perfective, the perfect always contains an event
(even with stative verbs). But in English, where the participle in aspectually unspecified, purely stative interpretations are also allowed. The particular differences among the two languages will be shown to follow from this
central difference.
The paper is organized as follows. In section 2, I discuss the semantics
of until and its Greek counterpart mexri. It will be shown that UNTIL connectives express three possible meanings, one of which contributes a purely temporal dimension. In section 3, I discuss the aspectual system of
Greek, concentrating on the perfective-imperfective contrast which is observed in all finite verb forms. Perfect participles, on the other hand, are
only perfective. This fact will be decisive for the particular meaning of the
Greek perfect and its difference from the English one. In section 4, I concentrate on the semantics of the Greek and English present perfect within
A puzzle about until and the Present Perfect
103
the extended now approach (McCoard 1978, Dowty 1979). We see that the
Greek present perfect is an existential perfect (the perfect of result included), with stative as well as eventive verbs, because of the perfective morphology of the participle. Finally, in section 5, I revisit the data presented
here and show that they follow from the joint analyses I proposed for
UNTIL and the perfect.
2. The various meanings of until
In this section, we see that the meaning of UNTIL crosslinguistically is
actually a cluster of three related meanings: a durative meaning; an eventive meaning which is triggered by negation; and a purely temporal meaning that appears primarily in future sentences, but is also licensed with the
perfect, as we shall see in section 5. The common core in all cases is that
UNTIL contributes an interval or a time scale upon which eventualities are
mapped: states (durative UNTIL), or events (the other two meanings).
2.1.
Durative until
Durative until is known to modify durative eventualities – states or activities (Karttunen 1974, Mittwoch 1977, Hitzeman 1991, de Swart 1996 and
references therein):
(6) The princess slept until midnight.
(7)
The princess was writing a letter until midnight.
Eventive eventualities, on the other hand, are generally incompatible with
until:
(8) a. * The princess arrived until midnight.
b. *The princess wrote a novel until she married Bill.
I call achievements and accomplishments ‘eventives’, and do not distinguish between the two unless it matters (e.g. in the discussion of the perfect). States and activities will be referred to as ‘statives’.
The generalization, then, is that until is compatible with statives but not
with eventive verb forms.
In Giannakidou 2002, I showed that the Greek counterpart of until,
mexri, exhibits exactly the same property:
104
Anastasia Giannakidou
(9) I
prigipisa
the princess
egrafe
ena
wrote.imperf. a
grama
letter
mexri ta
until the
mesanixta.
midnight
‘The princess was writing a letter until midnight.’
(10) I
prigipisa
kimotane
mexri
ta
mesanixta.
the princess
slept.imperf. until
the midnight
‘The princess was sleeping until midnight.’
(11) * I prigipisa
eftase
mexri ta
mesanixta.
the princess
arrived.perf. until the midnight
‘* The princess arrived until midnight.’
(12) * I
the
prigipisa
princess
egrapse
ena
wrote.perf. a
mithistorima
novel
mexri pu
until that
pandeftike.
married
‘* The princess wrote a novel until she married.’
The durativity of UNTIL can be captured as follows. UNTIL introduces an
interval with a well-defined endpoint (Hitzeman 1991), which is supplied
by the clock expression contained in the until-phrase. (I do not discuss
clausal until, and assume that what I say for phrasal until carries over to its
clausal counterpart.). The verb contributes a state which is then mapped
onto the until-interval, as indicated below:
(13) ...______s: sleep_______]______...
midnight
States are homogeneous, i.e. they exhibit the subinterval property (Bennett
and Partee 1978): the predicate characterizing the state is taken to be true
at all subintervals contained in the until interval. This is reflected in the
semantics in (14), from de Swart 1996 (building on earlier work by Kamp
1968; for the AT-relation see Krifka 1989):
(14) semantics for durative until
for : s [P(s)  t AT (s,t)]; : t’Q (t’)
[[until (,)]] = stt’t’’’ [P(s)  AT (s,t’’’)  Q(t’)  t  t’’’
A puzzle about until and the Present Perfect
105
t’’ [[t  t’’ < t’]  s’[s’  s  P(s’)  AT (s’, t’’)]]]
The until interval t''' extends from some (not necessarily well defined)
point t to a point t' which is the time of the clock description Q of the until
phrase. This semantics also captures the scalar nature of until. The connective introduces a range of values on the time scale. These are the times t''
which precede the time t'. The verb contributes a state P, and P is asserted
to hold at all subintervals t'' prior to t'. This semantics implies that there is a
change of state at t', and that P does not hold at t'. This, however, is a Qimplicature in the sense of Horn (1989), albeit a strong one; as such it can
be cancelled, as in (15):
(15) Sure, the princess slept until midnight. In fact she only woke up at 2
am.
Hence it seems appropriate to include t' in the P-holding interval; I indicate
this by using square brackets in (13). The scalar condition must then be
altered as t’’  t' instead of the existing t’’ < t’. For the purposes of comparison, I will also simplify the above definition and use the one below:
(16) durative UNTIL
[[mexri Q]]= P s tt’[ P(s,t)  Q(t')  RB (t,t')  t’’ [(t''t  t''
t’)  P (s,t'')]], where RB stands for ‘right boundary’
The left boundary of t remains unspecified, but this is not always so, as we
shall see later (section 5) in the discussion of boundary adverbials.
The data in (6)- (8) can now be easily explained: (6) and (7) are fine
because they contain statives: the condition that P holds is true for all relevant subintervals t''. On the other hand, the eventives in (8) are not compatible with until because events are either quantized (accomplishments) or
have no duration at all (achievements). In either case, the UNTIL requirement is not met.
106
Anastasia Giannakidou
2.2.
Eventive NPI until
This second meaning is triggered when UNTIL occurs with negation, and
corresponds to Karttunen's (1977) punctual UNTIL. In this meaning
UNTIL is an NPI triggered by negation and other antiveridical operators
(e.g. without; Giannakidou 1998, Zwarts 1995). According to Karttunen,
NPI-until, because of its punctuality, yields an inchoative meaning with
stative verbs5:
(17) The princess did not sleep until midnight. =
‘The princess fell asleep at midnight.’
(18) The princess fell asleep at midnight and not before that.
e t [midnight (t)  t< n  fall-asleep (princess, e,t )] 
e't' [tC  t'<t  fall-asleep (princess, e',t') ]
UNTIL again contributes a scale of contextually relevant times (tC, a
condition which I will henceforth drop, assuming that is always present).
But now there is indeed an event of falling asleep at midnight (or at little
later than that, as Karttunen notes; we will overlook this detail, as it is not
crucial for the present purposes), and no earlier than that:
e fall-asleep (princess, e,t)
(19) *…_____________ ____|__________...
t'
t
The actualization of an event that comes with eventive UNTIL is responsible for contrasts like below.
(20) a. * The princess didn’t get married until she died.
b. The princess remained a spinster until she died.
(21) t e e’ [die (princess, e, t)  t<n  get-married (princess, e’,t) 
e’’t’ [t’ < t  get-married (princess, e’’,t’)]]
In (20a), we have a situation where the princess both dies and gets married
at the same time t, hence the oddity. In a sentence with durative UNTIL,
however, we only have an implicature of actualization: the princess doesn’t
get married at the moment of dying and no oddity arises. The contrast argues against the stative analysis of negation: if not get married were equivalent to the stative remain a spinster, as argued in Mittwoch 1974, we
should not get a difference between the two sentences, contrary to fact.
