File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/79/j79-1071_metho.xml
Size: 107,684 bytes
Last Modified: 2025-10-06 14:11:15
<?xml version="1.0" standalone="yes"?> <Paper uid="J79-1071"> <Title>American Journal of Computational Linguistics Microfiche 71 SPATIAL REFERENCE AND SEMANTIC NETS</Title> <Section position="1" start_page="0" end_page="0" type="metho"> <SectionTitle> SPATIAL REFERENCE AND SEMANTIC NETS </SectionTitle> <Paragraph position="0"/> </Section> <Section position="2" start_page="0" end_page="0" type="metho"> <SectionTitle> SUMMARY </SectionTitle> <Paragraph position="0"> This paper presents an analysis in a semantic net formalism of the semantic structure of English sentences containing references to spatial- location.</Paragraph> <Paragraph position="1"> Spatial reference, hereafter - SR, provides either static location or motional information John is at home, Fred ran across the street to the store.</Paragraph> <Paragraph position="2"> .The task for the semantic analysis of sentences with SR's is to,make clear what is being positioned. THis has been difficult to do. Previous proposals have left unanalyzed many phenomena including important motional references. This paperv* main conclusion is that a much improved analysis can be obtained by representing the SR's as positioning abstract events and states of affairs.</Paragraph> <Paragraph position="3"> The analysis in semantic nets has the location of an event or state of affairs represented as a node which is linked to the node showing the event or state by arcs: indicating its staus as the spatial attribute. A few SR's are shown as naming these locational entities, which we call place ,object. These SR' s involve examples with &quot;where&quot;, &quot;here&quot;, and &quot;there&quot; However, most SRts are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy ok prepositions is argued for in the paper. Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional fcms. The necessary ordering that'comes with motional references is allowed for by associating temporal elements with the functions.</Paragraph> <Paragraph position="4"> While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential semantic structures necessary to deal with SR's. A similar complexity is necessary to deal with the informakion on the location of objects which is gained from sentences with SR's. The paper suggests-the use of inference rules to allow for this.</Paragraph> <Paragraph position="5"> The most surprising of the paper's oonclusions is that a strong tie exists between referehces to space and temporal information. In fact, the locations of all events and states of affairs placed by SR's are argued to be locations in both space and time. The effect of this conclusion is most clearly seen in a formalized definition of the primitives of the semantic seructures , which is also presented in semantic nets. There, as ane possible interpretation of the place object, it is shown as a set of pairs of volumes in space and points in time.</Paragraph> </Section> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> TABLE OF CONTENTS </SectionTitle> <Paragraph position="0"/> </Section> <Section position="4" start_page="0" end_page="0" type="metho"> <SectionTitle> Acknowledgements ........................... Bibliography ............................. I. Introduction </SectionTitle> <Paragraph position="0"> This paper presents an analpis in a semantic net formalism of the semantic structure of English sentence8 containing references to spatial location.</Paragraph> <Paragraph position="1"> Spatial referefnce, hereafter - SR, provides either'static location or motional information :</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 1.1 John is at home 1.2 Fred ran across the street to the store. </SectionTitle> <Paragraph position="0"> The task for the semantic analysis of sentences with SR's is to make clear what is being positioned. This has been difficult to do. Previous proposals have left unanalyzed many phenomena including important motional references. This paper's main conclusion is that a much.improved analysis can be obtained by representing the SR's as positioning ab~tract events and states of affairs.</Paragraph> <Paragraph position="1"> The analysis in semantic nets has the location of an event or state ~f affairs represented as a node which is linked to the node showing the event or state by arcs indicating its status as the spatial attribute. A few SR's are shown as naming these locational entities, which we call pLaee object. These SR's involve examples with &quot;where&quot;, &quot;here&quot;, and &quot;therei'. However, most SR' s are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy of prepositions is argued for in the paper.</Paragraph> <Paragraph position="2"> Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional forms.</Paragraph> <Paragraph position="3"> The necessary ordering that comes with motional references is allowed for by associating temporal elements with the functions.</Paragraph> <Paragraph position="4"> While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential semantic structures necesaaBy to deal ~ith SR's. A similar complexity is necessary to deal with the information on the location of objects which is gained from sentences with SR's. The paper suggests the use of inference rules to allow for this.</Paragraph> <Paragraph position="5"> The most surprising of the paper's conclusions is that a strong tie exists between references to space and temporal information.</Paragraph> <Paragraph position="6"> In fact, the locations of all events and s tetes of af fairs placed by SR' s are argued to be locations in both space and time. The effect of this conclusion is most clearly seen in a formaIized definition of the primitives of the semantic structures, which is also presented in semantic nets. There, as one possible interpretation of the place object, it is shown as a set of pairs of volumes in space and pointe in time.</Paragraph> <Paragraph position="7"> The paper has nine sections following this one. In the first, the limitations of previous analyses of the semantic function of SR's is considered. Thea in one section, the semantic net formalism and, in the next, the syntactic distinctions used in the study are introduced. The next four sections present epcr more complex situations. The first section shows simple direct analyses involving one event or state. The next section presents complex sentehtial structures with non-movement SR's. Motional references are analyzed in the next. The connection between time and SR's is diswed in the fourth section.</Paragraph> <Paragraph position="8"> Following these analyses of sentential semantic structures, a section hs given over to the formalization of the definition of the st tuctures used. The paper ends with a discussion of the limitations of the proposal and possible extensions to it.</Paragraph> <Paragraph position="9"> There is available a discussion in greater detail of a preliminary analysis to the one given here (~ondheimer, 1975).</Paragraph> <Paragraph position="10"> There is also available for comparison an analy is by this authbr of the same meaning phenomena, in the competing paradigm of model-theoretic semantics (Sondheimer , 1978) The current.</Paragraph> <Paragraph position="11"> paper is distinguishable by its better developed semantic net formalism nnd itn emphasis 3n producing computationally justified structures.</Paragraph> <Paragraph position="12"> XI. Prev ous Efforts Ihe pazt has seen many studies of SR phenomena.</Paragraph> <Paragraph position="13"> There ha8 been interest in connecting. language and scenes, e. g. , Coles ( 19681, Kochen ( 19691, Winograd (1972), Badlcr (1975), and Tsotsos (1976). The use of langdage to capture the bpatial 8;ru:ture of the physical world has been studied, e.g., Hobbs (1975) and Kuiper 197 The conceptual structure of the terms used in SR and the pragmatics of, evaluating them has been studied, e. g., Cooper (1968). Bennett (19751, and Denofsky (1976). Finally a number of studies have considered our topic: the position of a SR within the semantic structure of a sentence.</Paragraph> <Paragraph position="14"> Studies of our soet tend to be distinguishable by the type of entities SR'S are claimed to locate. In some cases, the SR'B apply to only physical objects.. In athers, they apply to only abstract forme identifying events and states o affairs. A broad third type of analysis shows different sorts of entities being modified. Each has.its limitations.</Paragraph> <Paragraph position="15"> 11.1 Analyses using Physical.Objects The paradigmatic phenomenon for tte analyses that claim physical objects as the referents of SR1s is the noun phrase modifier: 2.1 The man in the car left.</Paragraph> <Paragraph position="16"> The SR in the above is the phrase &quot;in the car&quot;. The proposals of Norman and Rumelhart (1975), Abrahamson (19751, Gei~ (1975a, b, anc c) and Schubert (19761, among others, would try to show the relation of the SR to &quot;the iuan-&quot; directly. Figure 2.1 illustrates the typical structure in the style of Schubert (1976). This figure shows &quot;the man&quot; being located (LOCI at a time, indicated by the T - link, and at a location which was in &quot;the car&quot;.</Paragraph> <Paragraph position="17"> This style of analysis seems simple and direct.</Paragraph> <Paragraph position="18"> It appeals to the intuition that only physical objects take up space.</Paragraph> <Paragraph position="19"> It promises t~ be easy to apply, FIGURE 2 .I &quot;The man in the car left&quot; in the st-yle of Schubert (1936).</Paragraph> <Paragraph position="20"> since all that is required is to associate SR's with the sentential elements which are modified which reference physical objects. Unfortunately, there are problems.</Paragraph> <Paragraph position="21"> It can be difficult to find all or any objects with which to associate an SR. Often there is more to an event than. just its partiripahts' locations:</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.2 John is playing solitaire in the basement. </SectionTitle> <Paragraph position="0"> If we hear example 2.2 then more than John is known to be in the basement. His cards are, for. example. Further, the location of the action is more than the instantaneous position of John and his cards. For example, space where the cards may, potentially be plaoed must be included. Similarly, the following does not indicate that John is next to the school:</Paragraph> </Section> <Section position="3" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.3 John is playing baseball next to the school. </SectionTitle> <Paragraph position="0"> Hemi.ght be playing outfield 300 feet from it. It can be difficult to find any objects to associate with an SR: 2,'4 * In France, literary criticism is a high art form.</Paragraph> <Paragraph position="1"> 2*.5 The explosion was in the garage.</Paragraph> <Paragraph position="2"> tn both the above examples, only complex analyses showing many understood ad potential participants can allow PS01 object-reference.</Paragraph> <Paragraph position="3"> Allowing fox motional sentences is a very serious problem for objectreferenee analyses. The typical proposal is to show motion as change from one static location to another: Examples like the above involve duration in a key way and can not be shown with reference to one position. For example, at no time was the man &quot;across&quot; the puddle like ~aleigh's cloak was across it. Similarly, two points showing the man's change of position are inadequate since the same initial and final positions are a'cceptable in all three cases. Finally, adding an intermediate point will not be adequate, since the man might reach that point while on a path that otherwise holds a different relation to the puddle. As shall be seen, the lesson to be learned from these examples is that in allowing for motion, it is the entire path that must be considered and not selected positions of objects. A second uniform type of analysis postulates events and states of aftairs as the subject of SR's (see for example, Davidson, 1967, Lakoff, 1970, and Harman, 1972). Events and states of affairs are said to be the two types of situations that utterances describe. Taking them 88 the subjects of SR' e claims that it is not the participants but the overall situation that is being referenced. This can be seen in Figure 2.2, which shows one of Davidson's analyses in a eemantfc net notation. The diagram shows that there is a strolling by John which has a particular time and space coordinate. The benefits of this analysis include the independence of event and state existence from discussion of spatial location, the ability to handle location of vaguely bounded events and states, and the simplicim of application. However, again the simple direct methods that have been proposed are unsatisfactory.</Paragraph> <Paragraph position="4"> It is often difficult to simply associate SR'S with a central event or state since SR's in some utterances must modify different entities: 2.11 John held the ice bag to his head in the car.