Semantics of motion verbs

Sumit Sen, Krzysztof Janowicz

Institute for GeoInformatics (IFGI), University of Münster,

Robert Koch Str 26, 48149 Münster, Germany

{sumitsen, janowicz}@uni-muenster.de

Abstract

Motion verbs in languages convey the meanings of spatiotemporal change. Ontologies that deal with

spatial actions can benefit from the study of such verbs and their usage. This paper explores semantic

components of motion related verbs and verb complexes. Such semantics components assume certain

spatial primitives and intuitions which need to be considered in the context of ontologies derived from

textual descriptions. We formalise such constraints for concepts of motion derived from motion verb

complexes and discuss their use in ontology specification.

1. Spatial changes and motion verbs

Motion verbs have been widely studied, as compared to other classes of verbs and are of significant in the

context of Ontologies in the geospatial domain (Sen, 2004). Verbs of spatial configuration, lodge verbs,

avoidance verbs and occurrence verbs have strong relevance in motion representation as well.

Asher and Sablayrolles (1994) discuss a complex formal calculus for motion complexes (verbs with

prepositions) to obtain their semantics. The Levin’s classification (Levin, 1991) is relatively simple because it

deals with motion verbs in nine categories. The spatial alternations of verbs are dealt separately. These

classifications are not formalized as compared to the work of Sablayrolles (1995), Landau and Jackendoff (1993)

both of which have a similar notion of classification of motion based on the motion starting, path and final

positions. This is done primarily on the basis of partition of space into inside, external zone of contact and

1
outside. Landau and Jackendoff (1993) use such structuring for static spatial prepositions. Conceptual structures

consider semantics of motion with both direction and manner.

Zlatev (1997) presents seven spatial semantic categories which include Trajector, Landmark, Motion,

Frame of Reference, Region, Path, and Direction. These categories help to specify motion concepts specified by

the verb complex. For example we can choose to define our object and person trajectors bound to Ground and

Relative Landmarks. Further we could also choose the cases of perceived motion in alleocentric, and deictic

reference frame. It is important to note that English mostly uses the deictic verbs. The first step towards

extraction of motion concepts from verbs is to classify verbs based on these semantic components. A cross

tabulation of verbs based on such properties can be seen in table 1 below. This table is only an attempt to show

how these categories can be used to distinguish between motion concepts and cannot claim to be a complete

tabulation of all possible categories.

Table 1 Categorization of spatial verb complexes to distinguish between motion concepts

Categorization tool Categories

Trajector Object (come near Person ( come to me ) Event ( come
the earth) closer to a fight)
Landmark Ground(Run away from Rome) Relative (Ran away from here)
Motion Perceived motion (walk, run) No Perceived motion (move up in rank)
Frame of Reference Alleocentric (stand Geocentric ( fly away Deictic (stand
beside me) or take-off from earth) behind the tree)
Region Interior (enclosed) Exterior (touch) Lateral(lie next to)
Path Beginning (start from) Middle (go through) End (to end at)
Direction Distal (going away) Geocentric Upward (going up)

2
2. Cross linguistic classes of motion verbs

Since our attempt is to study motion verbs at the semantic level it is useful to discuss the relevance of spatial

verb classes and lexical conceptual structures in the context of cross-linguistic studies. The first distinction about

motion verbs is the relative significance given to the manner and the path of motion. As discussed by Fellbaum

(2002), some types of languages (like Germanic, Chinese) conflate the fact of motion and manner while some

others (Romance, Greek, Semitic) conflate the fact of motion and path. It is also discussed that there is greater

salience for the path element in the domain of motion verbs and that path could be omitted only in cases where

the activity had no explicit end state. Japanese is reported to have mostly path encoded simplex motion verbs

besides compound verbs which encode both manner and path. This is similar to English Path is therefore the

necessary component of motion verb complexes since the two languages are genetically unrelated.

Acland et al (2004) have discussed that directionality needs to be a part of either the verb or its

prepositional part in a motion complex. This suggests that directionality (as a part of the path component) is one

of the primitives of motion representation and is consistent with cognitive observation discussed above.