According to Karttunen, English until is ambiguous between a durative
A puzzle about until and the Present Perfect
107
and an NPI-meaning, but Greek possesses a lexically distinct scalar expression for NPI-until – para monon, lit. but only. This item does not have the
durative semantics we described in the previous section:
(22) * I prigipisa
kimotane
para monon ta
mesanixta.
the princess
slept.imperf. but only
the midnight
‘The princess was sleeping until midnight.’
Instead, para monon is used with negated eventive verb forms, which always appear in the perfective in Greek – to be discussed shortly. It also
entails actualization, as I showed in Giannakidou 2002.
(23) I
prigipisa *(dhen) eftase
para monon ta
mesanixta.
the princess not
arrived but only
the midnight
‘The princess did not arrive until midnight.’ = It was only at midnight that the princess arrived.
That actualization is an entailment is evidenced in (24), where negating the
arrival leads to a contradiction.
(24) * I prigipisa dhen eftase para monon ta mesanixta. Dhen eftase kan
ekino to vradi.
‘* The princess did not arrive until midnight. In fact she didn’t arrive that night at all.’
The fact that the English simple past gives rise to actualization suggests
that its default value is that of a perfective past, as I noted in Giannakidou
2002. The true imperfective reading of Mittwoch (1977), which would not
entail actualization, seems to be absent with the simple past, and negation
only triggers the NPI-reading.
Crucially, durative mexri cannot be used with negated eventive forms:
(25) * I prigipisa dhen eftase
mexri ta
mesanixta.
the princess not arrived.perf. until the midnight
‘The princess did not arrive until midnight.’
This is another argument against the analysis of negation as a stativizer.
Putting the pieces together, we can safely conclude that para monon is the
lexical realization of Karttunen’s punctual NPI-until. The semantics below
captures its scalarity and punctuality (from Giannakidou 2002):
108
Anastasia Giannakidou
(26) scalar semantics for eventive UNTIL
[[dhen P para monon Q]]= e t [Q (t)  P(e,t) 
 t’ e’ [ t’<t  P(e’,t’)]]
Eventive UNTIL contributes a scale of times t' leading to an endtime t, at
which an event occurs. This is a purely scalar reading – it is no accident
that it involves an expression – para monon – that is scalar but not exclusively temporal (cf. French ne... que which has a similar use; de Swart
1996). Just like Greek, other languages seem to possess a distinct NPIUNTIL, e.g. Icelandic (see Giannakidou 2002 for details). Other languages
exclude durative UNTIL from this use altogether and employ a positive
polarity item instead, e.g. German and Dutch (Declerk 1995). Given its
non-durative semantics, para monon will not be central to the discussion in
this paper.
2.3.
Purely temporal until
This meaning of UNTIL is primarily visible in future contexts. I illustrate it
below with mexri and German bis. We see that English until does not have
this meaning. In a future context, by must be used instead:
(27) Tha to telioso afto mexri avrio.
‘I will finish this by (*until) tomorrow.’
(28) Ich erledige das bis morgen.
‘I will do it by tomorrow.’
In this reading, UNTIL introduces an interval with the UNTIL time as its
right boundary, as usual; unlike durative UNTIL, however, by-UNTIL does
not impose the requirement that a state holds throughout that interval. Rather, it allows for a free mapping of events (which may or may not have
duration).
e finish (I, this, e,t'' )
(29) #…________|____
_|__________...
t''
t'
The semantics is given in (30), where UNTILby stands for the purely temporal UNTIL whose English equivalent is by:
(30) purely temporal UNTIL
[[UNTILby Q]] = P t t' [Q(t')  RB (t,t')  e t'' [t'' t  P(e,t'')]],
A puzzle about until and the Present Perfect
109
where ‘RB’ stands for the right boundary of the interval introduced
by the UNTILby
It is easy to see the similarity of the purely temporal UNTIL with the durative one: they both introduce an interval with the UNTIL argument as its
right boundary. But unlike durative UNTIL, UNTILby contributes just that.
It then can be asserted that an event takes place at some subinterval t'', possibly event at the final subinterval t'. This is why a sentence like the one
below can be true even if the assignment is actually delivered at 9 am.
(31) Tha paradoso tin ergasia mexri avrio stis 9 am.
‘I will deliver the assignment by tomorrow at 9 am.’
Although UNTILby is characteristic of future contexts, there is nothing in
the semantics that implies that. We will see in section 5 that UNTILby readings are licensed also with the present perfect – and it is these readings to
be held responsible for the improvements we noted in section 1. The property of UNTILby introducing an interval makes it highly compatible with
the extended-now interval of the perfect, for which it can specify a right
boundary.
The purely temporal UNTIL is a subcomponent of the durative UNTIL,
hence it is expected that languages may employ one item to express both
meanings; Greek and German are cases in point. (Notice that these languages exclude their interval UNTIL from the scalar NPI-use, thus lexicalizing the distinction between interval and scalar UNTIL). But, as we saw,
the purely temporal is not a possible meaning of English until – possibly
because this item also encompasses a non-durative meaning.
Next, I discuss the contribution of Greek aspect because it will be crucial to the analysis of the present perfect and its interaction with UNTIL.
3.
Tense and aspect in Greek
3.1.
Tense and aspect in finite verbs forms
The Greek verb is obligatorily inflected for tense and aspect. The four possibilities for the verb grafo ‘I write’ are given in (32) (cf. Mackridge 1985,
Holton et al. 1997):
(32) a. graf-o. (INP)
write.imperf-1sg.nonpast
‘I am writing (right now).’
110
Anastasia Giannakidou
b. grap-s—o. (PNP)
write-perf.1sg.nonpast
‘I write (generally).’ [no English equivalent]
(33) a. e-graf-a. (IP)
past-write.imperf.1sg.past
‘I used to write.’
‘I was writing.’
b. e-grap-s-a. (PP)
past-write-perf.1sg.past
‘I wrote.’
The basic temporal opposition is between a morphological past, which is
marked by the prefix e- attaching to the verbal stem and exclusive inflection, and a nonpast which is signaled by the absence of the prefix e- (hence
the label nonpast), and which has its own inflection. Aspectual choice in
Greek is unavoidable in all tenses (including the future, which we ignore
here). Perfective aspect takes a bare verb meaning – bare in the sense that it
contains just the lexical entry and its argument slots, as in (34a) – and
gives back a predicate of events, as indicated in (34b):
(34) a. PFT [P] = xyet [P (y,x,e)  e  t] (Giannakidou 2002)
b. [[PFT]] = P e t [P (e)  e  t]
c. [[P (x,y)]] = xy P(y,x)
Events can take time to culminate (accomplishments), in which case t is an
interval; or they start and culminate at the same time (achievements), in
which case t is an instant. The condition ‘e  t’ expresses the relation that
e takes place at t; the same thing can be expressed by including t as an argument of the verb (I may switch back and forth between the two variants).2 At a higher level, tense contributes the information that the event is
located in the past (t<n), or in the future (n <t); or at some interval that includes n(ow), the utterance time (for more discussion along these lines see
von Stechow 2002:9-10).
The perfective nonpast does not occur as a free form, an issue discussed
in Giannakidou and Zwarts (to appear). Roughly, the reason is that PNP is
an eventive form which, however, cannot locate an event in time: it cannot
locate it in the past, since it is non-past, but it can’t locate an event in the
present or the future either, since it lacks more specific tense specification.