</Paragraph> <Paragraph position="5"> In 2.11, only the ice bag is to John's head but John and the ice bag are in the car. Motion is still a proslerb: 2.12 John walked from his car across the yard to the ilouse.</Paragraph> <Paragraph position="6"> How the event of 2.12 can be &quot;from&quot;, &quot;across&quot;, and &quot;to&quot; simultaneously and also have these aspects temporally ordered is nowhere explained in these analyses. Finally, even if SR's are associated with events and states of affairs, the fact that something is often learned about participants' location must be explained. For example, from the sentence of Figure 2.2, the fact that John was in the streets of Bologna is clear, but from the semantic structure only the location of the strolling is clear. No coherent way has been presented to allow for this kind of relationship.</Paragraph> <Paragraph position="7"> Strol ling, FIGURE 2.2 &quot;John strolled through the streets of Bologna at 2 a.m.&quot; in the style of Davidson (1967).</Paragraph> <Paragraph position="8"> W.3 Nonuniform Analyses The third style of SR analysis is nonuniform in nature. These either mix the two uniform analyses or elaborate on the simple event or state analysis. Mixed analyses claim that some SR's locate concrete objects while some locate events or states of affairs (see for example, Winograd, 1972, and Schank, 1973) By sacrificing the simplicity that comes from uniformity , these analyses avoid the uniform analyses' complementary problems. However, the mutual problems, especially motion, are left unsolved.</Paragraph> <Paragraph position="9"> The nonuniform analyses that elaborate on the nature of events and states of affairs are best represented by Case analyses, see Bruce (1975). They claim that either the overall location or specific aspects of events and states are located. Takin~ Fillmore (19711, as opposed to the betterknown but earlier Fillmore (1968), as the model, four spatial cases can be seen. An SR can either reference a static locatioh (the Location case), place of origin (the Source case), place of termination (the Goal case), or location of intermediate motion (the Path case). In terms of events and states of affairs, the first case can either be used for overall event or state location or it may be used to locate an aspect of the event. The final three cases all relate to different aspects of a motional event. This allows for examples like 2.12, with inherent temporal ordering among the cases allowing for the ordering of the SR's.</Paragraph> <Paragraph position="10"> The Case analyses still has problems. The two uses of the static case conflict in sentences with two static locations such as the one where the ice bag is held to the man's head while the man and ice bag are said to be in the car. Two instances of the Location case seem to be required, but if both appear, there is no way to identify their differing function. Also, motion is still troublesome. As Fillmore (1971) point^ out, instances such as the underlined phrases in the following seem tc indicate a need for an unbounded number of instances of the Path case.: 2.13 He walked down the hill across thebridge through the pasture to the chapel.</Paragraph> <Paragraph position="11"> The underlined phrases refer to motion ordered in time, e.g., he walked the hill before the bridge.</Paragraph> <Paragraph position="12"> However, Case analysis gives no way to order instances of the same case. Gruber (1965) points out the same problem with the Goal case: 2.14 I walked to New York to my mother's.</Paragraph> <Paragraph position="13"> Finally, the Case proposal rnuat be given some physical interpretation. Any representation of meaning must at some point be related to a m~del of the world. In this instance the idea of a source, goal, and path must be somehow related to models of motion.</Paragraph> <Paragraph position="14"> This paper presents a proposal for an analysis that is nonuniform in the same way the Case analysis is. A uniform source for locations modified by SR's is given, but the predication of these spaces by SR's is shown to be much more complex than previously thought. Futther, sentences are not seen as being as simple with respect to SR's as previously supposed. Before presenting the analysis, two sections will be devoted to preliminary topics: our semantic net formalism and the eyntactic status of the phenomena considered.</Paragraph> <Paragraph position="15"> 111. Semantic Nets The resblts of this analysis of SR phenamenaare formalized in semantic nets or networks. This %ormalism is currently a popular choice for semantic analyses. It allows clear, expressive graphic presentations and possesses many positive computational properties. Because of its popularity, it also allows wide dissemination of ideas. Working against this last claim is the $roliferation of versiods of the representation, for example, Hendrix (1976), Norman and Rumelhart (19751, Shapiro (,1971), Simmons (19731, and Woods (1975). This section clarifies what is meant here by the formalism, which can be seen to most closely resemble that af Brachman (19773.</Paragraph> <Paragraph position="16"> Semantic nets have been used for representing many aspects of htelligence. Of ten they are used to represent factual information concerning ob jerts, actions, and states. They have separately been used to show the semantic structure of utterances It is this use that mainly interests us. However, there is a connection between the two uses. All semantic structures must he related to structures that represent facutal inbrmation and each use of a type of object, event, or state of affair must be related to a concept that explains it. This can be thought of as paralleling the relationship between a semantic structure shown iq the predrcate calculus and a model in which that structure has a truth value. In a complete net, the above translates into the necessity of nodes for concepts representing types of events, states of affairs, and objects and nodes for instances of tokens of these concepts. The &quot;token&quot; nodes must link to &quot;type&quot; nodes that define them. These definitions must include specification of abtributes of an instance in terms of restrictions on values, fuwtional role of thd attribute, and other things. The instance nodes must be connected to instantiations of the attributes. Concept nodes must also be related to dther concepts., have overall structural conditions, locate inference rules that map apply, etc. All this information is essential to any artificially intelligent entity, just as the model is essential'to any analysis in the predicate calculur.</Paragraph> <Paragraph position="17"> However, for showing the semntic relations in which we are mainly interested, an abbreviation is sufficient just as only the formulas are sufficient in most studies using symbolic logic.</Paragraph> <Paragraph position="18"> Hence a special abbreviation will be used in all sections except fX where the definitional level wili be discwsed.</Paragraph> <Paragraph position="19"> Central to our abbreviation will be nodes that collapse types and tokens.</Paragraph> <Paragraph position="20"> These will identigy the verbal concepts that characterize the events and states of affairs. We will call them &quot;event/state&quot; nodes. They will be circled and capital letters will be used for abstract types, such as CAUSING. Nonabstract forms will be shown with names that suggest the interpretation, e.g., Sleeping will suggest the sleeping state. When a node represent8 a physical object, identifying information will be included in quotes, e.g., &quot;the bus&quot;. Names placed on ascs will abbreviate and suggest the functional roles of attributes. For example, - ANTE for antecedent and - CONS for consequence will be used with CAUSING. Case names will be used with many event and state of affairs types. These will include: T for &quot;Time&quot; showing the time an event occurred or state held.</Paragraph> <Paragraph position="21"> A for &quot;Agent&quot; showing the instigator of an event or state.</Paragraph> <Paragraph position="22"> 0 for &quot;Object&quot;, the neutral case (as Fillmore (1971) explains it &quot;the wastebasket&quot;) .</Paragraph> <Paragraph position="23"> Restrictions on types of entities which will be necessary will be shown by nonoval shapes for nodes. For example, time instances will be shown in parentheses and time intervals in square brackets.</Paragraph> <Paragraph position="24"> Finally, because it is not essential for our purposes, specification of time will often be left out of most semantic structures. Similarly, ire will consider only declarative statements. Figure 3.1 shows a typical struoture.</Paragraph> <Paragraph position="25"> Some concepts that act as functions will also be used.</Paragraph> <Paragraph position="26"> Each of these will look like a relation associating parameters with a value. The value will be identifiied by a - VALUE arc.</Paragraph> <Paragraph position="27"> Inference rules will be presented in the form of FIGURE 3.1 &quot;John loved Mary all last year.&quot; &quot;subnetll' 2 &quot;subnet2&quot;, where on seeing slibnet 1' subnet* is to 'be added to the semantic net. These rules will include variables within nodes, where the variables are to be bound on matching and referenced on inferencing. These variables will be in the form of capital letters, e.g., X.</Paragraph> <Paragraph position="28"> To summarize, our semantic net formalism uses concept names, descriptions of objects, mnemonic arc names, and mnemonic shapes for nodes to abbreviate the two levels in a semantic net. Also used are functions and inference ruLes. This will be enough to represent the semantic relations involving reference to space that are being considered. Unfortunately, it is one more unique formalism.</Paragraph> <Paragraph position="29"> However, it sdds no new structures, only abbreviating others. We leave as an unproven claim that it will fit in with any formalism which shows identifiable event and state of affairs nodes such as Norman add Rumelhart (1975) and Schank (1973). IV.</Paragraph> <Paragraph position="30"> The Syntactic Structure of Spathl References and the Primacy of Prepositions Semant'ic structure ia the topic of our paper, but the syntactic structure of sentences with SR s is also important.</Paragraph> <Paragraph position="31"> Us consideration clarifies the range of phenomena king etudied. With SR, the basic syntactic structures involve prepositional phrases.</Paragraph> <Paragraph position="32"> All other SR are analyzable in terms of these structures.</Paragraph> <Paragraph position="33"> In this sectian, the syntax function of prepositional phraees will be considered and arguments for their primacy will be presented.</Paragraph> <Paragraph position="34"> Our main interest in syntax is in structuring our didcussion of semantics. However, the problems of parsing and generation make the syntax of' SR's independently imprtant. These are not our topics here. However, in an earlier issue of this journal we presented a parsing scheme that produces semantic from syntactic structure and applied the scheme to current clags of phenomena (Sondheimer and Perry, 1975).</Paragraph> </Section> </Section> <Section position="5" start_page="0" end_page="0" type="metho"> <SectionTitle> IV.l Locative Prepositions </SectionTitle> <Paragraph position="0"> Prepositional phrases that express SR's can be called locatiye. They appear conti~~ously, as in example 4.1, or diacontiguously, as in example 4.2:</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 4.1 I put it on table. 4.2 The table I put it on is broken. </SectionTitle> <Paragraph position="0"> The discontiguous example can be taken as derivable from (reducible to) the cont iguous f oms in generation ( interpretation).</Paragraph> <Paragraph position="1"> Hence only contiguous examples will be considered. These are primarily employed in four syntactic roles: complement, quhlifier, adjunct, and locative object.</Paragraph> <Paragraph position="2"> There is also one special dependent usage that will be described at the end of this section. The complement usage of locative prepositions arises only when they are the &quot;complement&quot; of the verb &quot;be&quot;:</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 4.3 He is in the kitchen. </SectionTitle> <Paragraph position="0"> Quirk et al., (1972)'distinguishes them from predicate adjective and nominal usages. Locative objects and adjuncts with copulative eentences can be distinguished f rw compl ements by the presence of these ad jec tivee and noun phrases :</Paragraph> </Section> <Section position="3" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 4.4 There are lions in Africa. 4.5 He wes important in Chicago. </SectionTitle> <Paragraph position="0"> The qualifier usage of locative prepositions is part of noun phrases and shows the location of the reference of the noun phrase: 4.6 The man in ths car left.