Download lin3021_lecture10

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts

Latin syntax wikipedia, lookup

Pipil grammar wikipedia, lookup

Spanish grammar wikipedia, lookup

Georgian grammar wikipedia, lookup

Polish grammar wikipedia, lookup

Turkish grammar wikipedia, lookup

Serbo-Croatian grammar wikipedia, lookup

Lexical semantics wikipedia, lookup

Yiddish grammar wikipedia, lookup

Italian grammar wikipedia, lookup

Portuguese grammar wikipedia, lookup

Ancient Greek grammar wikipedia, lookup

Modern Hebrew grammar wikipedia, lookup

Esperanto grammar wikipedia, lookup

French grammar wikipedia, lookup

Japanese grammar wikipedia, lookup

Vietnamese grammar wikipedia, lookup

Transcript
LIN3021 Formal Semantics
Lecture 10
Albert Gatt
In this lecture
 We shift our focus to events, and ask:
 How should we think of events in natural language?
 How do we treat semantic issues related to events, specifically:
 The nature of events and event modification
 Time and tense
 Aspect and event structure
 Modality and possibility
What is an event?
 So far, we’ve thought of verbs (transitive and intransitive) as
predicates and relations:


 [[sleep]] = x sleep (x)
 [[eat]] = xy eat ( x, y)