PNP is, then, a truly nonveridical verb form and can thus only be used with
nonveridical particles like the subjunctive, prin ‘before’, and xoris ‘without’, receiving a temporal interpretation anaphoric to that of the main
clause (see Giannakidou and Zwarts for more details).6
A puzzle about until and the Present Perfect
111
So, a typical sentence with past perfective is interpreted episodically; this
is straightforward if the verb itself is eventive, as shown in (35):
(35) a. I
Ariadni filise
ton Pavlo.
the Ariadne kissed.perf. the Paul
‘Ariadne kissed Paul.’
b. e t [kiss (Ariadne, Paul, e)  t<n  e  t]
Statives can also be modified by the perfective-- but in this case the semantics of the perfective triggers an eventive reading of the stative. This reading can be an achievement as in (36):
(36) a. I
Ariadni agapise
ton Pavlo.
the Ariadne love.perf. the Paul
‘Ariadne fell in love with Paul’
b. e t [love (Ariadne, Paul, e)  t<n  e  t]
This is the inchoative reading: there is a loving event which is both included in, and culminates at, a time t. This yields the falling in love meaning
which is seen as an instantaneous event like e.g. notice. The resulting state
of being in love is of course true of the interval that follows the falling in
love event.
Activities, like kimame ‘sleep’ often receive accomplishment readings
with the perfective. The culmination in this case would be the end of the
activity:
(37) a. I
Ariadni kimithike ja mia ora.
the Ariadne slept.perf. for an
hour
‘Ariadne slept for an hour.’
b. e t [sleep(Ariadne, e)  e  t  one-hour(t)  t<n]
Here, Ariadne was in an extended event of sleeping which lasted for an
hour. Activities can also obtain achievement meanings: this happens in the
sentence below, which contains a definite locating adverbial:
(38) a. I
Ariadni kimithike stis enia.
the Ariadne slept.perf. at nine
‘Ariadne fell asleep at 9.’
b. e t [sleep(Ariadne, e)  t<n  e  t  t=9 o’clock]
The impact perfective aspect has on stative verbs will be important later,
when we consider the perfect. But let me note here that the aspectual shifts
we observe in Greek are not at all peculiar; comparable shifts are in fact
112
Anastasia Giannakidou
quite common (see Zucchi 1998 for recent discussion and references).
Since they are systematic, it makes sense to treat these shifts as the result
of modification by aspect, rather than lexical ambiguities. At the lexical
level, the V-meaning is bare, i.e. it does not contain any event information;
it is aspect that contributes this information.3 Another way of looking at
this is to say that the bare V-meaning is actually a family of meanings, and
that each aspect triggers those meanings compatible with it, e.g. perfective
aspect will trigger the eventive meaning with statives. The choice between
the two ways of formulating seems harmless, at least for the present purposes.
Sentences with imperfective verbs in Greek are generally ambiguous
between the habitual and the progressive, as is often the case (Comrie
1967). Imperfective aspect provides an interval, indicated here as i. The
habitual generalizes over eventualities (events or states) in that interval:
(39) the habitual: generalization over eventualities (Krifka et al. 1995)
HAB [P] = i HABt [tC  t  i ;  P (,t)],
where  is a variable ranging over eventualities
(40) a. I Ariadni filuse ton Pavlo sixna.
‘Ariadne used to kiss Paul often.’
b. i [i <n  MOSTt [tC  t  i ; e kiss (Ariadne, Paul, e, t)]]
HAB has by default the quantificational force of most, which can become
explicit with an adverb like usually.
By our definition, statives may again be coerced into achievement
meanings. When this happens, they can be modified by a locating adverbial:
(41) a. Ekino to
ximona, I
Ariadne kimotan stis
9.
that the winter, the Ariadne slept
at
9
‘That winter, Ariadne used to go to bed at 9.’
b. i [i <n  that-winter(i)  MOSTt [tC  t  i ; e fallasleep(Ariadne, e, t)  t = 9 o’clock]]
In this sentence we have a generalization over Ariadne’s falling asleep
events. But activities can also receive their expected durative meaning, as
in (42), where they are modified by a for-adverbial.
(42) a. Ekino to ximona, I Ariadne kimotan 10 ores tin mera.
‘That winter, Ariadne used to sleep for 10 hours a day.’
A puzzle about until and the Present Perfect
113
b. i [i <n  that-winter(i)  MOSTt [tC  t  i ; s sleep (Ariadne, s, t)  duration (t, 10 hours)]]
The progressive creates true stative forms: it maps a state onto the interval
provided by the imperfective aspect. I will assume that the (quite simplified) semantics in (43) will suffice for our purposes (the complications of
the progressive are well-known and discussed in, at least, Bennett and Partee 1972, Bonomi 1997, Dowty 1979, Landman 1992, Zucchi 1999):
(43) the progressive
PROG [P] = s i t [(t C t  i)  P(s,t)]
(44) a. I Ariadni filuse ton Pavlo ja pende lepta.
‘Ariadne was kissing Paul for five minutes.’
b. s i [five-minutes(i)  i < n  t [(t C  t  i )  kiss (Ariadne, Paul, s, t)]]
To summarize, then, Greek verb forms, unlike English, are unambiguously
stative or eventive, depending on whether they have perfective or imperfective aspect. We see next what the tense/aspect system of Greek implies for
the participial forms which are used for the present perfect.
3.2.
The perfect participle
The perfective-imperfective contrast we observe in the finite verb forms is
removed in the non-finite ones: the gerund and the participle. Gerunds appear only in the imperfective and participles in the perfective. The former
is expected, if gerunds have meanings close to those of progressives, as is
often assumed. Gerunds are not relevant to our discussion here so I will
ignore them. I will also ignore the passive participle as it does not appear in
the perfect.
The active participle, however, is crucial because it participates in the
formation of the perfect. This participle shows no agreement and also no
tense – and it appears invariable as a perfective. In fact, it is identical to the
third person singular of the PNP paradigm. The absence of aspectual alternation in the participial domain (the passive participle is formed using the
imperfective stem only), suggests that the respective forms are interpreted
unambiguously as perfective or imperfective. Hence the perfect participle
must be interpreted as an episodic form. The semantics of the participle
grapsi ‘written’ must then be an abstract of events as in (45):
(45) a. PFT [grap] = et [write (e)  e  t]
b. [[PFT]] = P e t [P (e)  e  t]
114
Anastasia Giannakidou
As in the other cases of the PNP, the participial meaning will be filled with
locating temporal information that will come form tense. In the case of the
perfect, it will come from the auxiliary exo ‘have’.
4. The present perfect
The complications that arise with the analysis of the English present perfect are well-known in the relevant literature (McCoard 1978, Dowty 1979,
Kamp and Reyle 1993, Iatridou et al. 2001 among many others; also Rathert 1999, von Stechow 1999, 2002 for comparisons between German and
English, and Psaltou-Joycey 1993, 1994 for comparisons between English
and Greek); they hardly need be repeated here. I will only summarize the
most important characteristics of the English present perfect which will be
useful in the comparison with Greek.
4.1.
Extended now
The perfect occurs with all types of eventualities, and it introduces an extended now (XN) interval (McCoard 1978, Dowty 1979). The locus of XN
is the auxiliary verb ‘have’, which is defined as a purely temporal abstract:
(46) [[have]] = P t t’ [XN (t',t)  P (t')]
where XN = t t' [t is a final subinterval of t'] (Dowty 1979)
(In the remainder of the paper I use t for both points in time and intervals).