</Paragraph> <Paragraph position="1"> The strings in some sentences may make it appear that locative prepositions are part of noun phrases when they are not: 4.12 *What she took care of was John in Chicago.</Paragraph> <Paragraph position="2"> The asterisk &quot;*&quot; here and throughout marks ungrammatical sentences. The ungrammaticality of the above examples indicate that the strings in question are not noun phrases. Hence the prepositional phrases cannot be qualifiers. Adjunct usages are prepositional phrases that are external to the clause of a scntence: 4.13 I met John on the train.</Paragraph> <Paragraph position="3"> Locative object usages are objects of verbs and internal to clauses: 4.14 I put the lamp in the corner.</Paragraph> <Paragraph position="4"> 4.15 He yelled at John.</Paragraph> <Paragraph position="5"> 4.16 He saw her in the park.</Paragraph> <Paragraph position="6"> There is some controversy on the distinction between these two types. We can present two syntactic and one semantic classification procedures. First, adjuncts are never required for granrmaticality, while locative objects can be: 4.17 I met John, 4.18 *I put the lamp.</Paragraph> <Paragraph position="7"> Second, adjuncts always allow shifting to presubject position without loor of gramatical&ty or shift in meaning: 4.19 On the train I met John.</Paragraph> </Section> </Section> <Section position="6" start_page="0" end_page="0" type="metho"> <SectionTitle> 4.20 *At John, he yelled. </SectionTitle> <Paragraph position="0"> 4.21 In the park, he saw her.</Paragraph> <Paragraph position="1"> Note that in 4.21, the man is definitely placed while in one interpretation of 4.16, the locative object one, he is not.</Paragraph> <Paragraph position="2"> Semantically, we claim that adjuncts locate the entirety of events and states discussed, while locative objects can locate only part of what ir described. For example, in 4.16, the locative object reading shows only the woman's position in the park, not the location of the &quot;seeing&quot; as a whole. The following is also informative: 4.22 He dropped it behind the door.</Paragraph> <Paragraph position="3"> 4.23 Behind the door, he dropped it.</Paragraph> <Paragraph position="4"> Both examples are similarly ambiguous with respect to the SR. One sense, the most likely to be identified in 4.22, is that the end result of the dropping wan that the object came to be behind the door. The second sense, the most likely for 4.23, is that the dropping took place behind the door. The firet sense ehows partial predication and a locative object usage. The second ehows overall predication and an adjunct usage.</Paragraph> <Paragraph position="5"> Some forme that seem to be adjuncts do not at first glance appear to make overall predication: 4.24 On the train, he commented on the Empire State Building.</Paragraph> <Paragraph position="6"> 4.25 In Chicago, John wrote to his mother.</Paragraph> <Paragraph position="7"> The Empire State Building's and John's mother's position are independent of the train and Chicago. However, we can claim there is still overall predication since the commenting and the writing were done on the train and in Chicago, respectively.</Paragraph> <Paragraph position="8"> Durational adjuncts also complicate the semantic test: 4.26 He cried through the tunnel.</Paragraph> <Paragraph position="9"> 4.27 He sat still from New York to Chicago.</Paragraph> <Paragraph position="10"> l'hese prepositional forms show duration of the crying and sitting and should be taken as adjuncts. The first gives overall predication. The second example shows two phrases that individually give ~artial predication. However, together they give overall predication. Further, they cannot be used individually: 4.28 *He sat still from New Y~rk.</Paragraph> <Paragraph position="11"> 4.29 *He sat still to Chicago.</Paragraph> <Paragraph position="12"> These are the four primary uses of locative prepositions. We claim that the semantic structure of other SK'B can be represented through these forms. We will now show this. In general, this will be done by observing the SR's structure or by paraphrase arguments.</Paragraph> <Paragraph position="13"> IV.2 Other Spatial References as Locative Prepositions Some spatial terms can have syntactic and semantic functions similar to prepositions in that they directly serve to relate two forms: 4.30 San Francisco is north of Los Angeles.</Paragraph> <Paragraph position="14"> 4.31 The car is to the left of the building.</Paragraph> <Paragraph position="15"> These examples can immediately be given prepositional-like semantic structures. In other sentences, these terms appear as nouns and adjectives: 4.32 The North is desolate.</Paragraph> <Paragraph position="16"> 4.33 He hit my left leg.</Paragraph> <Paragraph position="17"> Here, th~ forms can be paraphrased in the prepositional-like form which can be taken as their underlying semantic form: 4.34 The part of the country to the north of the rest is desolate. 4.35 He hit one of my legs that is to the left of the other.</Paragraph> <Paragraph position="18"> Another category, the locative prepositional adverbs, although lacking syntactic objects have assumed semantic objects. This is shown by our ability to question the missing object, which is a means for distinguishing this category from verb particles (~uirk et al., 1972, p. 103) : 4.36 He went up.</Paragraph> <Paragraph position="19"> 4.37 Up what did he go? 4538 He picked it up.</Paragraph> <Paragraph position="20"> 4.39 Up what did he pick it? A diverse variety of non-prepositional locative adverbs can be handled with prepositional forms. Assumed objects can also be seen in cases of paired prep-ositianal-adverbs and prepositions. These are suggested in parentheses below: 4.40 He walked across (a walkable space) to the blackboard. 4.41 He jumped from (a jumpable place which was on) the table. Some adverbs can be straightforwardly treated as the equivalent of prepoaitiod phrases. These appear as the concatenation of a preposition and a noun and refer to the spatial relation referenced by the preposition with respect to the type of object referenced by the noun: 4.42 He ran uphill.</Paragraph> <Paragraph position="21"> 4.43 He is overseas.</Paragraph> <Paragraph position="22"> The suffix &quot;-ward1' following a preposition or preposition-like term produces an adverb that can be treated as having a destinationel- or orientationel-like meaning as shown by the following,paraphrases: 4.44 He moved leftward.</Paragraph> <Paragraph position="23"> 4.45 He moved to the left.</Paragraph> <Paragraph position="24"> 4.46 It pointed upuard.</Paragraph> <Paragraph position="25"> 4.47 It pointed up the space.</Paragraph> <Paragraph position="26"> Other adverbs can be treated as having a neutral prepositional sense like &quot;at&quot; or 11 to&quot; in their semantic representation: 4.48 He is home.</Paragraph> <Paragraph position="27"> 4.49 He is at home.</Paragraph> <Paragraph position="28"> Finally, many noun phrases that indicate position can be seen as havipg prepoeitione subsumed by the verbs they appear with and hence can be represented as containing prepositional phrases, see Gruber (1965) for elaboration: 4.50 He gave Susan the ball.</Paragraph> <Paragraph position="29"> 4.51 He gave the ball to Susan.</Paragraph> <Paragraph position="30"> 4.52 He jumped the fence.</Paragraph> <Paragraph position="31"> 4.53 He jumped over the fence.</Paragraph> <Paragraph position="32"> There are a few forms in SR's that I can not always claim to be represented by preposi tidnal f om& Theq e are where&quot;, &quot;here&quot;, &quot;there&quot;, and measures of distance. These will be dealt with separately. In general, we will deal wit% prepositio~lal phrases with the aseumption that all SR phenomena are covered. Beyond the examples already g-n, it is hard to say what should be considered an SR. Adjectivw such as &quot;10ng&quot;~ and &quot;fat&quot; inwolve the abstract properties of ~bjects more than their properties as objects momentarily situated at a point in space. Many examples apsear to be metaphors of SR or make oblique reference to space: 4.54 I stood trial.</Paragraph> <Paragraph position="33"> 4.55 I go to Ohio State.</Paragraph> <Paragraph position="34"> All of these will be ignored. Doubtlessly, there are unarguable cases of SR that are being overlooked. For this, I can only apologize.</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> IV.3 Semantic Structure of Pre~ositional Phrases </SectionTitle> <Paragraph position="0"> Since prepositional forms are the basic method of making SR, their representation is central to this analysis. They will be given a semantic representation as concepts relating what is referenced by the SR to their own complements, see Figure 4.1. The referenced entity will be identified by the - F link for &quot;figure&quot; and the complement by the - G link for &quot;ground&quot; (Talmy , 1975)*. Each prepositional concept will be defined as comparing the figure's space to the location of the ground's object at the time associated with the figure (Section *There would have to be a second ground link for &quot;between&quot;: I left it between the window and the dbor.</Paragraph> <Paragraph position="1"> FIGURE 4.1 A prepositional semantic structure.</Paragraph> <Paragraph position="2"> VIfI contains more discussion on this point).</Paragraph> <Paragraph position="3"> Prepositional concepts will all be considered abstract and written in capital letters. The reason abstract forms are used will become clear in the following sections.</Paragraph> <Paragraph position="4"> One particular dependent use of the preposition &quot;from&quot; f all8 outside the simple pattern shown in Figure 4.1 as well as outside of the four claeeerr of prepositional usages: 4.56 John .is far from home.</Paragraph> <Paragraph position="5"> 4.57 John is acr-oss the street from home.</Paragraph> <Paragraph position="6"> In both of the above, John ip distant from home. But in 4.57, John is not across the street in the usual aenee of &quot;across&quot; stretching the width of the street. As Bennett (1975) points out, the &quot;across1' and &quot;from&quot; phrases combine in euch a way that we understand that it is the way that must be travelled in starting from home and going to John that is &quot;across the streetB'.* This can be allowed for in semantic nets with a function, - WAY, producing a path through space joining two points identified by INIT for Initial and FIN for Final links, see Figure 4.2. - For example 4.57, G would identify the cltreet, INIT the home, and FIN where John -e rame meaning also ariaer in sentences such as the following where there is an understood &quot;from point&quot; that muat be represented: He died acroae the river.</Paragraph> <Paragraph position="7"> is. How John's location is to be shown is explained in the next section, where the baaic and simpler SR's are analyzed.</Paragraph> <Paragraph position="8"> FIGURE 4.2 A prepositional semantic structure for the special &quot;from&quot; usage.</Paragraph> <Paragraph position="9"> V. Static Adjunct, Complement and Qualifier usages Section I describes our basic claim: the source of the locations being referenced by SB's can be represented as being the locations of events and states of affairs. In this section, this claim is associated with the amantic net model and applied to those types of SR's for which it works immediately. These are the static adjunct, complement and qualifier usages.</Paragraph> <Paragraph position="10"> V. 1 Basic Structures In our semantic net model, the locations of events and states of affairs will be shown as attributes of eventhate nodes through arcs leading from the nodes to locational entities. For each event/state node involved with an SR there will be only one such arc and locational entity. Theee atcs will be labelled - P to suggest a spatial attribute or &quot;Place&quot; case. The locational entities will be referred to as place objects. They are the basis of our analysis. These place objects can be taken for the time being as volumes in space. The sort of ~olume they are will be ela6orated upon. Place objects will be identified by boxes. Figure 5.1 gives a typical diagram.</Paragraph> <Paragraph position="11"> It must be askedwhether place objects are required in semantic representations or simply ad hoc creations. The answer is that they are required since speakers treat tb as existing byreferring directly to them with some user of &quot;where&quot; :</Paragraph> </Section> <Section position="2" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 5.1 Where is John living? 5.2 I found it where John was sleeping. </SectionTitle> <Paragraph position="0"> Place objects can not be outlined strictly in space like a solid can.