ter Meulen, Alice G.B., (1997) Representing Time in Natural Language, The
Dynamic Interpretation of Tense and Aspect, MIT Press, Cambridge
Mass., 160 pages, ISBN 0-262-70066-2, $12.50 (paperback).
reviewed by Anne Reboul, LORIA-CNRS, France. Anne.Reboul@loria.fr
Ter Meulen's book is a good introduction to the work which she has been doing
during the past few years on both the representation of eventualities and
temporal reasoning. The book combines her approach's desire not to lose
information previously given in discourse with semantic principles which
are squarely based on Situation Semantics. This allows her to introduce
Dynamic Aspect Trees (DATs), which can loosely be described as temporal
inference pumps. Ter Meulen points out that, though common sense reasoning
plays an important role in temporal reasoning, it is not her aim to deal
with it. Finally the present paperback edition of the
book incorporates an appendix giving some exercises on DAT construction and a
short account of the resemblance and difference between DRT and DATs.
1.0 BOOK SUMMARY
Ter Meulen aims to account for the linguistic and
informational balance between static and dynamic information, proposing
abstract semantic principles relative to tense and aspect and using tools
derived from Situation Semantics.
Language describes the world as being composed of two main types of temporal
objects: events which introduce changes and states which do not. This
linguistic capacity is accompanied by an inferential ability in interpretation
of what is or is not the case when. Ter Meulen deals only with the simple
past, the perfect, the progressive and some uses of the present, that is with
tenses which indicate in which order events occur, through weak constraints on
the temporal relations between events. The difference between the simple past
and the perfect is aspectual, as is obvious when they are examined in the light
of the difference between events and states: the simple past describes events
in a context-dependent way while the perfect only gives stative information.
Aspect can be seen as imposing control on the flow of information and here ter
Meulen introduces three main universal aspectual classes, holes, filters and
plugs which are closely related, respectively, to Vendler's classical
ontological classes of activities ("John ran for half an hour"),
accomplishments ("John built a house") and achievements ("John won the race").
Ter Meulen introduces the notion of situated reasoning about time: according
to her, there is a temporal vantage point in a text from which reasoning
about the text can proceed. This is the first (descriptive) factor. The second
(aspectual) factor is the source of information, while the third factor is
perspective information. New temporal information is processed against a
context (the given) which incorporates at least the lexicon and syntax of the
language, situated and logical inference rules, other semantic constraints
(presuppositions), the available information and the elements of the situation
2. The aspectual verbs
The aspectual verbs describe the internal structure of events and are, in
English, "start", "begin", "commence", "initiate", "resume" (for event
beginnings), "continue", "keep" (for event middles) and "end", "finish",
"terminate", "halt", "cease", "complete" (for event endings). Aspectual
verbs describing the middle or the end of an event carry anaphoric
presupposition (assumptions about the previous existence not of an event,
but of THIS event). Aspectual verbs describing the onset of an event
semantically have an indefinite and existential character. Verbs
describing the middle of an event are quantificational verbs and holes.
Verbs describing the ending of events may be filters or plugs depending
on whether they can be finished (plugs) or ended (holes). There are
pairs of aspectual verbs linked by a semantic contrary relation, such as
"start/finish", "stop/resume", "resume/finish", "keep/end", "continue/end".
In DAT's, events are complex objects, constituted of individuals, their
relations and a positive or negative polarity. Events can be classified in
similarity classes or types based on their parts. A type consists in a
relation, objects and a positive or negative polarity, indicating whether
the relation holds or does not hold between the objects. Types can also
include parameters or indeterminate objects and can be used for any
constituent apart from polarities. A given situation supports the type
if it matches with its constituents. For types including parameters
(parametric types), support depends on the existence in the situation
of some objects corresponding to the parameter and satisfying both
the relation and the polarity. Aspectual verbs can also be described
in terms of their monotonicity: if inferences, for a given aspectual
verbs, are valid from smaller parts to larger
parts, the verb is monotone increasing, while if inferences are valid from
larger parts to smaller parts, the verbs are monotone decreasing.