2.1 Troponymy, Events and States.

Fellbaum (1999) describes entailment relations in English verbs using the troponymy, causation, inclusion and

pre-supposition relation between two verbs. Among these troponymy which represents hierarchical relations in

verbs is discussed to be analogous to the taxonomical relations prevalent in noun classes and can be argued as an

instrument for deriving root verbs or verb classes. The main aspects about troponymy relations are:

− hierarchies do not exceed 3 or 4 levels

− polysemous in nature

− involves “way-of” or “manner-of” relations (e.g., to walk is to move)

− different from “function-of” relations (e.g., to walk is NOT to exercise)

3
 The concept lattice formed (figure 1) from most frequently occurring verbs of motion in a traffic text based on

Levin’s verb classes. The top 20 verbs such verbs result in 16 concepts based on troponymy relations using

WordNet lexicology. These could form the set of actions discussed . The concepts themselves are not verbs but

are actions that convey the sense of the verbs that occur at the corresponding node.

Figure 1 A Concept lattice of English motion verbs (occurring in traffic code text) based on entailment relations

A similar analysis of entailments in German traffic code by Kuhn (2001) shows entailment relations although the

elements of this analysis is activity based rather than verb based.(Activities could be seen as sequence of actions;

actions in turn are represented by individual verb/verb-complex.) What is interesting is that the event

granularity1 of these is not similar in these cases. This is because verbs like swerve pass and merge has been

1
This term refers to the granularity of the time scale with reference to which the activities denoted by the verbs correspond
to. Thus walk can have events startwalking and stopwalking where as step has events start_to_move and finish; clearly
these two verbs have associated events that are not defined for same time scales. We deal with this aspect in more detail in
section 3.2

4
merged into one activity by Kuhn (2001). It suggests that multiple hierarchies can exist at different levels and are

collapsible, expandable and mappable as required. This mechanism of switching between levels of granularities

is critical to the use of motion verbs in ontologies. The processes of matching and merging of ontologies based

on text analysis needs to examine this aspect. We discuss granularity transformations in section 3.2 but it is

important to mention the notion of inclusion (part-of relations for verbs) as a basis of such granularity

transforms.

Slobin (2005) discusses the linguistic representation of motion events which shows that languages differ

in their preferred means of encoding events in general and motion events in specific for example, English has a

stronger encoding of events relating to manner of motion. We therefore refer conceptual structures proposed by

Jackendoff (1990) which assumes events as conceptual primitives (also called conceptual constituents) along

with states, objects, path, place and property. We can thus represent our spatial action concepts using an event

definition represented as.

FROM ([AIRPORT])
EVENT [
GO( ThingCAR , ] PathTO ([CITYCENTER ])
Car goes from Airport to City Centre.

Since events relate to states we inspect the relation expressed as

[EVENT ] → [ Event INCH ([StateBE Indent ([Thing CAR ], [Place AT ([Pr opertyCITYCENTER ]) ]) ]) ] Inchoative Event changes the state of Car

of being at city centre

One can see that this event can be true if and only if the State of the Car being at the CityCenter exists. Thus if

the Car is doing a shuttle service from the airport to the city centre, one could argue that “Car goes from Airport

to City Centre” is an expression of its spatial action. The problem of such statements however is that, in English

one can choose to declare events only without declaring the state it because there is not compulsion to do so.

Pustejovsky (1995) discusses three main event types on the basis of stative verbs (stand, be), process verbs

(walk, run) and transition verbs (start, stop). It is evident in stative verbs where the occurrence of the event

5
results in only one state. (To Stop results in a state of being stopped). In the case of process verbs, it can result in

alternative states (In the example the Car was at the city centre but could go elsewhere). Enforcing the

requirement of state change would be redundant for the semantics of some verbs but is a necessary condition for

the generic case. Hence we define a set of actions A (such that At1 ∈ A at time t1) & states S (such that At1 ∈ S

at time t1) in our model as below to ensure that there are corresponding stable states t1 , t2 at the start and end.

(At1∈A)G (At1,σt1 ) → ∃σ t2 such that ({σt1, σ t2}⊆ S)G (1)

Here G represents the level of granularity at which A and S are defined. We use σ1, σ2 in place of t1 and t2.