</Paragraph> <Paragraph position="1"> This is not important, because there is no way in language to directly and completely locate any object. In the last section, it was argued that except -for &quot;where&quot;, &quot;here&quot;, and &quot;there&quot;, every SR is like a preposition. Hence they all give relative position. With those that do not, &quot;where&quot; can be show</Paragraph> <Paragraph position="3"> FIGURE 5.1 &quot;John is sleeping here. II as referencing place objects not definite locations. &quot;Here&quot; and &quot;there&quot; both predicate spatial qualities of place objects not specific locations:</Paragraph> </Section> <Section position="3" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 5.3 John was born here. </SectionTitle> <Paragraph position="0"> In 5.3, the location of the doing is simply associated with &quot;here&quot;. Hence a semantic analysis that associates SR with abstract locations can work if the means of predicating these locations and of fitting them into semantic structures can be found.</Paragraph> <Paragraph position="1"> V.2 Applying the Analysis With the place object, there is a Large class of phenomena that can be represented directly. These include static adjuncts (5.4) as opposed to durational ones (5.5): Applying the place object analysis to static adjuncts is easily defendable. One test for adjuncts in the last section was to see if it located the entirety of the event or state discussed. The static adjuncts are identifiable in this way. Since the place object shows the location of that entirety, static adjuncts can therefore be directly applied to them. Figure 5.2 gives a typical analpsir. FIGURE 5.2 &quot;~ohn is sleeping in the kitchen. II This basic treatment ex-tends to static complement usages. These relate an object to some location in space and time. To show this an abstract predicate, BEING-AT, can be postulated whose object case shows an entity whose spstiotemporal location ia specified by Plaee and Time cases, see Figure 4.3. Proposing a state of affairs to show an object's existence in space and time slay at first seem artificial. But in fact, it provides representations ismorphic to the usual &quot;direct&quot; repreeentatioa of object location. For example, Schubert (1976) uses a concept - LOC which by a link .I A identifies an object, a link B the object s location, and a link T its time frame (see Figure 2.1). These - v match our BEING-AT, 0, P, and T cases, respectively. Schubert sometimes abbreviates SR's when the preposition &quot;at&quot; is used. However, this is simply an abbreviation and his underlying form remains equivalent to ours.</Paragraph> </Section> </Section> <Section position="7" start_page="0" end_page="0" type="metho"> <SectionTitle> BEING -AT </SectionTitle> <Paragraph position="0"> FIGURE 5.3 &quot;John is behind the house. II Static qualifiers parallel either the adjunct or complement analysis. As m adjunct to verbal noun, we can claim that an event/state correspondiag to the event or state described by the nouns can be located by the SR in the same way as an actual adjunct. With qualifiers applying to concrete nouns there can be a BEING-AT eventheate showing that the existence in space and time of the object is being discussed. The qudifier can then modify its place object.</Paragraph> <Paragraph position="1"> This would then show thefollowing equivalently: 5.13 The man in the car yesterday left.</Paragraph> <Paragraph position="2"> 5.14 The man do was in the car yesterday left.</Paragraph> <Paragraph position="3"> The possibility of time modification as in 5.13 is good evidence for the treatment of qualifiers as having underlying complement structure (Winograd, 1972).</Paragraph> <Paragraph position="4"> V.3 Allowing for Object position Now that both adjunct and complement usages have been considered, our method of allowing for the positioning of objects while representing SR's as positioning eventlstates can be explained. As was discussed in Section 11, an event/state analysis must explain how ati artificially intelligent entity can discover that John vae somewhere from the representation of an event or state involving John being located there. This can be taken as being something like discovering the appropriateness of the complement form (5.16) from the truth of the ad jmct form (5.15): 5.15 John slept in the kitchen.</Paragraph> <Paragraph position="5"> 5.16 John was in the kitchen.</Paragraph> <Paragraph position="6"> Within the computational paradigm, the discovery of 5.16 from 5.15 is made easy by inference rules. Whenever the semantic analyses of a sentence like 5.15 is presented to a system, rules associated with the type of eventistate node involved can produce inferable information.</Paragraph> <Paragraph position="7"> This process allows for the human process of the deduction of specific information about participants in an event or state of affairs from kn~wledge~of the type of event or state of affairs. This is actually what is happening with SR's. From our knowledge of sleeping, we know that someone is where he is sleeping. From our knowledge of &quot;working for&quot;, we know that Bill but not necessarily John is at the store in the followirg: 5.17 BiU is working for John at the store.</Paragraph> <Paragraph position="8"> From our knowledge of contact cases such as in 5.18, we know that the location of the intersections of the objects is learnable: 5.18 The ball hit Mary on the ear.</Paragraph> <Paragraph position="9"> In semantic nets, these facts can be shown by inference rules associated with the appropriate concepts. In Figure 5.4, the &quot;subnetll' 9 &quot;subnet2&quot; form described in Section 111 is used to allow for the sleeping case. Other rules will, of course, be needed for other concepts.* The predication of pLace objects, hich are the locations of events and states of affairs, therefore stands as the core of our analysis. How it directly applies to represent certain SR's has been shown in this section. In the next, more indirect analyses are conaidered. Sleep1 ng BEING - AT FIGURE 5.4 The encoding of the rule &quot;If you know where something is sleeping, then you know where it is&quot;.</Paragraph> <Paragraph position="10"> *A potential critic may argue that the extra processing involved with inference rules should be avoided if at all possible.</Paragraph> <Paragraph position="11"> However, no other analysis of SR successfully avoids its use (~ondheimer , 1975).</Paragraph> <Paragraph position="12"> VI. Non~ovement Locative Object Usages The analysis of locative object usages is not a8 simple as that of other forms.</Paragraph> <Paragraph position="13"> Looking back to Section T', most of the problems with earlier studies arise from this class.</Paragraph> <Paragraph position="14"> The solution to tthese problems is found in an elaboration on the basic form of our event/state analysis.</Paragraph> <Paragraph position="15"> This elaboration proceeds in two directions. First, the semantic structure of the sentences containing the troublesome SR's is seen to be more complex than otherwise thought. Second, the nature of the SK is seen as more complex. The first case is best seen with non-movement and the second with movement sR's. This section covers the non-movement type of locative object. Section VII covers the movement type.</Paragraph> <Paragraph position="16"> We can review the problems with the use of event and skate location in the non-movement cases, briefly. There is a need to differentiate referents which can be seen in the following:</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 6.1 John held the ice bag to his head in the car. </SectionTitle> <Paragraph position="0"> The ice bag is to John's head, but both the ice bag and John are in the car. The first SR involves a locative object, the second an adjunct. With a simple approach to eventhate location, they would not be differentiated. There is a similar problem in some adjunct references to the location of only part of an event or state of affairs.</Paragraph> <Paragraph position="1"> For instance, in 6.2, only the boy isplaced which the hawk is definitely physically present: In an open field, a boy watched a hawk.</Paragraph> <Paragraph position="2"> In the latter case, although not the former, the use of inference rules might be suggested. However, a better answer can be found.</Paragraph> <Paragraph position="3"> VI.l Continuous Position and Perception Verbs The semantic structure of simple sentences have often been analyzed as involving multiple events and states of affairs, see far example, Schank (1973) and Norman and Rumelhart (1975). If we can see problematic sentences in this light, then perhaps we could assign the various SR1s to different event/states. Indeed, we can do both.</Paragraph> <Paragraph position="4"> Instances of calmative relations betweell events and states of affairs are found in many problem sentences. Change-of-state events applying to separate states of affairs are seen in others. Simple instances of embedded events and states of affairs are seen in yet others. &quot;Hold&quot; belongs to a clam of verbs that involve continuou~ position. Others in the class include &quot;adhere&quot;, &quot;cling&quot;, and &quot;keep&quot;.</Paragraph> <Paragraph position="5"> With Locative objects, these can all be seen ar causations. Each ha8 an action which causes some entity to remain somewhere. In our example 6.1, John's holding-type action causes the ice bag to remain somewhere. ~ealizkg thir allows us to analyze the SR's as locating events and states. The overall SR, &quot;in the car1', can be seen, as adjuncts were explained in the laet section, as locating the highest event/etate within the causation. The &quot;to his head1' can be seen as locating the resuftant state. This is shown in Figure 6.1. The TO in the diagram represents a static eense of &quot;to&quot;.</Paragraph> <Paragraph position="6"> the boy watched a hawk.&quot; &quot;Watch&quot; belongs to a class of verbs that includes &quot;hear&quot;, &quot;see&quot;, and &quot;taste&quot;. These can all be seen as involving the perception of another event or state of affairs.* In our example 6.2, it is the being somewhere, the existing, of a hawk that is watched. How this allows for the SR to be associated with the correct event/state is evident from Figure 6.2. This analysis may seem somewhat forced here, but other exmaples show more overt event or state forms: On these occasiow, an assumed entity can be added to the semantic structure: I heard ( sane thing) through the door.</Paragraph> <Paragraph position="7"> Inference rules play an important part in these analyses.</Paragraph> <Paragraph position="8"> For example, the positioning of John and the ice bag must be derivable from the structure of Figure 6.1. An inference rule must associate the position of the HOLDING-ACTION with their positions. Another rule must relate a place object for the HOLDING-ACTION as inside that of the CAUSING. Conversely, there should be no inference rule applying to the structure of Figure 6.2 to show the place object of the BEING-AT as being contained in that of the Watching.</Paragraph> <Paragraph position="9"> VI.2 Other Verb Classes There are a number of other classes of verbs that take static locative objects, see Table 1. We will survey their analysis in the remainder of the section and close with a comment on several related forms.</Paragraph> </Section> </Section> <Section position="8" start_page="0" end_page="0" type="metho"> <SectionTitle> TABLE 1,- A CATEGORIZATION OF SOME VERBS THAT ACCEPT NON-MOVEMENT LOCATIVE OBJECTS </SectionTitle> <Paragraph position="0"> 1. Continuous Position: adhere, cling, hide, hold, keep.</Paragraph> <Paragraph position="1"> 2. Perception: hear, see, taste.</Paragraph> <Paragraph position="2"> 3. Attachment, Containment, Posture, and Creation: build, close, crouch, draw, erect, glue, hang, lay, lean, lock, nail, paint, sew, shut, sit, stand, write.</Paragraph> <Paragraph position="3"> 4. Contact: grab, hit, kick, kiss, kneel, punch, slap, slug, touch. 5. Change of State: break, chop, cook, cut, fry, shatter, spill, split. 6. Discovery and Thought: dream, find, imagine, lose, recognize, remember, spot, think.</Paragraph> <Paragraph position="4"> 7. Copula-like: happen, gave, occur, remain, stay, take place. 8. Portability: bring, carry, send, take, wear.</Paragraph> <Paragraph position="5"> The next class of verbs adds anokher abstract predicate to the set of forms we have considered: c( him&quot; FIGURE 6.3 &quot;H& nailed it to the wall.&quot; The above sentences involve attachment, containment, posture, and creation. Each has an element of coming-into-being that must be represented. The standard form for these sentences shows the action of an agent causing the bringing about of a state of affairs. The locative object is shown locating this state of affairs, see Figure 6.3. The coming-into-being concept in this structure is labelled COMING-ABOUT. The segment of the structure inside the dotted line is there to show the analysis these verbs take in the second type of usage they allow: These; examples lack agents and any sort of causation. The farma within the dotted line in Figure 6.3 show exactly this structure. The prepositional form, TO, in Figure 6.3, is to be understood in the static sense just as with Figure 6.1. In fact, this is the case with all prepositional forms used here. It is an important advantage of this analysis that it uses only static senses in semantic structures. On the surface, it is often said that the locative objects of the current set of verbs have dynamic senses. However, with a separate inchoative eventletate, this is unnecessary. This allows. tb+ representation of presuppositions like &quot;to&quot; or &quot;on to&quot; either through &quot;at&quot; or &quot;on&quot; as Cruber (1965) does, or through their own static bense as in example 6.10. e his is one in a series of reductions. It was shown in Section IV that some double prepositional phrase structures involved &quot;from&quot; can be reduced to a simpler form. It will be seen elsewhere that other simplications can be made. That underlying senses of the prepositions are being used explains why our prepositional concepts have been capitalized.