Ter Meulen then proposes the aspectual cube, that is a cube allowing the
representation of the relations between aspectual verbs as well as of their
monotonicity properties. Each vortex in the cube has two vertical arrows of
which the first represents the dynamic plugs (upward arrow)/static holes
(downward arrow) while the second represents the increasing (upward arrow) or
decreasing (downward arrow) monotonicity. The cube visualizes the dynamic
transitions beginning with + start ; and stopping with "end" or "finish". Its
bottom corresponds to stative holes, while its top corresponds to dynamic
plugs, allowing four combinations of arrows in English : dynamic
plugs+increasing monotonicity, dynamic plugs+decreasing monotonicity, stative
holes+increasing monotonicity, stative holes+descreasing monotonicity. External
negation flips the left arrow changing the dynamicity, while internal negation
flips the right arrow. Finally, there is an arrow kinematics, whereby the
combinations of arrow given above gives a dynamic or stative aspect, this
being always determined by the right arrow (i.e. the value of the combination
is always the value of the right arrow (upward = plug; downward = hole).
3. Dynamic Aspect Trees (DATs)
DATs are a way of graphically representing information about events in directed
graphs where nodes, labeled with descriptive information, can be either holes
or plugs. Each DAT has a unique source node (a plug), which is the rightmost
terminal node, corresponding to the utterance event and determining the
perspective. Any information in the past tense has preceded the source and is
represented by nodes on a branch to the left of it. Present information include
the source. Thus left-to-right order represents the flow of time while downward
arrows represent temporal inclusion. DATs are used for temporal reasoning about
the past (though they do not allow modalities and counterfactuals). There is
also in a DAT a single root node, dominating all the other nodes, including the
source, and representing the entire episode described. There is also a current
node, which is the last constructed one and which can opened (hole) or closed
(plug), this property determining how the next information will be processed.
If the current node is a plug, the new information will be represented as a new
node descending from one of the current node's parent nodes, usually the lowest
compatible parent node. Simple past tense clauses introduce new nodes, while
states, quantificational relations, perfect clauses with "have" and
progressive clauses are introduced as stickers on the current node, if it is a
plug, or on the next node, if it is a hole. Stickers are portable and can be
transmitted to new nodes when the DAT grows.
Introducing a new node independent of the current one indicates a change in
perspective. A path connecting a set of labeled nodes to the root is a
chronoscope and the current chronoscope is the one containing the current node.
Compatibility relations (which are symmetric, reflexive and transitive) are
determined both lexically and through the entailment relations between types on
the same chronoscope.
The notion of chronoscope is one of the main device of temporal reasoning in
DATs, allowing for persistence, i.e. the preservation of information. However,
temporal reasoning depends on the current node, the search for a conclusion
starting at the current node.
DATs are structured semantic objects rather than linguistic expressions and, as
such they follow semantic rules. They are interpreted in event structures into
which they are embedded and which allow temporal reasoning if they obey
semantic constraints. Interpretability for a text depends on the possibility of
constructing a DAT which can be embedded in an event structure obeying these
constraints. An embedded DAT describes an episode if the event structure in
which it is embedded provides values for it. A text is true if the episode
described in the DAT is part of the world.
4. States, generic information and constraints
Stative information are always represented as stickers on nodes in DATs, but
some states are more permanent than others, this difference being represented
as portability conditions in perspective shifts. There are thus rules for
downward portability conditions of stickers and for portability conditions for
accommodation. Portability conditions differ for perfect states and for
progressive states. Generic stative information is also represented by stickers
and can be imported upward depending on whether the property is predicated
under default to the members of a given category or on whether it is predicated
of the category itself rather than of its members and on whether it appears in
conditionals or under temporal quantification.
A DAT gives a certain perspective on the episode it describes and the same
episode may be described in different DATs, each giving a different perspective
on it. This means that DATs on the same episode can lead to different
conclusions. Changes of perspective depend on the rule used for the updating:
hole or sticker rules do not change the perspective, plug or filler rules do.
In other words, changes of chronoscope correspond to changes of perspective
and depend on the status of the current node. There is, however, more to
perspective than just creating a new chronoscope when updating the DAT:
the other possibility goes through perspective refinement. Perspective
refinement corresponds to the unplugging of a plug which turns it into
a hole, allowing the addition of new information under it.
The notion of perspective is strongly dependent on the compatibility of
information or, more generally, on coherence and it is one of the merits of DAT
to allow for a more precise content to the rather slippery notion of coherence,
relying on the rules for DAT construction and the notion of chronoscope.
