File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/04/c04-1198_metho.xml
Size: 21,772 bytes
Last Modified: 2025-10-06 14:08:46
<?xml version="1.0" standalone="yes"?> <Paper uid="C04-1198"> <Title>Contextual processing of locative prepositional phrases</Title> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> 2 Locative prepositions are </SectionTitle> <Paragraph position="0"> contextual functions A lot of examples show that &quot;y&quot;, as a pronominal adverb, may have a NP as syntactical antecedent: null (5)Je vois la table. Tu y as pos'e tes clefs.(I see the table. You put your keys there.) If we replace the pronoun with its referent, we have: (6)Je vois la table. Tu as pos'e tes clefs sur la table.(I see the table. You put your keys on the table.) But, differences appear between this last example and a classical anaphoric relationship: first of all, the preposition &quot;sur&quot;/&quot;on&quot;, appearing in the rebuilt LPP in (6) does not come from either the antecedent (since this one is a NP), or the verb (&quot;sur&quot;/&quot;on&quot; is not a mandatory preposition for the verb &quot;poser&quot;/&quot;put&quot;, since this one accepts other prepositions like in &quot;poser dans&quot;/&quot;put in&quot; or &quot;poser sous&quot;/&quot;put under&quot;). Moreover, the coreference link between the pronoun and its antecedent (usually generated by the anaphoric relation) is unclear: can we really assume that &quot;y&quot;/&quot;there&quot; refers to the object designated by the NP &quot;the table&quot;? A first answer to these problems could be that there is a kind of metaphoric interpretation of the object designated by the antecedent NP, which would allow it to designate also a place, for instance the location where it stands, or its interior when the name of the object clearly suggests an interiority feature. This approach was studied and approved by (Mackenzie, 2004) and similar ideas (though in the conceptual domain) can be found in (Flageul, 1997). Nevertheless, we can notice that it is rare for the pronoun &quot;y&quot; to designate the place defined this way. In example (5), one cannot say that &quot;y&quot; refers to the place occupied by the table. Moreover, such an interpretation of the involved phenomenon does not allow to account for the following examples: (7)Va chercher pr`es de la table, je crois que ton ballon s'y trouve.(Go look near the table, I think your ball is there.) (8)Je suis all'e derri`ere la maison. Un v'elo s'y trouvait.(I went behind the house. A bicycle was there.) (9)Quand j'ai pos'e mon sac `a c^ot'e de la tasse, ton portable s'y trouvait.(When I put my bag next to the cup, your laptop was there.) If we replace the pronoun with its antecendent, we have: (10)Va chercher pr`es de la table, je crois que ton ballon se trouve pr`es de la table.(Go look near the table, I think your ball is near the table.) (11)Je suis all'e derri`ere la maison. Un v'elo se trouvait derri`ere la maison.(I went behind the house. A bicycle was behind the house.) (12)Quand j'ai pos'e mon sac `a c^ot'e de la tasse, ton portable se trouvait `a c^ot'e de la tasse.(When I put my bag next to the cup, your laptop was next to the cup.) The difference between these examples and (5) resides in the fact that, here, the preposition employed in rebuilt examples (10), (11) and (12) is necessarily the same as the preposition involved in the LPP of the previous sentence. We notice that the syntactical similarity between the antecedent LPP and the rebuilt one does not come from a pure syntactic relationship between the pronominal adverb and its antecedent. In addition, there really is a coreference relationship since, in example (10) for instance, both LPPs mention the same table and, in fact, the same 'place'. Therefore, we have to account for the coreference link between pronoun &quot;y&quot; and a 'place', described by the antecedent LPP. We must then consider that &quot;y&quot; is anaphorically related to the LPP and, hence, refers to an 'object' introduced by this LPP and different from that designated by the NP.</Paragraph> <Paragraph position="1"> Studying where-questions raises a valuable analogy. For instance, let us consider a static locative question like: (13)O`u se trouve mon portable ?(Where is my laptop?) One possible answer to this question, within a universe described by example (9) above, is: (14)`A c^ot'e de la tasse.(Next to the cup.) If we classically consider that the semantics of such questions (wh-questions) is the maximal set of individuals satisfying the property corresponding to the question (on this subject, see (Van Rooy, 1999), (Groenendijk and Stokhof, 1984)), then we must admit that &quot;`a c^ot'e de la tasse&quot;/&quot;next to the cup&quot; is an individual. As a matter of fact, the semantics of question (13) is: {x/se trouve(mon portable,x)}2 and one element of this set is therefore an object that can be designated by the LPP &quot;`a c^ot'e de la tasse&quot;/&quot;next to the cup&quot; (which is clearly a different object from the one designated by the NP &quot;la tasse&quot;/&quot;the cup&quot;).