</Paragraph> <Paragraph position="6"> Another class of verbs-shows contact. They take the two types of analyses just discussed. They also show a coming-into-being sense vhen no agent i~ present but a state is achieved. All three cases are shown in the following: 6.14 I touched her on her face with my hand.</Paragraph> <Paragraph position="7"> 6.15 The tree touches the window near the top.</Paragraph> <Paragraph position="8"> 6.16 The ball touched my leg near the knee.</Paragraph> <Paragraph position="9"> The semantic structures for each of these can contain an event/state showing contact between the two objects to which the SR's can be applied. Another class of verbs which show change of state have all three types of structures with locative object usages: 6.17 I broke it on the rim.</Paragraph> <Paragraph position="10"> 6.18 The cup broke on the rim.</Paragraph> <Paragraph position="11"> 6.19 The cup is broken on the rim.</Paragraph> <Paragraph position="12"> 6.17 is causative/coming-into-being, 6.19 is coming-into-being, and 6.19 is static only. The static form in each can again take the SR. In noncausative examples with thelse change of state verbs, SR's generally appear to act as locative objects, as a test from Section IV shows: 6.20 The cup broke on my knee.</Paragraph> <Paragraph position="13"> 6.21 *On my knee, the cup broke.</Paragraph> <Paragraph position="14"> The noncausative examples includes SR' s which reference ob jeats not inherently possessed by the changed object, such as 6.20, but which place the entire event. In these cases, the SR'r should be treated similarly to adjuncts and &own applying to the COMING-ABOUT event/state.</Paragraph> <Paragraph position="15"> Some verbe which take locative objects are like perception verbs in not requiring causative analyses to explain locative object usages. These lnclude discovery and thought verbs, such asl'spot&quot; and &quot;thought&quot;. They can be ahown with embedded event/states. With locative object readings, 6.22 and 6.23 involve only locating of the direct objects: 6.22 I spotted her behind the dresser.</Paragraph> <Paragraph position="16"> 6.23 I thought of Mary at the seashore.</Paragraph> <Paragraph position="17"> These entities can be shown in an event/state claiming they existed in a certain time and space with the SR predicating that event/state.</Paragraph> <Paragraph position="18"> This treatment would parallel the structure of overt examples of embedded events as in the following: 6.24 I spotted you stealing some bananas.</Paragraph> <Paragraph position="19"> 6.25 I thought of you dancing.</Paragraph> <Paragraph position="20"> Finally, as Lyons (1968) notes, some verbs, which we treat here as having locative objects, seem to relate to SR'S in the same way as the complement usages : 6.26 It occurred in Chicago.</Paragraph> <Paragraph position="21"> 6.27 It remained in New Orleans.</Paragraph> <Paragraph position="22"> These can be analyzed with one eventistate showing both adjunct and locative objects identifying the same entity.</Paragraph> <Paragraph position="23"> So we have seen that the complexity associated with many SR's comes from their semantic environment, not themselves. With the exception of a class of verbs covered in the next section this covers the range of verbs that take non-movement locative objects. Also covered but only indirectly are a few senses left from the last chapter.</Paragraph> <Paragraph position="24"> We can now see how qualifiers of verbal nouns that are acting as non-movement locative objects can be analyzed. We can also see that resultive senses of the copula can be shown with a COMING-ABOUT, see Figure 6.4.</Paragraph> <Paragraph position="25"> COMING-ABOUT FIGGRE 6.4 He is just now in the house.</Paragraph> <Paragraph position="26"> VII. Motion Problems with motion arise in every analysis of SR considered in Section 11. In this section, what is, as far as we know, an entirely unique approach to the semantics of motion is presented. Our analysisa centers on movement locative objects. As has been mentioned, this involves cmplex modification of the location of motional events. The section first presents a brief discussion of the structure of movement sentences, then motivates our view of motion predication, and finally presents the details of the representation. VII .1 The Structure of Movemement Sentences Thc verbs that take movement locative objects are numerous, see Miller</Paragraph> <Paragraph position="28"> puntf', set1', etc. The structure of the sentences with movement locative objects resembles that of sentences with non-movdment locative objects in being complex. Nearly all examples show causative structure with an actioin in one event/state causing motion in another illmo more, 1971). The appropriate analyses approximately pairs the following:</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 7.1 John threw the ball through the door. 7.2 John's throwing caused* the ball to go through the doox. 7.3 Mary walked out of the house. 7.4 Mary's walking caused her to go out of the house. </SectionTitle> <Paragraph position="0"> The only sentences which take simple, causative-like analyses are those with &quot;go&quot; and &quot;come&quot;.</Paragraph> <Paragraph position="1"> An important aspect of the analysis of movement SRfs is the concept to be</Paragraph> </Section> </Section> <Section position="9" start_page="0" end_page="0" type="metho"> <SectionTitle> 1 I </SectionTitle> <Paragraph position="0"> used in the motional event/state. The semantic equivalents of &quot;got1 or comef' will not do. These verbs have special deictic conditions on them (Fillmore 7-.5 Go there.</Paragraph> <Paragraph position="1"> 7.6 *Go here.</Paragraph> <Paragraph position="2"> 7.7 *Come there.</Paragraph> <Paragraph position="3"> 7.8 Come here.</Paragraph> <Paragraph position="4"> FIGURE 7.1 The basic structure for movement locative object sentences. Only &quot;take&quot; and &quot;bring&quot; show the same pattern. For this reason, an abstract concept of pure motion, called - GOING, will be used in our analysis. Figure 7.1 shows the sentential structure into which most movement SR's will t. The structures for &quot;take&quot; and &quot;bring&quot; will have Going and Coming, respectively, in place of the abstract form. For &quot;go&quot; and &quot;come&quot; themselves, the semantic structures will match the motional event/state shown wj th the other verbs with the exception of the type of eventhtate. The place objects of all the motional events can be considered the same, as can the way SR's apply to the different types of motion. We can also think of motional qualifiers as analyzable with the same structure. Because of this, the structure of movement predication will be considered in general and isolated from other forms.</Paragraph> <Paragraph position="5"> VZI .2 Thinking About Motion r As was pointed out in Section 11, one reason that motional SR1s are difficult is the multiple predications of different types which must be orderable in time.</Paragraph> <Paragraph position="6"> These problems can be overcome with appropriate consideration of the motion and the place objects of motional events.</Paragraph> <Paragraph position="7"> The ineight for a better analysis comes from considering answers to questions of where motion occurs. Consider the answer to where the first Marathon was run. It is probably some thing like &quot;in Greece&quot; or &quot;from Marathon to ~thens&quot;. These tend to place the entirety of motion.</Paragraph> <Paragraph position="8"> It is unlikely to be just &quot;from ~arathon&quot; or &quot;to Athens&quot;. These just place, part of the motioa. People tend It to locate motion as if it were a single thing, a motion&quot; so to speak. This is how we propose to think of the place object of motibnal events. Place objects of motional events can be thought of as showing that motion, essentially showing a trace of the path of motion. This trace yould be similar to the trace a piece of chalk leaves as it crosses a blackboard. But it should be the marks that would be made by the entire chalk if space was a three-dimeneional blackboard and the entire chalk could write. This idea is displayed pictorially in Figure 7.2 with another example where something approaching an overexposed photograph of a rolling ball shows a solid cylinder tracing a ball's movement. It is this type of cylinder that motional place objects represent. This trace idea has one great merit. It allows direct analysis of the most troublesome class of movements SR1s: 7.9 He walked through the puddle.</Paragraph> <Paragraph position="9"> 7.10 He walked across the puddle.</Paragraph> <Paragraph position="10"> 7.11 He walked around the puddle.</Paragraph> <Paragraph position="11"> 7.12 He walked over the.puddle.</Paragraph> <Paragraph position="12"> As was pointed out in Section 11, the aboue require a representation that considers every instance of movement. The trace idea does this in such a sidewal k FIGURE 7.2 A ball rolling across a sidewalk to a porch.</Paragraph> <Paragraph position="13"> way that the SR's can be shown applying to the trace directly. Further, it does it in a way that allows the basic static use of the preposition to be used in the representation: 7.13 The bridges across the Mississippi are closed.</Paragraph> <Paragraph position="14"> This was pointed out in Section IV to be the same sense that applied in the across-from&quot; form: 7.14 The man stopped across the street from here.</Paragraph> <Paragraph position="15"> Hence three usages collapse into one with this representation. This concept can be extended to allow for differentiating &quot;up&quot; and &quot;down&quot; by considering the solid traces to have an inherent ordering basedoon the direction of motion: 7.16 He walked up the hill.</Paragraph> <Paragraph position="16"> 7.17 He walked down the hill.</Paragraph> <Paragraph position="17"> Hence, the traces in 7.16 and 7.17 could be exactly the same except for the ordering and the preposition could be sensitive to this. This ordering sensitivity shows up with other uses of the prepositions and other prepositions: 7.18 The carotid arteries extend up the neck to the head.</Paragraph> <Paragraph position="18"> 7.19 A woman stood at the front of the line while a man stood at the rear.</Paragraph> <Paragraph position="19"> bnce its use is not arbitrary.</Paragraph> <Paragraph position="20"> The trace or path idea does not provide an immediate explanation for other movement 5#'s, those that reference instantaneous change: 7.20 He hit the ball into the cotner.</Paragraph> <Paragraph position="21"> 7.21 He walked out of the house.</Paragraph> <Paragraph position="22"> With the above we can not say that the overall path of motion wao either &quot;into&quot; the house or &quot;out of&quot; the house in the static sense of these prepositions Hoerever, there is a way we could use the static sense. If we could refer to podtione achieved by the moving object as it followed the path, we could say that there were positions where the object first got to be &quot;into the corner&quot; and &quot;out of the house&quot;. This would be like allowing reference to the position of the individual balls displayed in Figure 7.2. We can conclude that we ought to be able to reference parts of place objects.</Paragraph> <Paragraph position="23"> Being able to reference parts of motion actually leads to a solution of the problem of temporal ordering inherent in multiple SR, such a8 .. . across the yard up the stairs . . .&quot;. If these durational forms ere also thought of as modifying discrete, bounded parts of the kind of place objects that are being discussed, then they too can be compared. For the phrase just mentioned, a part of the motional object that~wae across the yard could be compared to a part that vas up the stairs as being lees further along it. The same could be done to capate the parte involved with instantaneous reference.</Paragraph> <Paragraph position="24"> To sunuuarize3 the idea is to think of movement as a trace of the event over tb, vbich has an inherent orientation and which cah be predicated in part. We can now almost present our representation. We will first present a slightly incapleke proposal and then rwlse it.