Hereafter we will use A(σ1, σ2) to represent the left hand side of equation (2) for sake of convenience. (“To be

stationed at a location” can be seen as an example of a state where as “to slide from one point to another” is an

example of an action) We can see that states and actions can be concurrent at the same granularity level and

therefore we need to derive constraints for such actions.

3 Motion primitives

MOVE is the only spatial verb that occurs in the current list of semantic primes (Goddard, 2002). The other two

action verbs that occur are HAPPEN and DO which can express semantics of motion, if combined with MOVE.

There are primes related to space WHERE, HERE, ABOVE, BELOW, FAR, NEAR, SIDE and

INSIDE. If we combine these and the three activity verbs we achieve a first list of primary spatial actions that

can be described as behaviours of motion. A cross tabulation is attempted to classify English verbs2 as shown in

table 2 below. It is strikingly close to the static spatial prepositions (Landau & Jackendoff, 1993) based on

motion verb-preposition complexes. Also these categories (and not the verbs themselves) mainly convey

directionality besides manner of motion. The reference location of the verb (Talmy, 1983) is utilized to classify

the verbs, assuming the deictic centre and could be further extended to existence primes like THERE IS.

2 It is important that the English representations of semantic prime is understood in the sense that it is originally intended. For
example the sense of MOVE that is considered as a primitive is contained in the sentence “I can’t move” (Goddard, 2002).

6
Table 2 Spatial actions from Semantic Primes

Above /
Primes Here/Where Below Far/Near Side Inside
Move visit, go, run jump, fall Migrate Skirt Enter
Do Move Fly
(Move) (something) (something) throw, push push aside Pull
Happen (something) (something) (something) (something) (something)
(move) moves flies shifts gets moved enters
Covered, Borders,
There is Be, is, exists floored Removed encircles Contained

3.1 Intuitions of motion

We can also inspect some intuitions about motion which are part of Jackendoff’s semantic structures

(Jackendoff, 1990). We restate them as rules as constraints of the definition (1)

− If someone stays at someplace for a period of time, he is in that place at any instant during that time.

A1 (σ1, σ2) (σ1=σ2 ) → ∃ (¬(A2 (σ1, σ3) (σ3≠σ1) ) ) where t1≤ t3≤ t2 (2)

− If someone doesn’t stay at someplace during an interval, there is a time during the interval he isn’t

there.

A1 (σ1, σ2)∧(σ1≠σ2 ) → ∃ (¬ (A2 (σ1, σ3) (σ3=σ1) ) ) where t1≤ t3≤ t2 (3)

− If and only if someone is not at someplace he is somewhere else.

− If and only if someone is at someplace he is not elsewhere.

A1 (σ1, σ2)∧(σ1=σ2 ) ↔ (¬A2 (σ1, σ2) )∧ (σ1≠ σ2 ) (4)

7
 − If something goes from one place to another it was at the first place first and the second place second.

( (A1 (σ1, σ2)∧ A2(σ2,σ3) ) → t1≤ t2≤ t3 (5)

− If you are going someplace you are not staying anywhere and if you staying someplace, you aren’t

going anywhere.

We do not state a constraint based on the last intuition since it can be shown same as (4) and (5). As a summary

we add that these rules on motion activities ensure some basic rules about spatio-temporal change axioms.

3.2 Granularity of actions and states

According to Reitsma and Bittner (2003) spatial and temporal grain and extent define the granularity structure in

processes. Granularity is also governed by the grain of observation and the type of view. Van Standen and

Narsimhan (2004) classify the notions of cross-linguistic notions of granularity into:

− Placing of event boundaries depending on the language and cultural setting in which the event is

encoded.

− Expressing elements within the set boundaries of a ‘single event’ in time

− The specificity with which these elements are characterized

These are consistent with observations of Hobbs (1985) which explains human intelligence in scaling between

levels of granularity using indistinguishablity relation (~) formalized as a second order axiom, applied on the

domain of interpretation of a first order global logic theory to obtain smaller local theories.