</Paragraph> <Paragraph position="2"> Thus, we conclude that LPPs introduce individuals that can be viewed as answers to locative questions. Then it appears that these individuals are precisely those that can be designated by adverb &quot;y&quot; in situations similar to the examples given above. Other examples give support to this point of view. The anaphoric use of locative adverb &quot;l`a&quot;/&quot;there&quot;3 or repetitions by definite description show us other cases that require to assume the existence of LPPsgenerated individuals: (15)Je suis en face du cin'ema. On se retrouve l`a.(I am in front of the cinema. Let's meet there.) (16)Je suis `a droite de l'arbre. L'endroit est magnifique.(I am to the right of the tree.</Paragraph> <Paragraph position="3"> The place is beautiful.) Therefore, we give the following rule: Some4 LPPs introduce/designate two individuals: one classical, given by the contextual interpretation of the embedded NP, the other given by the interpretation of the LPP itself (that is, by the compound [prep + NP]).</Paragraph> <Paragraph position="4"> The pronominal adverb &quot;y&quot; may be anaphorically linked to a LPP (that is, precisely, to the phrase made of [prep+NP]), thus referring to the individual introduced by the complete LPP.</Paragraph> <Paragraph position="5"> Then, what happens when &quot;y&quot; has a NP for an antecedent? The NP is its antecedent but it has no coreference relationship with it. On can think here that we are facing a relationship close to bridging anaphora (like &quot;la porte&quot;/&quot;the door&quot;...&quot;la poign'ee&quot;/&quot;the handle&quot;). But instead of having a classical part/whole relationship between the two members of the anaphora, we have an underspecified (the preposition is missing) relationship of type object/place. &quot;y&quot; involved in this rule is a difficult question, unanswered in this paper.</Paragraph> <Paragraph position="6"> is therefore involved in a coreference relationship with another object than the one designated by its antecedent. The semantics of this object depends on the antecedent NP and a (yet undetermined) preposition. We notice that even if the antecedent NP is embedded in a LPP, the preposition of the latter is not always referred to by the pronoun. The following example: (17)Max habite loin du parking mais j'y ai (quand m^eme) gar'e ma voiture.(Max lives far from the carpark, but I (nevertheless) left my car there.) can be paraphrased by: (18)Max habite loin du parking mais j'ai quand m^eme laiss'e ma voiture dans le parking.(Max lives far from the carpark, but I (nevertheless) left my car in the carpark.) In these situations, the preposition &quot;far from&quot; is not part of the anaphoric relation. We will say that the syntactic antecedent of the pronoun is the embedded NP (and not the LPP) and we will introduce a new locative preposition to calculate the referent of &quot;y&quot;/&quot;there&quot;. We notice that this process makes the interpretation of pronoun &quot;y&quot; potentially ambiguous when this one refers to a LPP: the pronoun can indeed refer to the LPP or have the NP (embedded in the LPP) as an antecedent. The following example shows such an ambiguity: (19)J'habite `a c^ot'e de Paris et j'y travaille.(I live near Paris and I work there.) There are two possible interpretations for this sentence: Finally, we notice that the behavior of relative sentences introduced by &quot;o`u&quot;/&quot;where&quot; is similar to the examples above, except the fact that the antecedent is always a NP: (22)Je vois la table o`u tu as pos'e ton portable.(I see the table where you put your laptop.) (23)Max habite loin du parking o`u j'ai laiss'e ma voiture.(Max lives far from the carpark where I left my car.) The antecedent of the relative pronoun &quot;o`u&quot;/&quot;where&quot; is the NP which is embedded in the locative complement and its semantics must be completed by a (implicit) locative preposition (it could be replaced in surface forms by &quot;on which&quot;/&quot;in which&quot;).</Paragraph> <Paragraph position="7"> Now that we showed that some LPPs introduce/designate one discourse individual, we must have a closer look at the underlying phenomenon. Many studies dealt with contextual effects of NPs (definite/indefinite, generic/specific, etc. see, for instance, (Corblin, 1987)). A summary of these can be found in (Salmon-Alt, 2001). Where do LPPs fit in there? First of all, we notice that the situation is complicated since it depends obviously on the nature of the NP which is embedded in the LPP and, for example, some cases of &quot;generic LPPs&quot; can be encountered, like in: (24)Habiter pr`es d'un cr'ematorium comporte 'egalement des risques : la mortalit'e natale y est plus 'elev'ee de 4%.