</Paragraph> <Paragraph position="25"> VII. 3 Semantic Structures for Motion Tentatively, we propose two different functions to produce parts from complete place objects. These are called SEGMENT and UNIT. They will be used with durational and instantaneous references, respectively. The durational function can be taken as picking off bounded parts of a place object. The instantaneous function can be assumed to pick off part of the trace beginning at the earliest point, and going up to the point of change. Both functions will have the place object they accept identified by an S link and the produced space identified by a VALUE link. To distinguish the two outputs, the SEGMENTized place object will have a colon inserted, and the UNITized one a period. The I I segments will be shown as ordered through numeric&quot; comparisdns. Pigcre 7.3 therefore gives a tentative analysis for the sentence &quot;The cat came across the yard up the stairs into the house&quot;. One SEGMJ3NT function picks out the motion across the yard while another picks out motion up the stairs. A UNIT function picks out motion into the house. The segments are all ordered by less-than-orequal links.</Paragraph> <Paragraph position="26"> The temporal ordering of the partial traces is the one tentative part of the analysis. To have it be sensible, some scale of comparison must exist. The appropriate choice appears to be the temporal scale. - When the locations were achieved is, of course, what is being ordered. There must also be conventions on application of the comparison. This is because there must be a way to force the comparison on only the appropriate end points of segments. We might develop a way of making these conventions inherent, but I propose to make them explicit. yard up the stairs into the house.</Paragraph> <Paragraph position="27"> Our final proposal for the structure of motional SR's is to include time parameters with the functions. In this way, both the end points of the segments and the temporal scale can be identified.</Paragraph> <Paragraph position="28"> For the SEGMENT function, two linkr, T1 and T2, will identify the times that initial and final points were occupied. - For the UNIT function, one link, T, will identify the time the final position war achieved. These structures are shown in Figure 7.4 and 7.5.</Paragraph> <Paragraph position="29"> FIGURE 7.4 The motional elements in &quot;John walked acreas the yard.&quot; FIGURE 7.5 The motional elements in &quot;I hit the ball into the khair.&quot; FIGURE 7.6 The motion component of &quot;John walked from his house to the cat.&quot;. I Now, in order to allow for multiple motional locative objects, two time instances can be related with a temporal relation, - LE, for less than or equal. This is done in Figure 7.6.</Paragraph> <Paragraph position="30"> An interesting aspect of the semantic structure of Figure 7.6 is the static representation of &quot;from&quot;. It is to be understood as showing that up to some point in the journey the moving object was not away from the house, but that it eventually got to be away from it.</Paragraph> <Paragraph position="31"> &quot;Out of&quot; and &quot;off of&quot; are analyzed similarly. VII.4 Static Spatial References Applying to Motional Events Besides the ,durational and instantaneoue predications of motion, there can be overall predications of moving objects.</Paragraph> <Paragraph position="32"> These come in two forms. Adjuncts in movement sentences place the entirety of motion: 7.22 In Chicago, he walked around the downtown.</Paragraph> <Paragraph position="33"> 7.23 John came to Chicago in a plane.</Paragraph> <Paragraph position="34"> One class of verbs, which allows both movement and non-movement locative objects, allows the moving object to be statically placed during movement: 7.24 He carried the dog onto the bus in a box.</Paragraph> <Paragraph position="35"> 7.25 He. brought John to Chicago in a plane.</Paragraph> <Paragraph position="36"> This class is the portability verbs left over from the last section. These verbs take causative analyses with a motional event/state as the caused event. In both of these kind of examples, the motional event must have its motional properties represented at the same time as its static properties. Instantaneous and dutational SR'S must be shown predicating special place objects which are parts of whole place objects. Therefore, we must show the over411 predication applying to different forms. these must be the complete place objects representing the entirety of motion. This is consistant with our other anlayees, as will shortly be seen in more detail. It .will also simplify the inference rules that bring down overall spatial predications from higher levels to the motional place objects. This analysis is seen in Figure 7.7 which essentially summarizes this section. We have introduced twd new functions and types of place objects. These have allowed for movement locative objects.</Paragraph> <Paragraph position="37"> We must, however, realize that there are other uses-for this analysis.</Paragraph> <Paragraph position="38"> We will see why in the next section.</Paragraph> <Paragraph position="39"> The last section may have given the reader the impression that the analyses for motiopal SR's are really different from those given other SR1s. Motional place objects have been set out as a history of movement in space and time.</Paragraph> <Paragraph position="40"> Nomotional place objects are left as &quot;just&quot; the location of certain events and states of affairs.</Paragraph> <Paragraph position="41"> In this section, we argue that this should definitely not be assumed. The place object of nonmotional SR's must be seen to have the same space-time structure as motional place objects.</Paragraph> <Paragraph position="42"> These are several arguments for this point.</Paragraph> <Paragraph position="43"> Relative motion has been considered only for sentences with movement verbs, but relative motion and references to motion are common as adjuncts of &quot;nonmotional&quot; sentences:</Paragraph> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 8.1 John held the ice bag to his head in the moving car. 8.2 Jane sat on her purse from New York to Los Angeles. </SectionTitle> <Paragraph position="0"> In each of tlie above examples two objects are statically related, i.e., John and the ice bag, and Jane and her purse, respectively. However, all are moving. One pair moves but remains static with respect to a car. The other pair is moving and changing with respect to -two cities. Hence, motion must somehow be allowed for in these &quot;nonmotional&quot; analyses. Further , change of relative position must be allowed for in at least one. No hint was given of how this last problem is to be solved in any of our diacussions of nonmotional SR's.</Paragraph> <Paragraph position="1"> Even when motion is not overt, time may have to be considered with SR1s: 8.3 He died in his car.</Paragraph> <Paragraph position="2"> As we have seen in the analyses of sentences like 8.3, the car is to be related to the event of the dying.</Paragraph> <Paragraph position="3"> Consider the fact that the car is moveable.</Paragraph> <Paragraph position="4"> If we were to check to see if this were true, we must have either a history of the car's location or have the ability to find its location at the time of death. In oher words, time must be available for even instantaneous events.</Paragraph> <Paragraph position="5"> Jane</Paragraph> </Section> </Section> <Section position="10" start_page="0" end_page="0" type="metho"> <SectionTitle> FROM TO -A<-&quot;) </SectionTitle> <Paragraph position="0"> fi I FIGURE 8.1 &quot;She sat on her purge from New York to Los Angeles. . The way to extend our analysis to cover these facts is to recognize the connection between moti'onal and nonmotional place objects. Motional events involve the location over time of moving objects. Ndnmotional events and states of affairs do not necessarily involve moving objects, but they can involve location over time. This location is the space of spaces occupied by an event or state of affairs during its holding. To analyze the case8 of relative motion and motion with nonmotional events and states of affairs, these locations-over-time must be taken as the locations bf the events and states of affairs. Note that this does not change of the analyses presented earlier, only the way they are understood. With the same tyw of place object in all SR's the problematic examples that began this section can be elegantiy alloved for with the use of the motional functions. This is shown for example 8.2 in Figure 8.. So in conclusion we propose an analysis that treats all SR's the same.</Paragraph> <Paragraph position="1"> IX. Interpreting the Representation The last several sections presented the &quot;syntax&quot; of our semantic analysis. The term syntax is appropriate since the form of the analysis was presented.</Paragraph> <Paragraph position="2"> The semantics or interpretation to be given the proposed structures was only informally discussed, as when the trace or path analogy for motion was introducld. We noted in Section I11 that semantic nets do not just allow for the syntactic aspect of meaning structures, but also for the representation of the interpretation or definition of the concepts used in these structures. In this section, this property will be used to help formalize an interpretation of our analyses. This is only one of many possible interpretations, but showing it will help clarify the semantic structures. The formaloism for the conceptual definitions is based on Brachman (1977). Again, many abbreviations of a complete formalism are used.</Paragraph> <Paragraph position="3"> The center of our previous discussions was the place objecc. This must also be true in discussing conceptual definitions. The nature of the place objects must first be defined, followed by the definition of everything that relates to place objects. ~ventlatates will be discussed first, then the SEGMENT and UNIT functions, and then the prepositional concepts. The definition of the WAY function wiil not be attempted.</Paragraph> <Paragraph position="4"> IX. 1 The Place Object Our interpretation of the place object wiil be based on a discrete representation of time. Time can be considered as composed of arbitrarily densely packed time instances. A place object can show the location of an event/atate at one time for instantaneous events/states, a set of consecutive instances for durations1 event/states, and any set of instances for intermittent event/states. Structurally, we can take a place object to be a set of what we can call platelets, each of which is an ordered pair whose first element is a volume in space and whose second is an instant in time.</Paragraph> <Paragraph position="5"> This is formalized in Figures 9 .la and b.</Paragraph> <Paragraph position="6"> In Figure 9.la, the node labelled Place Object stands for the concept of a place object. An arc with the special label DATTR points to its one defining attribute. This attribute and a11 others in this section are shown by a special node shaped as a square. The fact that placelets are members of the set that compose place objects is the defining attribute of place objects, This node captures this by using an arc labelled ROLE to point to the special name Membe'r and one labelled V-R for value-restriction to point to the restriction on any member, namely that it must be a Pldcelet. In Figure 9.lb, the concept Placel* is defined. In this case, the concept has two defining attributes since a placelet must have a space and a time. The two attributes are shown accordingly with the one in the role restricted to be a SPACE and one in the role called - t being a - TIME. The concepts of SPACE and TIME will be treated as primitive, here. IX.2 Event/States and Place Objects The structure we defined for place objects will be referenced whenevel place objects are used. One reference will be in eventlstatea where place objects are involved with the case P. Hence, with the conceptual definition of every type of event/state that has a location, there will be a definingattribute with role P and value-restriction Place Object.</Paragraph> <Paragraph position="7"> It is also the case that with each event/state, there will be a way to ehaw how the place object fits in with the .definition of the event/state. Thig will include the way in which the place object will be related to the participants in the eventlstate and structural restrictions on the place object Consider theabstract eventlstate GOING. GOING requires of its place object that the placelets showlwhe.re the moving object was at each instance during the movement., The placelets must refer to the time of the GOING.</Paragraph> <Paragraph position="8"> Since a discrete representation -of time is used, placelets for successive instances of time during the movement must show an overlap in positions occupied. Further, since movement is FIGURE 9.2 Partial Definition of GOING.</Paragraph> <Paragraph position="9"> necessary at least two positions among the placelets must be different. All these facts will have to be shown in the definition of GOING.