This semantics gives anteriority and past reference. Unlike the simple past,
however, the perfect denotes an indefinite past time; hence it is incompatible with definite adverbials like on Monday, at three o clock, yesterday.
The indefiniteness is a consequence of the XN semantics.
4.2.
Result state
Under the XN-analysis, the perfect is not, strictly speaking, a stativizer.
But it is often observed that the perfect contributes a result state (see for
instance Parsons 1990, Vlach 1993, Kamp and Reyle 1993, and references
therein). Result states become salient when we consider eventive verbs like
in (47a). Compare this sentence to its simple past version in (47b):
A puzzle about until and the Present Perfect
115
(47) a. I Ariadne exi xasi ta
klidia tis.
Ariadne
has lost the keys hers
‘Ariadne has lost her keys.’
b. I Ariadne exase
ta
klidia tis.
Ariadne
has-lost the keys hers
‘Ariadne has lost her keys.’
(47a) involves the complex semantics in (48). There is an event e of Ariadne’s losing her keys, e happened at a time t' prior to the speech time n, and
the time t' provides the left boundary (i.e. the beginning) of the XN interval. The event also signals the beginning of a result state s which abuts e,
and which holds in the rest of the XN interval (but, of course, not at t'; a
point which will become crucial in 5.1):
(48) present perfect
e t t’ [XN (t, n)  lose (Ariadne, her keys, e, t’)  LB (t’, t)  s
[e  s  lose (Ariadne, her keys, s, t)]]
(49) simple past
e t [lose (Ariadne, her keys, e, t)  t < n]
The result state is additional information, on top of the perfect XN meaning. Given that result states become available only with events, it is reasonable to assume that result states are not part of the meaning of the perfect itself, but a contribution of the lexical meaning of the verb.
Crucially, the result state is true of n, since we cannot negate it without
contradiction.7 The simple past, on the other hand, yields no conflict, since
there is no result state:
(50) * I Ariadne exi
Ariadne
has
apo
before
xasi ta
lost the
klidia tis, ala
keys hers but
ta
vrike prin
them found 3sg
ligo.
little
‘* Ariadne has lost her keys, but she found them a little while ago.’
(51) I Ariadne exase ta
klidia tis, ala ta
vrike prin
Ariadne
lost the keys hers but them found 3sg
apo
ligo.
before little
‘Ariadne lost her keys, but she found them a little while ago.’
116
Anastasia Giannakidou
As Parsons notes, once an event is culminated, its result state holds ‘forever after’ (Parsons 1990: 234; see also Kratzer 2000). The contradiction
then follows since ‘forever after’ includes n. With stative meanings we do
not get result states, but we may get the so-called universal reading which
gives a similar effect, as we see next.
4.3. Universal and existential readings with stative verbs
The English present perfect exhibits an ambiguity between a universal and
an existential reading with stative verbs. Consider, e.g., the activity below:
(52) a. Ariadne has lived in Paris.
b. (-reading)
t s [XN (t, n)  live (Ariadne, in Paris, s, t)]
c. (-reading)
t [XN (t, n)  t’ e [t’ t  live (Ariadne, in Paris, e, t’)]]
(52) says that there is a state of Ariadne’s living in Paris that extends
though now, i.e. Ariadne lives in Paris in the present time. This is the reading. This reading, which is characteristic of the English present perfect
but doesn't seem to be generally available with perfects of statives crosslinguistically, appears to be the counterpart of the result state we just mentioned in the domain of states: if a state/activity does not culminate, then it
must be true for the whole XN interval, and later on, until it culminates.
The sentence also has an existential reading in which the activity occurs
and ends at a subinterval of XN. According to this reading, Ariadne lived
in Paris for an extended period of time in the past, but she longer lives
there.4 This reading is supported with the non-contradictory continuation
below:
(53) Ariadne has lived in Paris in the past but she no longer lives there.
In the existential reading, the activity is interpreted as a culminated one; it
is thus no longer an activity but an accomplishment. Result states of accomplishments, unlike those of achievements e.g. lose my keys, are complex: they are states of both having done the activity P and no longer doing
P. In other words, the result state of an activity contains a state of the activity no longer being true. It is this state that holds of now, hence the absence
of contradiction in (53). This point will be crucial when we consider the
Greek perfect and UNTIL in 5.1.
A puzzle about until and the Present Perfect
4.4.
117
The contribution of adverbials
Adverbials are important in bringing about the various readings of the perfect. For instance, the - versus - ambiguity is resolved in the presence of
a since- adverbial, which is known to be a perfect level modifier (Dowty
1979). The sentence below only has the -reading that includes now.
(54) *Ariadne has lived in Paris since 1998, but she doesn’t live there
anymore.
A since adverbial modifies the XN of the perfect – this is why it only combines with the perfect. Von Stechow 2002 defines since as follows:
(55) [[since t]] = P t’ [P(t’)  LB (t,t’)],
where LB (t, t’) reads as ‘the left boundary of t’’.
In other words, since- adverbials provide the beginning of the XN interval.
Adverbials like from … to, and for three years retain the ambiguity and are
not perfect level. Iterative adverbials like two times bring about the existential reading, even with statives, which in this case are shifted to accomplishments:
(56) a. Ariadne has lived in Paris three times since 1984.
b. (-reading)
t [XN (t, n)  LB (1984,t)  3t’ [t’ t  live (A, in Paris, t’)]]
In the XN interval that stretches from 1984 until n, there were three subintervals during which Ariadne lived in Paris. (I suppressed here the event
information – I will take the freedom of doing so when it seems harmless.)
Likewise, with achievements we have iteration of events:
(57) a. Ariadne has lost her keys three times since Monday.
b. t [XN (t, n)  LB (Monday, t)  3t’ [t’ t  lose (Ariadne,
her keys, t’)]]
Hence iterative adverbials always bring about the -reading in an XN interval. Interestingly, this reading does not yield a unique result state that
holds ‘forever after’. Rather, iteration of events implies iteration of potential result states, one after each event – but these do not hold for forever
after, as Parsons puts it; they hold only until the next event takes place.
Such result states are thus transient and consequently trivialized.
118
Anastasia Giannakidou
4.5.
The Greek present perfect
A detailed study of the Greek present perfect and a comparison with English is found in Psaltou-Joycey (1991, 1993, 1994). Although space prevents a full review of these works, the central conclusions are reflected in
my discussion here. The Greek present perfect is similar to the English one
in one important respect: it exhibits the indefiniteness induced by XN and
it contributes a result state (recall 4.2.). The indefiniteness of XN is evidenced in the incompatibility between the Greek perfect and definite adverbials:
(58) a. * I Ariadne exi fiji stis pende.
‘* Ariadne has left at five.’
b. I Ariadne efije stis pende.
‘Ariadne left at five.’
The two perfects also differ in a way summarized by Psaltou-Joycey as
follows:
‘We can see more clearly now the different lines along which the two Perfects have developed: whereas the English Perfect has developed by indicating the link with the present (MOS) more directly, the Greek Perfect has
maintained this link rather indirectly by emphasizing the end of the situation
before the MOS. This difference ties the English Perfect more strictly to the
present while the Greek Perfect has more freedom of expression in the past.’
(Psaltou-Joycey 1993: 11)
In the discussion below I try to formalize in what sense the English, but
not the Greek, perfect is more strictly ‘tied’ to the present. Unlike English,
the Greek present perfect does not allow the universal reading with stative
verbs. This is first shown with an activity:
(59) a. ?I Ariadne exi zisi sto Parisi.