Perspective refinements usually correspond to flashbacks and consist of three
operations: unplugging nodes in old chronoscopes, resetting the current
node to an old one and making the existing nodes and the new nodes just added
independent though they are dominated by the same ancestor. Ter Meulen proposes
a tentative definition of such a flashback process. She notes that perspective
allows perspective binding of NP anaphora and proposes a perspective
constraint to deal with it. Finally ter Meulen outlines the possibility of
three-dimensional DATs, the third dimension allowing for the representation
of independent structures of nodes, though communality of the root is
preserved. Such three dimensional DATs would be called scenarios though
each part of such DATs with a unique source would
still describe an episode. Parts of scenarios which incorporates different
simultaneous episodes are scenes.
6. A fragment of English
In this chapter, Ter Meulen gives a syntactic and semantic account of a
fragment of English, as "a laboratory environment for the theory of
interpretation" (93), as well as a recapitulation of the rules for DAT
construction and updating.
DATs allow for the simultaneous encoding of three types of information:
aspectual control of information through the open/closed property of nodes,
temporal part/whole relations between events, and descriptive information
through the types labeling the nodes. DATs are compatible with Aristotelian
Realism as defined by Barwise and Perry, that is, with the view that the
structure of information depends on the natural world. A distinction
should be made, according to Ter Meulen between interpreting an utterance
to extract the information it contains and evaluating the
truth-value of that information.
In this appendix which has been added in the paperback edition, Ter Meulen has
provided exercises on DAT construction as well as a comparison between DRT and
DATs. I will only comment on that second part of the appendix. As Ter Meulen
notes, DATs stems partly from DRT. Nevertheless and though DAT logic is not
developed enough to allow for a complete comparison, there are quite a few
differences between them: DATs account for inference from simple past to past
perfect or progressive, for portability conditions of stative information in
chronoscopes and for aspectual verbs, which DRT does not; DRT accounts for
temporal adverbials and for temporal interpretation of subordinate clauses,
which DATs do not; DATs substitute the notion of current chronoscope to the
notion of temporal reference point used in DRT. Finally DATs characterize valid
logic inference as depending crucially from the current node, while DRT
characterizes valid logic inference classically, as logical consequence.
1.1. CRITICAL EVALUATION
Ter Meulen's book is highly stimulating, original and interesting. It must be
read by anyone interested in time, tense and aspect. All the criticisms which
can be leveled against it come from one reason: the
fact that the book is much too short. This means that though it is clear
enough, it is rather difficult to read. It is also quite frustrating in that,
for instance, though Ter Meulen does explain how to construct DATs (in chapter
3) and gives detailed rules, there is not enough step by step examples of DATs
in the book (as for example, there was in Kamp and Reyle 1993). In much the
same way, one would have liked to have a much more detailed comparison with
alternative accounts, such as DRT, the Parsonian account of events or, more
generally, Aktionsart. The appendix, which has been added in the paperback
edition (1997) and which did not exist in the first hardback edition (1995), is
not really enough to satisfy both of these complaints.
Ter Meulen's book hints about a relation between reference to objects and
reference to time (the indefiniteness of the argument NPs of a given verb
may influence the interpretation of the clause, and, notably, aspect). All
of these hints are very interesting and, again, a longer book might have
given the author more scope to develop those intuitions. Thus one can
only hope that this interesting book will be followed by a bigger one,
in which DAT logic could be developed more fully, contrasted with
alternative approaches and where ter Meulen could describe more precisely
her suggestions regarding perspective and the interaction between the
semantic aspect in DATs and the commonsense reasoning whose
necessity she acknowledges but which she does not deal with in her book.
Finally, ter Meulen's book was published at the same time as Pustejovsky's
work on the lexicon, thus she doesn't make use of Pustejovsky's system. It
would be nice to know if an intersection between these approaches can be made.
Kamp, H. & Reyle, U. (1993): From Discourse to logic, Dordrecht, Kluwer.
Parsons, T. (1990): Events in the Semantics of English, A study in subatomic
semantics, Cambridge Mass., MIT Press.
Pustejovsky, J. (1995): The Generative Lexicon, Cambridge Mass., MIT Press.
Reviewer: Anne Reboul, Research Fellow at the CNRS (National Center for
Scientific Research) France. PhD. in Linguistics, PhD in Philosophy, currently
working in The Center for Computer Research in Nancy, in the team dedicated to
man-machine Dialogue. She has written quite a few papers both in French and in
English, is the co-author of the Dictionnaire Encyclopedique de Pragmatique
(Paris, Le Seuil. English translation in preparation for Basil Blackwell,
Oxford) and is the 1997 bronze medalist for Linguistics at CNRS.