(∀x,y)x ~ y ≡ (∀p∈ R) (p(x) ≡ p(y) from Hobbs (1985)

8
In this axiom R is the set of predicates of the subset of the global theory and it means that x and y are

indistinguishable if there is no relevant predicate distinguishing them. We state that the combination of actions

A1 and A2 of granularity level G can be equivalent to A3 of granularity F if the definition of indistinguishablity

(~) is applied as follows.

(A1G (σo, σm )) ∪ ( A2G (σm,σn )) ≡ A3F (σo, σn ) where (σo ~ σm ) F (6)

4 Extracting actions from texts

The text analysis of the German traffic code (Kuhn, 2001) provides a methodology to derive action components

in task ontologies. The main result of this paper is a lattice structure of actions based on text analysis and

concepts of Entailment, Causation, Inclusion and Troponymy as discussed by Fellbaum (1999). The analysis of

these activities using verb classification (Levin, 1991) into the main categories and subcategories is shown in the

table 3 below3.

Table 3 Actions from traffic code texts

German traffic Main Verb Spatial Action (At) & Current spatial
Code activity class/Sub class target object type(α) state(σt)
Stand Lodge/Verb of Stand at a place At some place
spatial configuration
Enter Motion/Directed Enter a Road Not on the road
Motion
Drive Motion/Not vehicle Drive on Road On some place
name-vehicle motion on the road
Keep Distance Hold-Keep/Keep Drive on Road at a dist. from
car
Yield Possession Yield to car behind Ahead of a car
change/Future
having
View Perception/Sight (no action of motion) -
Change Lane Possession Change to a lane In a lane
change/Exchange

3 WordNet lexicology was used to interpret the sense of the verbs with synonyms because in many cases the exact verbs did
not appear in the Levin’s list of verbs.

9
 Control Speed Motion/Not vehicle Control speed on At a dist. from
name-vehicle motion road car
Turn Motion/Manner-Roll Turn on a road Towards one
direction
Turn around Motion/Manner-Roll Turn around a Towards one
corner direction
Stop Aspectual/Begin Stop on the road Moving at some
speed
Swerve Motion/Manner-Roll Turn sharply on a Towards one
road direction
(go) Pass Motion/Manner-Run Overtake a car Behind a car
Merge Combining- Change to a lane In a lane
Attaching/Mix
Wait Existence/Exist Stand at a place On some place
on the road
Continue Aspectual/Begin Drive on Road On some place
on the road
Be passed Appear/Occurrence Be passed by car Car behind

The issue is whether these activities can constitute the spatial action elements A. View and be passed

clearly do not qualify as spatial (constraints 3-6). Yield and change lane seem to suggest spatial change but they

themselves do not have the motion components and hence the entailment of motion is based on inclusion

principle. Thus yield entails move right and change lane entails both move right or move left.

While merge and keep distance can be also argued to have a similar entailment with drive, aspectual

verbs begin and stop mainly indicate the general task component. Existence and motion verbs which form the

main bulk of these verbs thus constitute the spatial action set A along with the entailed verbs. One could also

argue that these actions could be further broken down on the lines of primitives discussed in section 3.1 and this

is true. This also important in the context of the granularity discussion in section 3.2 but our current aim is to

show how our notions restrict the choice of spatial actions from a description of tasks. This example also

presents us with the spatial states –action combinations like waiting-continue-waiting, driving in a lane-change

lane- driving in a lane.

10
 The resolution of granularity of the states and actions is done on the basis of the indistinguishablity

relation (Equation 6)4. Equations 2-6 help to analyze whether the actions and states are consistent. The motion

concepts from table 1 from the traffic code text are constrained using these and a table 3 represents such

information. Thus the semantics of motion verbs based on the actions and states that they represent are captured

in this representation.

5 Conclusion and Future work

We have explored the semantics of motion verb and verb complexes based on existing work in linguistics. The

categorization of such verbs based on their semantics and the capture of such semantics from formal texts using

concept analysis, entailment relationships and the notion of granularity has been discussed. We have also

presented a formalization of the semantics using the action and state components.

The formalization of the semantics is presently simplistic and future work is required to develop axioms

for representation of such concepts in geospatial ontologies. The use of such concepts of spatial function and

behaviour components in ontologies and ontological reasoning would thus be a natural further extension of this

work.