‘Ariadne has lived in Paris.’
b. (-reading)
t [XN (t, n)  t’ [t’ t  live (Ariadne, in Paris, t’)]]
To the extent that this sentence is acceptable – for most speakers, myself
included, it can be accepted only with special intonation, e.g. pitch accent
on the participle zisi – it only has the existential reading we described previously. It says that at some, in the absence of other adverbials, unspecified
time in the past Ariadne lived in Paris. Unlike English, we cannot continue
the Greek sentence by asserting that she still lives there:
A puzzle about until and the Present Perfect
119
(60) I Ariadne exi zisi sto Parisi; * malista zi akoma eki.
‘Ariadne has lived in Paris. In fact she still lives there.’
This indicates that Greek lacks the -reading with activities. For this reading, Greek employs the simple present like many other languages (see examples in Psaltou-Joycey 1991, 1993).
States like agapo ‘love’ have a complex interpretation that encompasses
a falling in love event (achievement) and a result state of love which extends through the utterance time, as indicated below. This all is in agreement with the workings of the perfective with stative verbs.
(61) Apo ti stigmi pou ematha ti exi kani ja mena, ton exo agapisi vathia.
‘Since the moment I found out what he has done for me, I have
loved him deeply.’
(61’) continued:
*Ala dhen ton agapao pja.
‘* But I don’t love him anymore.’
(62) e t [t = the moment I found out what he did for me  love-deeply
(I, him, e, t)  t’ s [XN (t’,n)  LB(t, t’)  e  s  love-deeply
(I, him, s, t’)]]
This is the typical result reading that arises with achievements; as we see,
the continuation negating this result is not possible. Iatridou et al. (2001:
208) note in passing that “statives lack the -reading”, but our observations here prove this inaccurate: the present perfect of a state does yield a
result-state reading which is similar to the -reading in that they are both
true of states holding at the XN interval which includes n. Activities like
live or sleep don’t license such results: as we saw (53), the result of an activity contains the end of the activity as a proper subpart.
Nevertheless, why don't we get the -reading with activities in Greek?
The answer is obvious. In Greek, the participle is a perfective form which
receives always an eventive interpretation. This rules out -readings. In
English, the participle, just like most verbs forms, lacks overt aspectual
marking, and is thus able to receive eventive as well as stative interpretation, which is responsible for the universal reading.
To give a simple derivation, I assume that the auxiliary exo ‘have’ introduces the XN, just like in English. Notice that the auxiliary appears only
in the imperfective in Greek, a language which otherwise makes obligatory
aspectual distinctions. The imperfective is consistent with the assumption
that exo introduces an interval:
120
Anastasia Giannakidou
(63) [[exo]]= P t t’ [XN (t',t)  P (t')]
where XN = t t' [t is a final subinterval of t']
In the present perfect, the final subinterval t is n; in the past perfect, it is
the reference time in the past. In the future perfect, it will be some time in
the future (Iatridou et al. 2001 claim that tense sets the right boundary of
XN).
Greek lacks a perfect level adverb like since. Instead edo ke and apo can
be used, glossed here as from. These can also be used with the simple past.
(64) I Ariadne {exi fiji/efije} edo ke
deka meres.
Ariadne
has left/left
here and ten days
‘It is ten days now that Ariadne (has) left.’
Given the meaning of the VP containing the participle in (66), a sentence
like (65) is interpreted as in (67):
(65) I Aridane exi xasi ta klidia tis.
‘Ariadne has lost her keys.’
(66) [[xasi ta klidia tis]]= et' [lose (Ariadne, her keys, e, t’)]
(67) [[(65)]] = t [XN (t,n)  e t' [t'  t  lose (Ariadne, her keys, e, t')
 s [e  s  lose (Ariadne, her keys, s, t)]]]
If we add an adverbial like tris fores ‘three times’, the result state is lost,
and we are just counting the losing-events in the XN interval:
(68) I Ariadne exi xasi ta klidia tis tris fores apo ti deftera.
‘Ariadne has lost her keys three times since Monday.’
(69) [[(65)]] = t [XN (t,n)  LB (Monday, t)  3e 3t' [t'  t  lose
(Ariadne, her keys, e, t')]]
The interpretation of a perfect activity follows the same format. Recall first
that a bare present perfect is a little odd with an activity unless it has special intonation – cf. example (59). Because we employ a perfective participle, the stative meaning shifts to an accomplishment: the activity of living
in Paris culminates at some time t'. We thus end up with an existential semantics for activities, unlike in English (where, as we said, the participle is
unspecified for aspect):
(70) I Ariadne exi zisi sto Parisi 3 fores apo to 1995.
A puzzle about until and the Present Perfect
121
‘Ariadne has lived in Paris three times since 1995.’
(71) [[zisi sto Parisi]] = et’ [live (Ariadne, in Paris, e, t’)]
(72) [[(70)]]= t [XN (t,n)  LB (1995, t)  3e 3t' [ t'  t  live (Ariadne, in Paris, e, t')]]
To sum up, the Greek present perfect is existential because a perfective
participle is employed.8 This explains why it is less tied to the present time,
and allows more ‘freedom of expression’ in the past, as Psaltou-Joycey
puts it. We are now in position to explain how this property of the Greek
perfect is responsible for blocking until-phrases in the present perfect.
5.
The present perfect and until
5.1.
Back to the puzzle: the core cases
Consider our initial data, repeated here:
(73) * I Ariadne exi zisi
sto Parisi mexri
tora.
the Ariadne has lived in
Paris
until
now
‘?Ariadne has lived in Paris until now.’
(74) * I Ariadne exi xasi ta
klidia tis
mexri tora.
the Ariadne has lost the keys hers until now
‘*Ariadne has lost her keys until now.’
Recall that the sentences improve with iterative adverbials and negation;
also that the oddity is weaker in English with the stative verb. Take (73)
first. Mexri is the durative UNTIL with the semantics below:
(75) durative mexri
[[mexri Q]]= P s tt’ [P(s,t)  Q(t')  RB (t,t')  t’’ [(t''t  t''
t’)  P(s,t'')]], where RB stands for ‘right boundary’
(76) [[mexri tora]] = P s tt’ [P(s,t)  now (t')  RB (t,t')  t’’ [(t''t
 t'' t’)  P(s,t'')]]
The perfect exi xasi ta klidia tis has the by now familiar semantics in (77):
(77) [[(65)]] = n t [XN (t,n)  e t' [t'  t  LB (t’,t)  lose (Ariadne,
her keys, e, t')  s [e  s  lose (Ariadne, her keys, s, t)]]]
This semantics says that at some time t’, which is the left boundary of XN,
122
Anastasia Giannakidou
there was event of losing the keys and this event resulted in a state s. When
we combine this meaning with the meaning of mexri tora, the mexri interval is mapped onto the XN interval and the argument of mexri contributes
the right boundary of XN. We thus get the following meaning for the
whole sentence:
(78) [[(73)]] = t [XN (t,n)  e t' [ t'  t  LB(t’,t)  lose (Ariadne, her
keys, e, t')  s [e  s  lose (Ariadne, her keys, s, t)  t'' [t''t 
lose (Ariadne, her keys, s, t'']]]]
This semantics creates a conflict because of the final universal clause:
mexri requires that the state hold in all times t’’ of the XN interval. These
times t’’ include, of course, the initial time t’. At that time t’, however, we
do not have a state, but an event: the event of losing the keys. Hence the
durative UNTIL condition is not met and the sentence leads to a contradiction.