 This puts them on a par with nouns:
 [[man]] =
xman(x)
Predicates of what?
 But notice:
 The argument of the semantic representation of the noun is an
individual (an “entity”)
 This is evident in a typical predicative construction:
 Sam is a man xman( x)(s)  man(s)
 And if we add premodifiers, we assume they’re
<<e,t>,<e,t>>, i.e. Take a noun predicate and return a
complex noun predicate:
 Sam is a tall man
[[tall ]]  NxN ( x)  tall ( x)
[[ tall man ]]  NxN ( x)  tall ( x)(man)  xman( x)  tall ( x)
[[Sam is a tall man]]  xman( x)  tall ( x)(s)  man(s)  tall (s)
Predicates of what?
 But what about events?
 With Sam eats, it seems to be similar to the nominal case: eat(s)
 But what about Sam eats quickly? What is the argument of quickly? Do
we want eat(s) & quick(s)?
 Many semanticists argue that verbs actually denote an implicit
event argument.
 I ate a meat pie roughly means:
 There is an event, and this event is an eating event, and the event involves me as agent,
and a meat pie as patient
 In other words, rather than a relation between two things (an eater
and an eatee), we might think of this as a relation between three
things (an event, and the eater and eatee involved in it)
A preliminary example
Strange goings on! Jones did it slowly, deliberately, in the bathroom,
with a knife, at midnight.What he did was butter a piece of toast.
Davidson, 1980 [1967], p. 105
 Some questions/observations:
 What does it refer to in that sentence? (The buttering event?)
 We have the intuition that slowly, deliberately, in the bathroom etc
modify the same buttering event involving Jones and the toast.
 There is a difference between:
 Adverbs like slowly, which actually modify the event itself
 Phrases like in the bathroom, with a knife etc, which seem to add arguments
to the event in addition to Jones and the toast.
The phenomena we want to look at
Strange goings on! Jones did it slowly, deliberately, in the bathroom,
with a knife, at midnight.What he did was butter a piece of toast.
Davidson, 1980 [1967], p. 105
 We’re going to consider three classes of phenomena:
Thematic roles
2. Event modifiers and their relationship to nominal modifiers
3. Event reference and nominalisation
1.
Part 1
Evidence from thematic roles
Thematic roles
 We’ve encountered thematic roles before...
 Roughly, these are semantic categories that specify the roles
of arguments of events:
 Agent, patient, location, instrument etc
 Their main theoretical function is to allow us to:
 Categorise the arguments of verbs (and of some other
predicates)
 Make generalisations about how they combine with predicates
An example
John buttered the toast.
The toast was buttered by John.
 John = the agent
 The toast = the patient
 The roles stay the same whether it’s active or passive.
 Notice that we don’t have a verb (say, sbuttered) which would
be synonymous with buttered-in-the-passive and would reverse
the roles:
 The toast sbutttered John (=the toast was buttered by John).
 Why?
Generalisations
 The apparent non-existence of verbs like sbuttered, with reversed
agent/patient roles, suggests that language:
 is biased towards having specific properties for agents and patients
(John is a more likely agent of a buttering event)
 maintains these roles even if the event is conceptualised in terms of
the reverse relation (as with the passive)
 Typically:
1.
2.
3.
Agent maps to subject
If there is no agent, but there is a patient, then patient maps to
subject
If there is both an agent and a patient, then (1) applies, and patient
maps to object.
An aside on Maltese and related
languages
 Superficially, we do seem to find languages that allow the
equivalent of “sbuttered”, i.e.:
 Take a verb with Agent and Patient
 Add something to it to morphologically to reverse the roles.
 Pawlu qatel raġel
 Raġel inqatel minn Pawlu
 But note:
 The roles are not being reversed (Pawlu is still agent)
 The verb nqatel is arguably marked with respect to the base
form qatel.
 This suggests that the “passive” is not the basic form.
Agent and patient of what?
John buttered the toast.
The toast was buttered by John.
 Notice that we have the same event viewed in different ways:
It’s the same action in both cases (a buttering)
 It involves an agent and a patient in both cases.
 We might take this as prima facie evidence for the argument
that:
 There is a single event underlying these two sentences (call it e)
 The agent and patient roles are relations between the event and an
individual. E.g.
 AGENT(e) = John
 PATIENT(e) = the toast
Things to note
 If we adopt this view, then we no longer think of butter as a 2-place
predicate involving 2 arguments. We would think of butter as also
involving an implicit event argument.
 Semantically, it might look like this:
ebutter(e)  AG(e)  j  PAT (e)  t 
 In other words, we think of event sentences as:
 Implicitly involving an event argument
 Relating the explicit arguments directly to the event itself
 So we no longer analyse this as a 2-place predicate along the lines
of butter(j,t).
Some further evidence
John buttered the toast.
John buttered the toast with a knife.
 Here, we seem to have introduced additional arguments!
 If we assume that [[butter]] is a two-place predicate in the
first example, what happens in the second case?
 Do we want to have to say that butter is ambiguous?
 2-place butter: agent, patient
 3-place butter: agent, patient, instrument
 ...
Some further evidence
John buttered the toast.
John buttered the toast with a knife.
 Our event-based analysis would allow us to avoid this kind of
argument.
 In the second sentence, all we’ve done is introduce a third
role (instrument), but it’s still the same butter predicate (only
we’ve added an optional argument):