(It is also risky to live near a crematorium: birth mortality is higher by 4% there.) We will firstly focus on LPPs containing two types of NPs: indefinite with specific interpretation and definite descriptions (or any referential expression, for instance pronominal). In both cases we get, after the interpretation of the embedded NP, an entity, either pre-existing the interpretation of the NP or newly created. This entity serves as a basis to determine the new entity generated by the interpretation of the LPP. Clearly, the other tool allowing to determine this new entity is the locative preposition involved in the LPP: what is its status? Several studies on space representation give definitions (either qualitative or quantitative) of the part of space designated by a spatial function applied to an object or a place. These works usually focus on explaining the conditions under which one can say that a point of space (or an object) is or not within such a part of space (see, for example (Vieu, 1991), (Schang, 1997)). This belonging is then most often represented by a spatial relation (like to be on the right of) between objects, which will be either true or false. These processes do not highlight this part of space that we aim to raise as a discourse object, but they show the matching between some locative prepositions and spatial functions5. Other 5Here, again, it seems difficult to state that every locative preposition matches a spatial function. This question will not be discussed in this paper. But, such a study could probably rely on the classification of locative prepositions proposed in (Borillo, 1993) and (Borillo, studies also highlight this role of locative prepositions. V. Flageul (Flageul, 1997) considers locative prepositions as conceptual-level operators, which turn a type object into a type place. A. Peyraube (Peyraube, 2003) describes the historical evolution of Chinese localizers and notes that &quot;their main function in pre-medieval Chinese is to follow nouns and transform them into place words, tending to behave like functional words&quot;. We take one step further and state that locative prepositions are contextual-level operators (or &quot;functions&quot;, from a logical point of view) which produce an object from another, the former being the part of space cited before.</Paragraph> <Paragraph position="8"> The work of V. Flageul allows us to specify the conceptual type of this object according to the involved preposition.</Paragraph> <Paragraph position="9"> Concerning LPPs that contain NPs quantified by 'every', the situation slightly differs: typically, such NPs imply universally quantified variables. We will get back on this point when examining examples in the next paragraph.</Paragraph> <Paragraph position="10"> As a conclusion for the current section, we state that in the situations described above: Some locative prepositions denote contextual functions: they associate to one individual X from the context another one, noted f(X). Some LPPs produce an object f(X) where f is the function associated to the locative preposition and X, the object designated or produced by the embedded NP.</Paragraph> <Paragraph position="11"> 3 Processing preposition functions with DRT and DPL We are now going to show how to integrate these functions when representing discourse and referential links. First of all, we show on examples6 how this works within DRT-like formalism. Then we show results obtained in a DPL approach.</Paragraph> <Paragraph position="12"> 2000).</Paragraph> <Paragraph position="13"> 6The examples we use here are simple on purpose and may therefore seem odd or ill-formed. Nevertheless, we can give many others where &quot;y&quot; refers to a LPP. A brief search on the Internet allowed us to reveal examples of various types: (25)Nous nous sommes promen'es pr`es d'un volcan et nous y avons trouv'e des roches int'eressantes.(We had a walk near a volcano and we found interesting rocks there.) (26)Les terrains primitifs, en faisant irruption dans les terrains primordiaux, y ont amen'e un certain nombre de min'eraux int'eressants.(Primitive terrains, when rushing into primordial terrains, brought a certain number of interesting minerals there.) DRT ((Kamp, 1981)) is a theory proposing a discourse representation formalism, essentially oriented towards (pronominal and others) reference problems. We show here with the cases studied in the previous section how a LPP and a referent to a LPP can be accounted for within this formalism.</Paragraph> <Paragraph position="14"> (30)Un arbre se trouve derri`ere une maison.</Paragraph> <Paragraph position="15"> Il y fait de l'ombre7.(There is a tree behind a house. It casts shadow there.)</Paragraph> <Paragraph position="17"> Comments: indefinite NPs &quot;un arbre&quot;/&quot;a tree&quot; and &quot;une maison&quot;/&quot;a house&quot; classically introduce two new individuals in the Discourse Representation Structure (DRS). The LPP introduces a new individual z, being the result of the application of function behind on reference marker y.</Paragraph> <Paragraph position="18"> Example continuation: (27)Au moment du projet, la densit'e des moules z'ebr'ees,..., 'etait de faible `a mod'er'ee pr`es de Burritts Rapids, mais elle y augmentait rapidement.(At the time of the project, the density of zebra mussels,..., was low to moderate near Burritts Rapids, but it was quickly increasing there.) (28)Durant le projet, c'est pr`es de chaque section qu'on a observ'e les plus bas niveaux de bact'eries E. Coli de toute la rivi`ere. Le niveau d'E. Coli y 'etait assez faible pour que la baignade y soit permise presque tous les jours.(During the project, the lowest levels of E. Coli bacteria in all the river were observed near every section. The level of E. Coli was low enough there to permit swimming on most days.) (29)Les gens pr'ef`erent acheter loin du centre ville car ils y trouvent des logements peu chers.(People prefer buying far from downtown because they find cheap residences there.)