Consider now what happens if we add ‘tris fores’ three times.
(79) I Ariadne exi xasi ta klidia tis tris fores mexri tora.
‘* Ariadne has lost her keys three times until now.’
The sentence is good in Greek (but still unacceptable in English, and it will
soon become obvious why). The improvement is due to the effect of tris
fores.
(80) [[(79)]] = t [XN (t,n)  3e 3t' [t'  t  lose (A, her keys, e, t')]]
Here we have three events of losing the keys that happened at subintervals
of the XN introduced by the perfect and identified by the mexri phrase.
Crucially, the semantics in (80) does not lead to a contradiction; rather, in
the absence of a result state, it triggers the purely temporal meaning of
mexri, in which mexri provides just an interval on which events can be located:
(81) purely temporal UNTIL
[[UNTILby Q]]= P t t' [Q(t')  RB (t,t')  e t'' [t'' t  P(e,t'')]],
where ‘RB’ stands for the right boundary of the interval introduced
by the UNTILby
So (79) is fine because there is no conflict between purely temporal
UNTIL and the existential perfect. That we have this meaning here is confirmed by the fact that the corresponding English sentence is good only
with by and not with until, which as we noted in section 2, lacks the purely
A puzzle about until and the Present Perfect
123
temporal reading:
(82) a. * Ariadne has lost her keys three times until now.
b. Ariadne has lost her keys three times by now.
Consider now the activity in (73), repeated here as (83). The participle has
the accomplishment reading we noted. So there was an activity of living in
Paris which ended at some time t’ in the past. This time t’ is included in the
XN interval. Again, the XN interval is mapped onto the boundaries provided by the UNTIL-phrase. The result in given in (85):
(83) * I Ariadne exi zisi sto Parisi mexri tora.
‘? Ariadne has lived in Paris until now.’
(84) [[zisi sto Parisi]] = et’ [live (Ariadne, in Paris, e,t')]
(85)
[[(83)]] = t [XN (t,n)  e t' [t'  t  LB(t’,t)  live (Ariadne, in
Paris, e, t')  s [e  s  live (Ariadne, in Paris, s)  t'' [t’’  t
 live (Ariadne, in Paris, s,t')]]]]
The italicized part spells out the result state of the accomplishment. This
state creates the anomaly when combined with durative UNTIL: since the
event of living culminated, the result state contains no living in Paris. This
leads to a contradiction: during the XN interval we have both a living in
Paris and a not living in it, hence the universal condition of durative mexri
is again not met.
As expected, the purely temporal reading that becomes salient with iterative adverbials can rescue the sentence:
(86) I Ariadne exi zisi sto Parisi 3 fores mexri tora.
‘Ariadne has lived in Paris 3 times {until/by} now.’
This is a good meaning, as we see below.
(87) [[(86)]]= t [XN (t,n)  3e 3t' [t'  t  live (Ariadne, in Paris,
e,t')]]
This meaning is compatible with purely temporal UNTIL, in which the
UNTIL phrase simply serves as the identifier of the right boundary of the
XN interval.
It is worth noting that the corresponding English sentence (83) is odd
for a different reason. Since the participle also receives a stative interpretation, the sentence allows for the universal reading of the perfect. In this
124
Anastasia Giannakidou
reading, the state of living in Paris would have to hold at all subintervals of
XN including now. As we noted in section 1, however, durative until gives
a strong implicature of change of state at the time denoted by the untilargument. It is the presence of this implicature that creates a conflict: in the
absence of additional information, it creates the expectation that the state
of living ends at now, which is at odds with the universal perfect that it
holds at now. The fact that we have a conflict with an implicature explains
why the oddity in English is rather weak.9
Apart from iterative adverbials, the odd sentences are saved by from...
to adverbials and negation. I consider these cases next.
5.2.
From… to-constructions10
Stative verbs become fine in apo... mexri, lit. from... until, ‘from... to’ constructions – which in Greek involve mexri in the place of to. Notice that
eventives, i.e. achievements, are still bad.
(88) a. I Ariadne exi kimithi apo tis 3 i ora mexri tora.
‘Ariadne (*has) slept from 3 o’clock to now.’
b. I Ariadne exi zisi sto Parisi apo to 1982 mexri tora.
‘Ariadne (*has) lived in Paris from 1982 to now.’
c. *I Ariadne exi xasi ta klidia tis apo tis xthes mexri tora.
‘Ariadne (*has) lost her keys from yesterday to now.’
An additional interesting feature here is that from... to adverbials exclude
the present perfect in English – that could be due to the definiteness constraint which seems to be stronger in English than it is in Greek.
Why are the a,b sentences above good? They are good because they
trigger the existential reading of the perfect – there was a living in Paris
and a sleeping which were true at some subinterval t' of the XN interval.
The from to construction provides the right and left boundary of the relevant subinterval:
(89) [[apo XP… mexri YP]] = t [LB (t) = XP  LB (t) = YP]
(90) [[(88a)]] = t [XN (t,n)  e t' [t'  t  LB (t') = 3 o'clock 
RB (t') = now  sleep (Ariadne, e,t')]]
(91) [[(88b)]] = t [XN (t,n)  e t' [t'  t  LB (t') =1982  RB (t') =
now  live (Ariadne, in Paris, e,t’)]]
Since from... to adverbials introduce intervals, it is predicted that they will
A puzzle about until and the Present Perfect
125
have no effect on achievements because achievements have no duration.
The contrast between (88a,b) and (88c) thus follows directly.
5.3.
Negation
Finally, consider the effect of negation:
(92) a. I Ariadne dhen exi kimithi mexri {tora/ta mesanixta}.
Ariadne
has not slept
until now
‘?Ariadne has not slept until midnight.’
b. I Ariadne dhen exi zisi sto Parisi mexri {tora/to 1984}.
‘?Ariadne has not lived in Paris until 1984.’
c. I Ariadne dhen exi xasi ta klidia tis mexri tora (= akomi).
‘Ariadne has not lost her keys yet.’
As we see, negation uniformly improves the status of the sentences in
Greek. The effect is visible in English too, although it is not as uniform as
it is in Greek, e.g. stative verbs remain relatively awkward. The improvement should not be viewed as an argument for a stative analysis of negation – in Giannakidou 2002 I offer extensive arguments as to why this
analysis is wrong; and as we see here, the improvement is not uniform in
English. Rather, negation rescues the sentences because it either negates
that there was an event, or that the event culminated at the time indicated
by the UNTIL argument. In either case, negation has an improving effect
because it licenses the iterative reading.11
Both readings of negation are visible in (92a). Consider first the version
with mexri tora, which, as indicated, is equivalent to akomi ‘yet’. The sentence has the reading below:
(93) [[(92a mexri tora)]] = t [XN (t,n)  e t' [t'  t  sleep (Ariadne,
e, t')]]
As we noted previously, the perfective of sleep here is an achievement: a
falling asleep event which starts and ends at the same time t'. The separation of the event and its culmination is thus trivial, and certainly not visible
for negation. In this reading negation takes in its focus both the event and
its culmination, and the sentence says that in the XN interval that ends
now, there was no event of Ariadne's falling asleep.