ebutter(e)  AG(e)  j  PAT (e)  t  INSTR(e)  k 
A complication
(1) Sam hit the table with a hammer
(2) Sam hit the hammer against the table.
 Do we have the same event here? Intuitively, perhaps, we do.
 Given an actual event in a world/model, could we describe it in either of these
two ways?
 But the thematic roles seem different in the two cases: Sam does
something to the table in (1), but to the hammer in (2). By our earlier
generalisations, table is the patient in (1), but hammer is patient in (2).
 If we we want to keep the traditional thematic role analysis, we’d be forced
to conclude that hit takes different arguments in (1) and (2).
 But then, we can’t really say it’s the same event! The two have different
participants.
 Two different entries for hit?
Dowty’s (1990) theory
Sam hit the table with a hammer
Sam hit the hammer against the table.
 Perhaps we should instead think of these as “prototypes”:
 Proto-agent entailments: volitional, sentient, causer, moves, exists
independently
 Proto-patient entailments: undergoes change, changes portion by
portion, causally affected, stationary, doesn’t exist independently.
 Under this view:
 Sam is more agent-like
 Both hammer and table are roughly equally patient-like; hence we have
a choice about which becomes the object.
 Hit is the same in the two cases; it’s just that we have an option about
which argument to map to the patient role.
Interim summary
 We’ve adduced some evidence for events involving an
implicit event argument:
 The same events can be conceptualised in different ways (e.g.
active/passive) but retain the same thematic roles
 The thematic roles relate the event to its arguments
 Additional roles can be introduced (often through PPs) and
these are just conjoined to the whole interpretation.
 We’ve also seen some reasons for thinking of roles as being
“prototypical”. This helps in maintaining the event analysis,
and dealing with variable mappings to the syntax.
Interim summary cont/d
 Things to note about the analysis:
 There is an implicit event argument
 Thematic roles are relations between the event and its
arguments
 The event argument is existentially bound (there is an event...)
Part 2
Event modifiers
Our original example
Jones buttered the toast slowly, in the bathroom, with a knife, at
midnight.
 Here we have:
 Prepositional phrases introducing “extra” or “optional”
arguments (instrument, location)
 Manner adverb (slowly)
 Time adverbial (at midnight)
 These are all “event” modifiers – they add some more
information to the basic event of John buttering the toast.
Questions
 How should event modifiers be analysed?
 Do verb modifiers have anything in common with adjective
modifiers?
 Can we have a single, unified theory?
Some observations
 Event modifiers exhibit two interesting phenomena, which, following
Landman (2000), we’ll call permute and drop
 Permute:
1. John buttered the toast slowly, in the bathroom, with a knife.
2. John buttered the toast in the bathroom, slowly, with a knife.
 We can permute the order of modifiers, and this seems to make no
difference to the interpretation. (One entails the other)
 Drop:
1. John buttered the toast slowly, in the bathroom, with a knife.
2. John buttered the toast in the bathroom with a knife.
 If (1) is true, then it entails (2).
 In a sentence S with n modifiers, if we form a new sentence S’ from S by
dropping one or more of these modifiers, then S  S’.
The parallel with adjectives
 Adjectival premodifiers of nouns seem to exhibit the same
properties.
 Permute:
 John is a dark, thirty-something, Maltese man
 Therefore, John is a thirty-something, dark, Maltese man.
 Drop:
 John is a dark, thirty-something, Maltese man
 Therefore, John is a thirty-something, Maltese man
More parallels with adjectives
 Adjectives:
 If we set things up so that we’re talking about the same individual,
then inferences of the following sort seem to be ok:
 John is a dark man.
 John is a basketball player.
 Therefore, John is a dark basketball player.
 But we’ve noted that some adjectives are exceptions to this (so-
called “intensional” adjectives):
 John is a former president.
 John is a basketball player.
 *Therefore, John is a former basketball player.
More parallels with adjectives
 Adverbs:
 If we set things up so that we’re talking about the same event, then
inferences of the following sort seem to be ok:
 Caesar stabbed Brutus with a knife.
 Caesar killed Brutus.
 Therefore, if the event here is the same one: Caesar killed Brutus with a knife.
 But some adverbs resist this (those related to belief, state of mind etc):
 Caesar stabbed Brutus intentionally.
 Caesar killed Brutus.
 *Therefore, if the event here is the same one: Caesar killed Brutus intentionally.
 NB: it is crucial that we assume that the event is the same one! (Just as
it’s crucial in the adjective examples that we’re talking about the same
individual)
Can we exploit the parallels?
 Except for a particular class of adjectives, we find remarkable
flexibility in premodification of nouns, given permute and
drop.
 Recall that, in our earlier analysis, we distinguished between:
 Predicative adjectives (John is tall) which are just properties:
tall(j). Type: <e,t>
 Attributive adjectives (John is a tall man) which take a noun
predicate (<e,t>) and return a complex noun predicate (also of
type <e,t>). So the attributives have type <<e,t>,<e,t>>
Attributives
 Attributive adjectives (John is a tall man) are of type <<e,t>,<e,t>>, so
they’re not simple predicates.
 What is their relationship to their predicative counterparts (John is tall)?
These have simpler types (<e,t>) and so seem more basic.
 Let A be an attributive adjective, and Ap be its predicative counterpart.
We might analyse the attributive as:

A  Nx N ( x)  Ap ( x)

 This says: the attributive adjective takes a noun meaning and applies this
to x. It also applies the basic adjectival meaning corresponding to A, to
the same individual x.
 Crucially, the adjectival meaning is conjoined to the nominal meaning
Attributives
 Now, if attributives take a noun predicate and return a new
complex predicate, we know that this can be done
recursively:
dark  NxN ( x)  dark p ( x)
 If we apply the above function to man we get dark man (which
is itself of type <e,t>)

dark man  x man( x)  dark p ( x)

 To get fat dark man, we combine the above with the entry for
fat:

fat  x N ( x)  fat p ( x)


fat dark man  x man( x)  dark p ( x)  fat p ( x)

Permute and drop with adjectives

fat dark man  x man( x)  dark p ( x)  fat p ( x)

 The crucial observation is that the semantic analysis of complex
NPs with multiple attributives views these as conjunctions.
 The predicates are applied to the same argument (the same
individual is a man, dark and fat)
 Logically, if we have P & Q, then:
 This implies P (i.e. From P & Q we can drop Q to get P)
 This is logically equivalent to Q & P (i.e. We can permute P&Q)
 This seems to be exactly what we want...
Adverbs
Brutus ate quickly with a fork
 We want to capture the same permute/drop phenomena
with verb modifiers.
 So, we might say that the above involves conjunction.
 Just like dark fat N is something like [dark(x) & fat(x) and N(x)]
 Let’s think of this example as something like:
 [quick(x) & with-a-fork(x)]
 Problem:
 With John is a dark fat man, we know what x is (namely, John)
 But what is the x with quick and with a fork?
Take 1: subject modification
Brutus ate quickly with a fork
 Here’s a suggestion:
 Maybe quickly modifies the subject (Brutus), which is also the argument of
the verb.
 eat(b) & quick(b) & with-a-fork(b)
 (Simplifying the analysis of with a fork for the moment)
 Problem:
 It doesn’t seem natural to say that quickly is a property of Brutus himself (it’s
a property of what he does)
 Consider:
 Kim tapped Susumo lightly (after Landman, 2000)
 Kim is a Sumo wrestler. He’s anything but light. We don’t want to say that
lightly(x) involves predicating light of Kim.
Take 2: the event argument
Brutus ate the toast quickly with a fork
 We can resolve this problem if we take up our earlier
suggestion:
 Events involve an implicit event argument
 The modifiers modify this argument directly.
Compositional interpretation
eat  yxeeat (e)  AG(e)  x  PAT (e)  y
Takes two arguments, but also introduces an implicit event e
quickly  VyxeV (e)  AG(e)  x  PAT (e)  y  quick (e)
Just like an attributive adjective, requires a verb predicate V to return a complex verb.
eat quickly  yxeeat (e)  AG(e)  x  PAT (e)  y  quick (e)
Compositional interpretation
Brutus ate the toast quickly
eat quickly  yxeeat (e)  AG(e)  x  PAT (e)  y  quick (e)
 To get the full interpretation, we first apply this to the toast:
yxeeat (e)  AG(e)  x  PAT (e)  y  quick (e)(t )
xeeat (e)  AG(e)  x  PAT (e)  t  quick (e)
 ...and then to Brutus:
xeeat (e)  AG(e)  x  PAT (e)  t  quick (e)(b)
eeat (e)  AG(e)  b  PAT (e)  t  quick (e)
Adding modifiers
Brutus ate the toast quickly with a knife
 To get the permute and drop phenomena, we want to have simple
logical conjunction of quickly and with a knife:
eeat (e)  AG(e)  b  PAT (e)  t  quick (e)  INSTR(e)  k 
 Observe that our logical form says:
 Quickly is a property of the event (as are the thematic roles)
 The thematic roles are relations between the event and individuals.
 We get the right entailments:
 Brutus ate the toast quickly (drop with a knife)
 Brutus ate the toast with a knife (drop quickly)
 Brutus ate the toast (drop both)
 Brutus ate the toast with a knife quickly (permute)
Part 3
Event reference
Event reference
 We are often able to refer back to an event, as though it were a
thing. Compare:
 John met Sally. She was very pretty.
 She clearly refers back to the individual Sally.
 This suggests that we have some “mental representation” of the individual to
refer back to.
 Usually, we think of these NPs as introducing a variable. So she can then hook
on to the variable introduced by Sally.
 John met Sally. It was really traumatic.
 It refers back to the event.
 This suggests that we have some “mental representation” of the event to refer
back to. Where does it come from?
 Just as Sally introduces an individual variable in the discourse, perhaps meet
introduces an event variable.
Event reference
 Language allows us to “nominalise” events, i.e. To take verbs and turn them into
nouns (or nouny things, like gerunds).
 In fact, we are able to quantify over events the way we quantify over individuals:
 Every farmer eats meat.
 Every burning consumes oxygen.
 We could think of these events as predicates.
 The noun farmer is semantically a predicate of individuals (the property of things
which are farmers).
 The noun theft is semantically a predicate of ... What?
 The things which are thefts are events.
 This would also capture the relationship between verbs and their
nominalisations quite straightforwardly.
Summary
 We’ve proposed (following Davidson) that events can be
viewed as introducing an implicit event argument.
 Under this theory, we are able to:
 Deal quite flexibly with events having different numbers of
participants in different contexts (eat, eat with a knife etc)
 Deal with event modification in much the same way that we
deal with adjectival modification
 Account semantically for permute and drop phenomena