</Paragraph> <Paragraph position="10"> In order to show the flavor of eyent/state definition, we show in Figure 9.2 the relation between moving object and place object for GOING. The definition shows that the event/state has three attributes corresponding to the cases, P, T, and 0. Names have been added to the attribute nodes to make this easier to see. The everitlstate also has a structure identified by a special 5-C link, tor structural conditions, which is used to identify how the event/stete is structured. The conditions for GOING are a set of conjuncts identified by the label in the diamond shaped node. This shape is an ald to the reader and indicates a logical operator. The structure necessary for the spatial relation is the leftmost of the conjuncta. It essentially takes the form of an implication statement eaying that for every placelet in the place object, the moving object, which is identified by the 0 role, will have a BEING-AT holding for the place and time in the placelet.</Paragraph> <Paragraph position="11"> The statement begins with a logic node, labelled EVERY, identifying a universal quantification.</Paragraph> <Paragraph position="12"> The domain of the quantified variable is shown by the link labelled - x.</Paragraph> <Paragraph position="13"> By po3nting to the appropriate attribute node, the restriction is to the set of placelets in the place object.</Paragraph> <Paragraph position="14"> The link labelled - P identif ies the proposition within the scope of the quantifiet which shows that for Cach placelet BEING-AT is the case for the entity in the 0 role at the place and time of that placelet. The representation of this last depends on the ability to focus in on attributes of entities being quantified over, for which see the FOCUS-SUBFOCUS mechanism in Brachman (1977). This ability is indicated here by the special representation of the P attribute.</Paragraph> <Paragraph position="15"> IX.3 The SEGMENT and UNIT Functions Formalizing the definitions of the SEGMENT and UNIT functions is fairly straightforward. Both can merely identify subsets of sets of placelets. The structural conditions for both can be shown with the same function which can be called GENERATE-RANGE. It can be asstuned to apply to any set and to produce the subset that fits a range defined by two limits and a measure. Since the definitions are similar, only the UNIT function will be shown, see Figure 9.3. The GENERATE-RANGE function for the UNIT definition can take as its input, identified by the SOURCE link, the place object marked by the S link of the UNIT function. The scale for measurement can be established by reference to a special Temporal scale. and apply to the time values in the set of placelets being operated on. The boundaries of the subset to be generated can be shown by FROM and - TO linke.</Paragraph> <Paragraph position="16"> The FROM value would be produced from the set of placelets by a special LOWEST function to produce the lowest time value from among the placelets. The TO value can be a placelet with time specified by the T role in the UNIT function and unspecified space. The RESULT link can show that the generated value should be connected to the VALUE role of the UNIT function.</Paragraph> </Section> <Section position="11" start_page="0" end_page="0" type="metho"> <SectionTitle> SPACE </SectionTitle> <Paragraph position="0"/> </Section> <Section position="12" start_page="0" end_page="0" type="metho"> <SectionTitle> IX,4 Pre~osi tions </SectionTitle> <Paragraph position="0"> Prepositions are the final concepts whose xelation to place objects will be considered. It is always the case that prepositional concepts relate place objects which aFe locations over time to simple objects. The suggestion in the last section was that the locations of the place objects are related to the position of the referenced object at the times of the place object. The nature of this relationship depends on the source of the place object being predicated.</Paragraph> <Paragraph position="1"> Consider first predication of place objects which directly show the location 9.1 In his new shoes, John walked through the barnyard.</Paragraph> <Paragraph position="2"> The above eqample asserts that at each instance during the walking, the walker was &quot;in&quot; with r@spect to the position of his shoes. Such examples require that the object's position at each instant during the eventlstate be compared to the location of the event at that instant.</Paragraph> <Paragraph position="3"> Ulowing for the preporititma]. concepts applying to place objects produced by the UNIT function must be done differently: 9.2 John went into the car.</Paragraph> <Paragraph position="4"> Exaq4-e-9-2 ite of thia class with the eemantic structure showing concept Im relating a part of the going to the location of the car. Here the position of the aiqle object must again be compared at each instant of the place object to the location of the place object at that instant. Rowever, only at the last instant wt the relation be ehown as holding.* Repositions predicating place objecte produced by the SEGMENT function are are caplex: 9.2 An ant is crawling up your arm.</Paragraph> <Paragraph position="5"> A s*le interpretation of the prepositional concept in the above may be problematic.</Paragraph> <Paragraph position="6"> Since your arm could be in motion, a stationary observer would include sare wtion attributable to your arm in the ant's path. Further, even if r wanted to take the position of the arm at erne one instant it is unclear wbich to take. These problems, however, dieappear with. the realization that the motion refereaced is not with respect to an arbitrary observer but to one on the am. For him, the SR can be treated as involving not a moving arm but one essentially static in space. This can be allowed for by requiring the conceptual definition of the prepositions to project the referenced objects' positions shown in the place object in. the P case onto the base object shown in the G case. This will be like taking the base as a static ground and the referenced object as a figure seen against it.</Paragraph> <Paragraph position="7"> *Since the change into the final state my be gradual and not dramatic, fuzzy relatioucl (~adeh, 1973) might be used.</Paragraph> <Paragraph position="8"> For instance, the degree of &quot;into&quot;' ednees could be quantified with the analysis showing that a certain degree was reached.</Paragraph> <Paragraph position="10"/> </Section> <Section position="13" start_page="0" end_page="2" type="metho"> <SectionTitle> FIGURE 9.4a The definition of ACROSS: higheat level. </SectionTitle> <Paragraph position="0"> The conceptual level definition of the prepositional cmaept ACROSS is sketched in Figures 9.4a, b, and c. Here again, it should be remembered that we are not trying t~ show the entire meaning of the prepositional form, only its relation to the place object. Accordingly, a number of unanalyzed forme will be used. Perhaps the most curious one of the forms will be one labelled acroes , this can be seen as the physical part of the concept ACROSS. It would have to be the next form developed if we are analyzing the meaning of the preposition.</Paragraph> <Paragraph position="1"> The definition of ACROSS begins in Figure 9.4a with an indication of the F and G nodes. The structural condition again shows the required connection between these two elements. It has three a1 ternative opportunities for eatisfaction, one for direct predication, one for UNIT functions, and one for SEGMENT funf tions.</Paragraph> <Paragraph position="2"> The three choices are reflected by the three arcs projecting from the OR node. Which case applies should be shown by one of the two arcs projecting from the corresponding AND node.</Paragraph> <Paragraph position="3"> The test for direct predication is identified by the object in the F role being pointed to by an Event/Statc.* This is shown by *It could just a8 well be pointed to by a WAY node.</Paragraph> <Paragraph position="4"> Thicl could be teetcd for with the addition of a disjunct.</Paragraph> <Paragraph position="5"> FIGURE.9.4c The definition of ACROSS: SEGMENT predication.</Paragraph> <Paragraph position="6"> a test labelled with these names. The condition that muat hold if we do have direct predication is shorn in Figure 9.4b. It states that for every placelet the physical across must hold between the space of the placelet and the space of the place object of a BEING-AT which locates the object identified by the object in the role at the ti* shown in the placelet. The superscripts on nodes in the figure establish co-reference between the different parts of Figure 9.4. The structure for the UNIT case is fairly similar and not shown. For the SEGMENT case, the condition is based on one space being phy~ical across from another. This is shown in Figure 9.4~. Both spices are shown being produced by a s~ecial proieqtion function which takes the place object in the F role and the object in the G role and produces the projection, shown by an arc of that name, and an abstract space to compare it to, shown by the Abstraction arc.</Paragraph> <Paragraph position="7"> To summarize, the section has shown how conceptual level interpretation can be given the semantic structures proposed earlier in the paper. Any system that uses the semantic structures can also use the interpretations. Of course, the interpretations are baaed on one way of structuring plade objects. Since there are other ways, other interpretations are possible.</Paragraph> <Paragraph position="8"> X. Limitation, Summary, and Conclusions There are definite limits to the claims we wish to make.</Paragraph> <Paragraph position="9"> In this concluding section, we point out several half-solved and unsolved problems, one area where we could conceivably expand our claims, and then end with a summary and final defense.</Paragraph> <Paragraph position="10"> Metaphorical usages are i~lportant but difficult subjects for semantic representation. Things like &quot;climbing the ladder of success&quot; are far enough away from spatial reference to be ignorable. However, some SR phenomena appear to be metaphors: 20.1 John yelled his greetings to John.</Paragraph> <Paragraph position="11"> In the above, an imaginary object, &quot;his greetings&quot;, seems to be sent through space. In the.following, a hypothetical journey is referenced: 10.2 The bridge goes from New York to New Jersey.</Paragraph> <Paragraph position="12"> Any direct representation of these phenomena using the definitions from the lasf section is unlikely since a non-instantaneous time interval must be present, while these sentences are basically instantaneous. My best suggestion is to represent these using the motional structures but to indicate by a function or operation applied to every appropriate foa that the actual sense is metaphorical. null Unfortunately, this leads to odious complexity. It is probably better to say our claims stop at this point.</Paragraph> <Paragraph position="13"> Many adverbial8 qan apply to modify SR'S or show spatial-like properties of events and states. We have no definite analyses for these, either. An incomplete study indicates that these may be analyzable within our model. For example, some forms can be shown as modifiers of a prepositional concept, such as in the following: -3 I put the ball completely under the car.</Paragraph> <Paragraph position="14"> Some seem to predicate place objects directly: 10.4 I walked two miles.</Paragraph> <Paragraph position="15"> Others seem to coordinate with SR's: 10.5 Go straight into the house.</Paragraph> <Paragraph position="16"> Here &quot;straight&quot; can be shown aa predicating the part of the journey up to the time the house was entered. However, I do not know how many other terms remain to be considered.</Paragraph> <Paragraph position="17"> Two problems remain completely unsolved. The first involves relative motion.</Paragraph> <Paragraph position="18"> 10.6 The ant walked over the rising pile of dough.</Paragraph> <Paragraph position="19"> Now, it is actually possible that-10.6 can be true but that the dough changed during the walking. Since there is no one static pile of dough, this would make which problematic the use of the dough as the object into,,the motion of the ant is projected. Secondly, it appears that inferences vary in habitual sentences: 10.7 He bought a present for her in New York.</Paragraph> <Paragraph position="20"> 10.8 He always buys a present for her in New York.</Paragraph> <Paragraph position="21"> The last examples differ because the former says an event occurred in New York but the latter says that a certain type of event must occur when the person is in New York. I have solutions for neither problem. We can only appeal to the fact that these phenomena do present problems in many other areas of semantics. Switching from difficult to promising areas, one strong possibility exists for expanding the analyeis and corroborating it.