In the version with mexri ta mesanixta, the sentence has the following
interpretation (where CUL is Parson’s culmination predicate):
(94) [[(92a mexri ta mesanixta)]] = t [XN (t, n)   e t' [t'  t  RB
(t', midnight)  sleep (Ariadne, e, t')  CUL (e, midnight)]].
126
Anastasia Giannakidou
In this sentence, the UNTIL phrase contributes another time, midnight,
which allows the creation of a subinterval going through midnight, inside
the XN interval-- notice that although midnight could in principle be a time
after now, the XN information of the perfect restricts reference to times
that precede now. The perfective sleep can thus be interpreted as an accomplishment which occupies that subinterval and culminates at its right
boundary, i.e. midnight. Negation then applies to negate precisely this: it is
asserted that during the XN interval there has not been a sleeping event
that culminated at the UNTIL time. So, there may have been a sleeping
event, but it didn’t culminate at midnight-- it might have culminated at a
time prior or posterior to midnight. Negation in this reading thus takes the
UNTIL subinterval in its focus.12
Two questions arise: first, why is this second reading possible when
UNTIL contains a time other than now, but not with now? Second, why is
the English sentence not as good in this reading? It is not hard to see what
the answers should be. When the UNTIL time is now, there is a perfect
mapping between XN and the UNTIL interval. This means that there is no
subinterval upon which the activity can be mapped, which explains why an
activity is necessarily interpreted as an achievement (same thing with perfectives and NPI-UNTIL, where again an perfective activity is interpreted
as an achievement, recall our discussion in section 2.2.). The subinterval
enables an activity to map onto it, and then perfective aspect says that that
the activity ends when the subinterval ends. Negation can then negate this
bit of information, which is available only in case a subinterval is available.
If this explanation is correct, then we expect some improvement with an
UNTIL phrase other than now, even without negation. This is actually the
case:
(95) ?I Ariadne exi kimithi mexri ta mesanixta.
‘?Ariadne has slept until midnight.’
This sentence, though not impeccable, is still much better than its corresponding version with mexri tora. Perhaps this reasoning can help us understand the improvement we witness with the past perfect (fn. 1).
Regarding the second question, in English there is no improvement with
a creation of a subinterval with an endpoint prior to now, because doing so
creates a conflict with the -perfect reading of the stative perfect. So again
the individual differences of the perfects in the two languages are responsible for the distinct effects of UNTIL adverbials.
The explanation of (92a) carries over to (92b) without any further adjustments. It is also predicted that an achievement as in (92c) will only
have the reading with negation negating both the event and its culmination,
A puzzle about until and the Present Perfect
127
since they both happen at the same time t'. This is exactly the reading we
have with (92c):
(96) [[(92c)]] = t [XN (t,n)  e t' [ t'  t  lose (Ariadne, her keys,
e,t')]]
The English sentence is likewise good because the perfect is existential,
and there is no conflict with a potential universal reading.
Hence, this analysis explains successfully the improvement of perfects
and UNTIL adverbials with negation, and at the same time makes the distinct meanings of perfect stative verbs follow from it, rather than stipulating them.
6. Conclusion
In this paper I discussed the interaction between the present perfect and
UNTIL adverbials in English and Greek. I showed that when UNTIL and
the present perfect don’t combine well, this is due to a clash between the
semantics of durative UNTIL, which requires that a state extend through
all subintervals introduced by it, and the perfect, which contains both an
event and a result state and does not satisfy this requirement. When the
perfect and UNTIL do combine well, this is because UNTIL has a purely
temporal meaning and an existential perfect is compatible with this meaning. The aspectual information coming from the participle has been important in trying to account for the differences between Greek and English.
In Greek, where the participle is perfective, the perfect always contains an
event, even with stative verbs. But in English, where the participle has no
overt aspect, purely stative interpretations are also allowed – this is why
English, but not Greek statives allow for the universal reading. Unlike
Greek, in English, it is the licensing of this reading that blocks durative
UNTIL in the present perfect with statives. The particular differences
among the two languages thus followed compositionally from the central
difference in the forms the two languages employ for the perfect.
Ackowledgements
Many thanks to Arnim von Stechow for his most helpful written comments, and to the editors of this volume as a whole for giving me the final
push to actually write down this paper. I also wish to thank the audience of
the Workshop on the Perfect at the University of Thessaloniki, especially
Elena Anagnostopoulou, Graham Katz, and Monika Rathert for their com-
128
Anastasia Giannakidou
ments; Jason Merchant and Angeliki Psaltou-Joycey for discussion and
judgments. Thanks to Angeliki also for promptly sending me her material
on the perfect. This research is funded by the Royal Dutch Academy of
Sciences (KNAW), whose contribution is hereby gratefully acknowledged.
Notes
1. It is a little unclear what the status of UNTIL phrases is with the past perfect.
For some speakers, myself included, there is partial improvement with stative
verbs, but for others, stative and eventive verbs are uniformly bad with the past
perfect and UNTIL, just like their present perfect counterparts.
(i) a. (?) I Ariadne ixe zisi sto Parisi mexri to 1984.
‘Ariadne had lived in Paris until 1984.’
b. * I Ariadne ixe xasi ta klidia tis mexri ta mesanixta.
* ‘Ariadne had lost her keys until midnight.’
I will not consider these subtleties here, and assume that what I say in this paper
about the present perfect roughly carries over, without major adjust-ments, to
the past perfect as well. In section 5.4, I revisit the interaction of UNTILadverbials with arguments other than now and show that my analysis actually
predicts improvement with stative perfects in this case.
2. Karttunen in fact argues for a systematic ambiguity of statives, but the inchoative meaning can be seen as a result of modification by perfective aspect, which
is indeed an option with statives, as we see below in Greek. What we see overtly in Greek can be argued to happen covertly in English.
3. There is no telicity involved in the Greek perfective forms, unlike e.g. Slavic
aspect, as we see in (i).
(i) Diavasa to vivlio, ala to pempto kefaleo dhen to thimame; isos na to perasa
poli grigora, i bori na min to diavasa ke katholu.
‘I read this book, but I don't remember chapter 5; perhaps I went through it
very quickly, or I haven’t even read it at all.’
Greek aspect is different from Slavic in another important respect: in Slavic, aspect applies lower, at the level of Aktionsart (hence it is more lexical in this
sense). Greek aspect modifies verb meaning at a higher level. The absence of
telicity is obviously related to this fact.
4. The behavior of the PNP is related to the well-known fact (Giorgi and Pianesi
1997) that, even in languages with no obligatory aspect such as English, eventives do not allow episodic interpretations in the present-- episodic as referring
to a single event (Giannakidou 2001):
(i) Bob kisses Mary. = Bob has the habit of kissing Mary.
NOT: There is an event of Bob kissing Mary right now.
5. Hence speaking about statives or eventives seems redundant in this context, as
V-meanings on their own do not contain states or events.
A puzzle about until and the Present Perfect
129
6. This contrasts with the claim in Rathert 2000 that the speech time is not included in the XN interval. The contradiction we observe here also suggests that the
inclusion of n in the XN cannot be just a ‘strong conversational implicature’, as
argued there.
7. Kamp and Reyle 1993 propose three rules for perfect formation in order to capture the attested ambiguity. These readings, however, seem to all follow from
the extended now semantics of the perfect (as suggested, for instance, in Iatridou et al 2001 and von Stechow 2002), and this is the position I adopt here.