Acknowledgements

The ideas presented have been shaped by discussions with members of the semantic interoperability research

group at the Institute for Geoinformatics, University of Münster (http://musil.uni-muenster.de).

We wish to express our gratitude to them, especially Prof Werner Kuhn to instigate many of the ideas

expressed in this paper.

4 The current scale can be argued to be of driving direction level [20]

11
References

Aceland, B., Baggetter, N., Bley-Vroman, H., Klein, N., Llewellyn, E., Osborne, G., Stiglitz, J., Waser,

A., Thornton, R., Hackkl, M.(2004) Integrating the Spatial Semantics of Verbs and Prepositions. In Proc of

Seventeenth Annual CUNY Conference on Human Sentence Processing, College Park, MD.

Asher, N., Sablayrolles, P.(1994) A Compositional Spatio-temporal Semantics for French Motion verbs

and Spatial PPs. In Proc of SALT4, Semantics and Linguistic Theory, NY (1994)

Fellbaum, C. (1999) A Semantic Network of English Verbs. In: C. Fellbaum (Editor), WordNet - An

Electronic Lexical Database. The MIT Press, pp. 69-104.

Fellbaum, C.(2002) On the Semantics of Troponymy. In Green, R., Bean, C., Myaeng, S. (eds.): The

Semantics of Relationships: An Interdisciplinary Perspective. Kluwer, Dordrecht, Holland 23-34

Goddard, C. (2002) The search for the shared semantic core of all languages. In Cliff Goddard and

Anna Wierzbicka (eds). Meaning and Universal Grammar - Theory and Empirical Findings. Volume I. John

Benjamins, Amsterdam 5-40.

Hobbs, J. (1985) Granularity. In Proc of Ninth International Joint Conference on Artificial Intelligence.

Los Angeles, California. (1985) 432-435

Landau, B., Jackendoff, R.(1993) "What" and "Where" in spatial language and spatial cognition.

Behavioral and Brain Sciences 16, pp 217-238

Levin, B.(1993) English Verb Classes and Alternations: A Preliminary Investigation. University of

Chicago Press, Chicago

Jackendoff, R.(1990) Semantic Structures. MIT Press, Cambridge, MA

Kuhn, W.(2001) Ontologies in support of activities in geographical space. International Journal of

Geographical Information Science, 15 613-631.

Pustejovsky, J.(1985) The Generative Lexicon. MIT Press, Cambridge, MA

12
 Reitsma, F., Bittner, T.(2003) Scale in object and process ontologies. In Kuhn, W. and Worboys, M.

and Timpf, S. (eds.) Spatial Information Theory. Cognitive and Computational Foundations of Geographic

Information Science. International Conference COSIT'03 Springer, Heidelberg

Sablayrolles, P. (1995) The Semantics of Motion. In Proc 7th Conference of the European Chapter of

the Association for Computational Linguistics, pp 281-283

Sen, S. (2004) Semantics of Spatial verbs and their role in interoperability. Position paper In Report of

the NCGIA Specialist Meeting on Spatial webs Santa Barbara, California.

Slobin, D. (2005) Linguistic representations of motion events : What is signifier and what is signified?

In C. Maeder, O. Fischer, & W. Herlofsky (eds.) Iconicity Inside Out: Iconicity. In Language and Literature

4.John Benjamins Amsterdam/Philadelphia

Talmy, L. (1983) How Language structures space. In Pick, H and L. Acredolo (eds),Spatial

Orientation: Theroy, Research, and Application. Plenum, NY

Timpf, S., Kuhn, W. (2003) Granularity Transformations in Wayfinding, In Spatial Cognition III,

Freksa, C., Brauer, W., Habel, C. and K. Wender (eds) Springer, Heidelberg

van Staden, M., Narasimhan, B.(2004) Granularity in the Crosslinguistic Encoding of Motion and

Location. Edited volume on Granularity and Resolution in Language and Space, Oxford University Press

Zlatev, J.(1997) Situated Embodiment. Studies in the emergence of spatial meaning. Stockholm: Gotab

13