</Paragraph> <Paragraph position="22"> As the analysis was developing, time could be seen t-9 become more closely associated with place.</Paragraph> <Paragraph position="23"> In the end, time was claimed to fit it every place object. Perhaps with our place object, no separate time attribute needs to be associated with events or states of affairs. We may be able to claim that, to quote E. J. Lemon (19671, we can associate I I events with space-time zones&quot; instead of times and spaces.</Paragraph> <Paragraph position="24"> How this would be done remains to be seen.</Paragraph> <Paragraph position="25"> However, if we have not already met our gdal of putting space on a par with tie, that would certainly do it.</Paragraph> <Paragraph position="26"> In suampry, this paper has ahown how the .emantic rtructure of 8pati.l reference8 can be shown as locating events and states of affairs. within a semantic net, this has the form of showing a location ae an attribute of event/state nodes. In line with this, the concept of a place object, showing where events and states of affairs held at instances of time, was developed. Several functions were developed for use in predicating locations. Inferencing of spatial facts, the use of prepositional-like concepts for showing spatial relatibnships, and the overall semantic structure of utterances was also discussed. null Throughout the paper, the main jus tif ication has been that the analysis handles phenomena that other analyses do not. However, there are other jurtifications. Only one source for space simplifies the modeling of spatial phenomena. Using only static forms simplifies the interpretation of spatial terms. Also, the use of static forms fits in with proposals for state-based semantic representations (~ercone and Schubert, 1975). Finally, we can see that the analysis of )I semantic structures, in general, fits in with deeper&quot; analyses of semantic structure such as Schank (1973) and Norman and Rumelhart (1975). In sum, there appears to be a strong case for the analysis.</Paragraph> </Section> <Section position="14" start_page="2" end_page="2" type="metho"> <SectionTitle> Acknowledgements </SectionTitle> <Paragraph position="0"> I wish to acknowledge the help of David Brown on an earlier draft of thia paper and the even earlier advice of Richard L. Venezky and Peter S. Schreiber who launched me into time and epace.</Paragraph> <Section position="1" start_page="2" end_page="2" type="sub_section"> <SectionTitle> Resident Carter's Reorganization PZan Jo. 1 of 1077, dealing with the </SectionTitle> <Paragraph position="0"> reorganiaat ion of the Executive Office of the President (EOP) , including bodies dealing with telecommunications, computers and information policies.</Paragraph> <Paragraph position="1"> Reoqanizutia Ph No. I, submitted July 15th, is the first of a series of plans to be presented by the ?resident reorganizing the Executive Branch of Governutent. The proposals directly affect the organization of groups within EOP including the Office of Telecommmicatians Policy (UP) ; the 0fiice of Science and Technology Policy (OSTP) ; the Intergovernmental Science, Engineering, and Technology Advisory Panel (ISETAP) ; the President s Cmittee on Science and Technology (PCST) ; and the Federal Coordihating Cameil for Science, ~ngineeking and Technology (FCCST) . The plan becomes law unless vetoed by Congress in 60 days.</Paragraph> <Paragraph position="2"> Details of lReorganization Plan No. 1. In hearings hst month before both the Subconittee on Legislation and National Security of the House bittee oh Government Operations and the Senate Conunittee on Governmental Affairs, Office of Management and Budget (OMB) Director Bert Lance contended that the Reorganization would strengthen the Cabinet form of Government. Mr. Lance added that it would also insure that &quot;interested individ~als~~ in the departments would be consulted early in the decisionmaking pmcess .</Paragraph> <Paragraph position="3"> In short the plan creates (in lieu of the present White Home Domestic Council) a &quot;Domestic Policy Staff1' composed of ad hoc working groups of Cabinet and agency officials who, under the Vice President, are charged with setting priorities among the issues.</Paragraph> <Paragraph position="4"> Specifics of the plan as it affects telecomunications, computers and infmtion pol icies groups within the EOP are: It abolishes OTP.</Paragraph> <Paragraph position="5"> It transfers varkous functions of OTP to the President.</Paragraph> <Paragraph position="6"> It grants the Domestic Policy Staff authority, once associated with OTP, to review policy options requiring presidential decisions.</Paragraph> <Paragraph position="7"> - It delegates to OMB responsibility for establishing policy for Government procurement of telec~nicatias facilities and services, formerly a function of UI'P.</Paragraph> <Paragraph position="8"> - It transfers all other functions of OTP, including policy development and the allocation and regulation of frequency assignments, to the Department of Commerce.</Paragraph> <Paragraph position="9"> - null It creates a new position of Assistant Secretary of Cbnmerce for Communicatfons and Information &quot;to serve as spokesman for the Administration on telec~unicat ions issues , and [to] assume responsibility for the functions transferred from OTP and those of the Office of Telecomunications in Commerce,ff according to OMB Director Lance.</Paragraph> <Paragraph position="10"> - It retains OSTP but transfers all functions vested in the director of OSTP to the President, who may than redelegate various functions.</Paragraph> <Paragraph position="11"> - It abolishes ISETAP, PCST and FCCST [established. with OSTP through enactment of the NationuZ Science and TeehZogy IbZicy, UrganCsation, and Prioritis8 Ad of lg76), and transfers these organizations' functions to the President, who may then redelegate some or all of them.</Paragraph> <Paragraph position="12"> Criticism of 'Reorganization Plan No. 1.' Congressional criticism of heorganization PZun No. I, as it pertains to teletommunications, computers and infomation policies groups, centered on the creation of the new Assistanit Secretary of Commerce for Communications and Information. Sen. Abraham A. Ribicoff (D-Conn.), chairman of thd Senate Committee on Government Affairs, suggested there wi 11 be &quot;a lack of coordinationv of telecommunications policies with the new Assistant Secretary of Commerce outside the White House. Sen. Ribicoff also expressed concern that the Assistant Secretary would &quot;get lost in the shuffle.&quot; In addition, Rep. Jack Brooks (D-Tex.) , chairman of the House Committee on Government Operations, questioned how the Assistant Secretary-designate, 'hot the juiciest job in the United States,&quot; could coordinate Govermnent;wide policy.</Paragraph> </Section> </Section> <Section position="15" start_page="2" end_page="2" type="metho"> <SectionTitle> BILL REGULATING EFTS ImRODUCED IN HOUSE </SectionTitle> <Paragraph position="0"> A bill that would extend Federal regulation to include control over electronic funds transfer systems (EFTS) was introduced in the House of ~e~resentatives in July by Rep. Wary Rose maker (D-Ohio) .</Paragraph> <Paragraph position="1"> The bill, B.R. 8387 (#I, would expower the Comptroller of the Currency to oversee EFTS for national banks; the Federal Home Loan Bank Board to oversee EFTS for savings and loan associations; and the Federal Ekposit Insvrance Corporation, the Federal Reserve, and the Nadonal Credit Union Administration to oversee EFTS for their member institutions.</Paragraph> <Paragraph position="2"> The bill incorporates some of the legislative recommendations of the National Comission on Electronic Fund Transfers (NCEFT) conkained in the NCEFTWs Fehuary, 1977, interim report. (The final NGEFT report is due to be released next month.) H.R. 8387 was referred to the House Committee on Banking, Finance and Urban Affairs in July.</Paragraph> <Paragraph position="3"> SEPTEMBER, 1977 AFIPS WASHINGTON REPORT The Department of Justice, the Ad Hoc Tblecormnunications Users Committee, ATGT, IBM Corp., the Computer and Business Equipment Manufacturers Association (CBEMA) , the Computer Communications Industry Association (CCIAJ, and the Association of Data Processi'ng Service Organizations (ADAPSO) have all filed recent comments in the Federal Comrmmications Comiss i on ' s (FCC] Second Canputer Inquiry.</Paragraph> <Paragraph position="4"> The InquZq is considering the role of a regulated monopoly, ATBT, and otherregulated CQmon carriers in providing unregulated data processing services, generally torbidden by the FCC.</Paragraph> <Paragraph position="5"> In filings last May, the Justice Department opposed broad regulation by the Commission of unregulated firms which, though they may compete in certain areas with AT6T, are not now regulated by the FCC. Justice also said it &quot;strongly supports the Commission' s intentions to base its rules on marketplace standards, rathet. than simply technolbgical standards.&quot; lle Ad Huc Telecommunications Users Committee (composed of 15 users companies such as the Ford Motor Co. and Sears, Roebuck & Co.) oppased &quot;a prohibition against the usc of new technology for communications purposes.&quot; The Users Committee suggested that prohibitions against regulated carriers' entry into unregulated data processing activities should be confined to the lJpurpose and effect of the services provided,&quot; rather than be based an the processes or equipment used.</Paragraph> <Paragraph position="6"> ATET argued against anbwoverly restrictive viewff of &quot;communications common carriage.&quot; IBM, CBESlA, CCIA and ADAPSO, however, supported a broad definition of data processing, leaving it unregulated, and opening the field to more competition, they said.</Paragraph> <Paragraph position="7"> FEDERAL MINICOMPUTER. TERMINAL STANDARDS TO BE ADOPTED BY NBS A new Federal interface standard, called '%S-XYZ;&quot; is planned by the National Bureau of Standards (NBS) to replace the current RS-232C interface between terminals and computers (especially minicomputers) and data communications equipment. According to a June 27, 1977, article in ComputemorZd, the RS-XYZ standard should be implemented this month. In addition, an eight-bit ASCII code (in lieu of the usual seven-bit code), and protocols standardizing the user-terminal interface, are under consideration by NBS. The eight-bit ACII code could be implemented as a Federal standard early iq 1978.</Paragraph> <Paragraph position="8"> A Federal Basic standard is also expected for 1978.</Paragraph> </Section> <Section position="16" start_page="2" end_page="2" type="metho"> <SectionTitle> FIRST NATIONAL FACILITY FOR ELECTRONIC DEVICES RESEARCH ESTABLISHED AT CORNELL BY NSF </SectionTitle> <Paragraph position="0"> The first national facility'for research on electronic devlces with dimensions of less than one micron is being established at Cornell University in, Ithaca, New York, with a five-year, five million dollar a grant from the National Science Foundat ion (NSF) . The faci li ty , known as the National Research and Resource Facility for sub-Micron structures, SEPTEMBER, 1977 3</Paragraph> </Section> <Section position="17" start_page="2" end_page="74" type="metho"> <SectionTitle> AFIPS WASHIKGTON REPORT </SectionTitle> <Paragraph position="0"> is designed to find better ways to produce tiny patterns that can be incorporated into various electronic devices.</Paragraph> <Paragraph position="1"> The technology, planned for development in the new facility, could allow researchers to increase the density of componengs in an integrated circuit by as much as 10 times.</Paragraph> </Section> <Section position="18" start_page="74" end_page="74" type="metho"> <SectionTitle> SECOND ANNUAL REPORT OF THE PRESIDENT RELEASED ON 'FEDERAL PERSONAL DATA SYSTEMS SUBJECT TO THE PRIVACY ACT OF 1974' </SectionTitle> <Paragraph position="0"> As required by the Privacy Act of 1074, President Carter in June submitted his Seaond Annuat Report of Fedem2 Personut Data Systsm Subject to tho Morn& Act of 1974.</Paragraph> <Paragraph position="1"> The teport covers personal data systems maintained by the Executive Branch in 97 agencies during calendar year 1976. The study concludes that there was no &quot;significant change1' in the &quot;scope and naturetf of Federal persoqal data systems, nor was there any &quot;significant changew in the use of computers to process personal data.</Paragraph> <Paragraph position="2"> Specific findings include: (1) Ninety-seven agencies maintained 6,753 personal data systems containing 3.85 billion individual records at the end of 1976, a net increase of 11 agencies and 30 systems, but a net decrease of 34 million individual records from 1975; and (2) at the end of 1976, 29 per cent of the personal data systems and 74 per cent of tho individual records were fully or partial.1~ computerized, as compared to 27 per cent and 79 per cent, respectively, at the end of 1975.</Paragraph> </Section> class="xml-element"></Paper>