</Paragraph> <Paragraph position="20"> Comments: classically, pronoun &quot;il&quot;/&quot;it&quot; introduces a new reference marker v linked to its antecedent by the equality v = x. &quot;y&quot;/&quot;there&quot; also introduces a new reference marker w linked to its antecedent by equality w = z. The fact that the antecedent z is itself linked to a functional term is without consequence.</Paragraph> <Paragraph position="21"> (31)Si un arbre se trouve derri`ere une maison, il y fait de l'ombre.(If a tree is behind a house, it casts shadow there.)</Paragraph> <Paragraph position="23"> Comments: this example is equivalent to a donkey sentence. Thanks to its location inside the DRS, reference marker z is accessible from variable v.</Paragraph> <Paragraph position="24"> (32)Un arbre se trouve derri`ere chaque maison et y fait de l'ombre.(A tree is behind every house and casts shadow Comments: we chose here the interpretation in which the quantifier &quot;chaque&quot;/&quot;every&quot; has a scope beyond the indefinite subject. We notice that the quantification associated to variable y is existential (from its presence in the right box of the implication) and not universal, like it would have been the case if variable y was present in the left box. This choice is not really important since the result of a function is, by definition, unique.</Paragraph> <Paragraph position="25"> Through these examples, we can see that repetitions of functional terms are always done within the scope of the argument quantification: indeed, either both terms are introduced in the same box - at the same level - or the functional term is introduced in the right box of an implication (it is therefore in the scope of the variable introduced in the left box). To remain correct, every reference to this functional term will occur in the same context.</Paragraph> <Paragraph position="26"> These processes are thus classical. Functional terms just have to be introduced in DRT notation. In the &quot;classical&quot; DRT definition (like it is presented, for instance, in (van Eijck and Kamp, 1997), p. 191), terms are either constants or variables used as reference markers or discourse referents. We added functions in the language. We assume that a set of function letters with their arities is given. Then the set of terms is augmented by functional terms. Practically, these functional terms are obtained by applying functions on variables.</Paragraph> <Paragraph position="27"> Thus, a locative preposition corresponds to a function from this language. One can see to which typed lambda-expressions lexical items or phrases must be associated to obtain the results given above. According to notations from (van Eijck and Kamp, 1997) p. 217, we can state that a locative preposition is of type < e,e > and has as lambda-expression lv (behind v) if behind is the function associated to the preposition. The LPP causes the new individual to appear in the DRS. It is of type << e,T > ,T > (like a NP) and its lambda expression is lP(ui;(NP lv(ui = (Loc Prep v) . P(ui)))) where NP denotes the lambda-expression given from the LPP-embedded NP, Loc Prep is the lambda-expression of locative preposition given above and ui the reference marker introduced by the LPP. Finally, pronoun &quot;y&quot; will be associated to lambda-expression lP(P ui) where ui is its antecedent reference marker8.</Paragraph> <Paragraph position="28"> 8We can not detail here the lambda-expressions associated to other phrases or items required for complete DPL (Groenendijk and Stokhof, 1984) is another discourse representation system allowing easy expression of referential links. To account for our LPP processing, the language must be augmented with functions and functional terms (which raises a priori no problem). Here are some DPL formulas corresponding to the following examples: (33)Un arbre se trouve derri`ere une maison.(A tree is behind a house.) [?]x [?]y [tree(x)[?]house(y)[?]is(x,behind(y))] (34)Il y fait de l'ombre.(It casts shadow there.) [?]x [?]y [tree(x) [?] house(y) [?] is(x,behind(y))] [?] casts shadow(x,behind(y)) The occurrence of the variable in the functional term is linked by its quantifier, like in case of pronominal repetition without function. The well-formedness of formulas is therefore guaranteed. null (35)Si un arbre se trouve derri`ere une maison, il y fait de l'ombre.(If a tree is behind a house, it casts shadow there.) [[?]x tree(x)[?][?]y [house(y)[?]is(x,behind(y))]] = casts shadow(x,behind(y)) In order to build these DPL formulas, we associate to locative prepositions the lambda-expression lv(behind v) of type < e,e >, and to LPPs the lambda-expression lP[NP ly[P [Prep Loc y]]] of type << e,T > ,T > ([ ] are used as application operator).</Paragraph> <Paragraph position="29"> These expressions are very close to those used in DRT, apart from the fact that they do not introduce reference markers. In the absence of these markers, the link between pronoun &quot;y&quot; and its functional antecedent is established by using the same variable appearing in the antecedent functional term and building a new functional term from the same function. The lambda-expression associated to &quot;y&quot; then becomes lP[P [Prep Loc x]] where [Prep loc x] is the antecedent of &quot;y&quot;.</Paragraph> </Section> class="xml-element"></Paper>