8. See also the article of Iatridou et al. in this volume.
9. Note in this connection that until is actually well formed with stative perfects in
some cases. The following are attested data from a web search, brought to my
attention by Arnim von Stechow:
(i)
Until recently, most NLP research has been focused on the understanding of a larger text.
(ii)
Until now, Apple has been the industry leader.
(iii) Founded in 1995, mySQL AB has until now been supported exclusively
through internal investments, and it has been profitable since 1996.
If, as I argue here, the oddity of until with stative perfects is not semantic, then
examples like the above, which involve stative verbs, are not surprising. Additionally, improvement occurs when the until adverbial is preposed. Preposing
weakens the change of state implicature considerably (see Giannakidou 2002
for data and more details).
10. See also the article of Veloudis in this volume.
11. Notice that the sentences are bad on the eventive NPI-reading: they do not
entail that Ariadne actually fell asleep at the UNTIL-time. For this, simple
past must be used instead.
12. This is, quite unexpectedly, the reading we otherwise observe in Greek with
mexri and perfective simple past, as I showed in Giannakidou 2002. This
reading is the distinct NPI-reading licensed with para monon that we saw in
section 2.2.
References
Bennett, Michael and Barbara Partee
1972
Toward a Logic for Tense and Aspect in English. IULC, Indiana
University.
Bonomi, Andrea
1997
The progressive and the structure of events. Journal of Semantics 14:
173-205.
Comrie, Bernard
1967
Aspect. Cambridge: Cambridge University Press.
Declerk, Renaat
1995
The problem of not… until. Linguistics 33: 51-98.
130
Anastasia Giannakidou
Dowty, David
1979
Word Meaning and Montague Grammar. Dordrecht: Reidel.
Giannakidou, Anastasia
1998
Polarity Sensitivity as (Non)veridical Dependency. Amsterdam/
Philadelphia: John Benjamins.
2001
The meaning of free choice. Linguistics and Philosophy 24: 659-735.
2002
UNTIL crosslinguistically, aspect, and negation: a novel argument
for two untils. To appear in the Proceedings of SALT 12.
Giannakidou, Anastasia and Frans Zwarts
(To appear) Temporal, aspectual operators and (non) veridicality. To appear in
Tense, Aspect and Mood Selection, Alessandra Giorgi, James Higginbotham, and Fabio Pianesi (eds.), Oxford: Oxford University
Press.
Hitzeman, Janet
1991
Aspect and adverbials. Semantics and Linguistic Theory 1: 107-127.
Holton, David, Peter Mackridge, and Irene Philippaki-Warburton
1997
Greek: A comprehensive grammar of the modern language. London:
Routledge.
Horn, Lawrence
1989
A Natural History of Negation. Chicago: University of Chicago
Press.
Iatridou, Sabine, Elena Anagnostopoulou, and Roumyana Izvorsky
2001
Observations about the form and meaning of the perfect. In Ken
Hale: A Life in Language, Michael Kenstowicz (ed.), 189-238. Cambridge, MA: MIT Press. (reprinted in this volume)
Kamp, Hans
1968
Tense and the Theory of of Linear Order. PhD thesis, UCLA.
Kamp, Hans and Uwe Reyle
1993
From Discourse to Logic. Dordrecht: Kluwer.
Karttunen, Lauri
1974
Until. Proceedings of Chicago Linguistic Society 10: 284-297.
Kratzer, Angelika
2000
Building statives. Proceedings of Berkeley Lingusitic Society.
Krifka, Manfred
1989
Nominal reference, temporal constitution, and quantification in event
semantics. In Semantics and Contextual Expressions, Renate
Bartsch, John van Benthem and Peter van Emde Boas (eds.), 75-115.
Dordrecht: Foris.
Krifka, Manfred, F. Jeffry Pelletier, Gregory Carlson, Alice ter Meulen, Gennaro
Chierchia, and Godehard Link
1995
Genericity: An introduction. In The Generic Book, Gregory Carlson
and F. Jeffry Pelletier (eds.), 1-124. Chicago: Chicago University
Press.
A puzzle about until and the Present Perfect
131
Landman, Fred
1992
The progressive. Natural Language Semantics 1: 1-33.
McCoard, Robert W.
1978
The English Perfect: Tense choice and pragmatic inferences. Amsterdam: North-Holland.
Mackridge, Peter
1985
The Modern Greek Language. Oxford: Oxford University Press.
Mittwoch, Anita
1977
Negative sentences with until. Proceedings of Chicago Linguistic
Society 13: 410-417.
Newton, Brian and Ioannis Veloudis
1980
Necessity, obligation, and the Greek verbal aspect. Lingua 50: 25-43.
Parsons, Terence
1990
Events in the Semantics of English. MIT Press: Cambridge, MA.
Psaltou-Joycey, Angeliki
1991
The Temporal, Aspectual, and Pragmatic Functions of the Perfect in
Modern Greek. PhD thesis. Aristotle University of Thessaloniki.
1993
Disambiguating temporal phrases: the apo and since: comparison.
Proceedings of the 6th Symposium on the Description and/or Comparison of English and Greek, 1-12. Thessaloniki.
1994
Temporal Intervals and Situations in the Perfect. In Themes in Greek
Linguistics, I. Phillipaki-Warburton, K. Nikolaidis, and M. Sifianou
(eds.), 153-160. Amsterdam/Philadelphia: John Benjamins.
132
Anastasia Giannakidou
Rathert, Monika
2000
A semantics for durative adverbials. In Proceedings of ConSOLE
VIII, Christine Czinglar et al. (eds.), 267-280.
von Stechow, Arnim
1999
Eine Erweiterte Extended-Now Theorie für Perfekt und Futur.
Zeitschrift fur Literaturwissenschaft und Linguistik 113: 86-118.
2002
German seit 'since' and the ambiguity of the German Perfect. In More
than words. A Festschrift für Dieter Wunderlich, Ingrid Kaufmann
and Barbara Stiebels (eds.), 393-432. Berlin: Akademie Verlag.
de Swart, Henriette
1996
Meaning and use of not… until. Journal of Semantics 13: 221-263.
Vlach, Frank
1993
Temporal adverbials, tenses and the perfect. Linguistics and Philosophy 19: 271-292.
Zucchi, Sandro
1998
Aspect shift. In Events and Grammar, Susan Rothstein (ed.), 347370. Dordrecht: Kluwer.
1999
Incomplete events, intensionality, and imperfective aspect. Natural
Language Semantics 7: 179-215.
Zwarts, Frans
1995
Nonveridical contexts. Linguistic Analysis 25 (3-4): 286-312.
A puzzle about until and the Present Perfect
A
antiveridicality 106–8
Aspect
imperfective 112
perfective 110
D
de Swart, Henriette 104
Dowty, David 114
durative adverbs
from...to 124
mexri (‘until’) 101, 103–9
since 117–18
until 101, 103–9
F
Frequency adverbs
tris fores (‘three times’) 101–2
tris fores 'three times' 122
G
Giannakidou, Anastasia 110
H
habituality 112
Hitzeman, Janet 104
Horn, Lawrence 105
K
Karttunen, Lauri 106
133
M
McCoard, Robert W. 114
N
Negation
in Greek 101–2, 106–8, 125
NPI-until 106–8
P
past participle
in Greek 113–14
Perfect
Extended-Now-Theory 114, 118–21
in Greek 118–21
progressive 113
R
Resultativity 114
S
Stechow, Arnim von 110
U
universal-existential ambiguities
complex 117
Z
Zucchi, Alexandro 112