<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-0601">
  <Title>The representation of syntactically unexpressed com to nouns</Title>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Standard treatment in argument
</SectionTitle>
    <Paragraph position="0"> structure The standard treatment of syntactically oriented representations introduces some level of abstract description of argument structure in syntax, so that general grammatical relations can be accounted for in a straightforward way. Thus, content relations can be stated between arguments and their heads that differ from the surface syntactic ones. In HPSG, for example, control relations are expressed as coindexation between syntactic and argument values, in such a way that a single valence element realises two argument slots; and passive is (lexically) dealt with as a change in the correlation between valence elements and argument ones. 5 But this level where argument structure is usually represented is a direct projection (although neither simple, nor one-to-one) of the surface structure. That is to say there are difficulties in incorporating new argument slots when no correponding syntactic slot exists. And this is so simply because the argument values are not thought as full semantic representations but simply as the deep representation of syntactic structure. 6 Being syntactically oriented, the standard treatments of argument structure can only represent the types of complements that correpond to the simple twofold distinction mentioned in the previous section: strongly restricted obligatory complements and optional adjuncts. ~ Nonetheless, there are at least three aspects that are not considered in a satisfactory way: complements to verbs and to nouns which are not linked with a strong syntactic and semantic relation, but cannot be treated as ordinary adjuncts, i.e., intersectively (as for instance the kind of complements mentioned in footnote 4); complements to nouns generally, because they usually are optional (as in (1) to (3) above); and some complements to verbs which are manifestly optional (as in (5)). This is so because in the standard typing of arguments  argument structure is embedded within the content part of the sign, where a lot of terminolgy is used that comes from situation semantics.</Paragraph>
    <Paragraph position="1"> 7In HPSG for example the representation for the formet is strictly based on the coindexation between syntactic and argument values, whereas the treatment for the latter assumes an event structure for predication and intersects it with the optional adjunts (see Badia &amp; Colominas (1996)).</Paragraph>
    <Paragraph position="2"> there is no way to introduce arguments (or other kind of slots) that are not induced syntactically (or grammatically) in a restrictive way. A full theory of lexical semantics is needed to do so in a consistent way. For example, in Badia (1997) a particular case is considered and solved in an ad hoc way: the nominal translation (as in 9b above) is said to have an extra argument (to those inherited from the verb) so that its result reading can have its denotation properly derived. This of course is only adequate for a small class of the complements we have here in mind (see examples in (8) and (9) above) and does not derive from a general theory of argument structure.</Paragraph>
    <Paragraph position="3"> Such optional complements have only been treated in standard treatments by creating distinct lexical entries, where the omitted complement is either just not present in the argument structure representation or represented by an abstract existential quantification (without a correspondence to any syntactically realised element). However this treatment neither reflects the relation between the different lexical entries that exist for a single word, nor is adequate in avoiding lexical redundancy. An alternative option that could be considered is to treat those complements as adjuncts. However the treatment of adjuncts has not been fully developed yet, so that most of the proposals at hand are only adequate for intersective adjuncts. 8 Furthermore, complements to nouns are particularly difficult. From a semantic (and argument structure) point of view many of them are like complements to verbs, in that they contribute entities to the relation denoted by the noun; thus an adequate treatment might introduce a particular number of arguments for every relation class denoted by nouns (just as it is done for verbs) (Badia &amp; Colominas, 1996). But from the syntactic point of view the difficulty arises because they cannot be simply introduced in the valence lists (as verbal complements are).</Paragraph>
    <Paragraph position="4"> And a simple treatment as adjuncts is not very satisfactory either since it would increase the complexity of the semantic calculations.</Paragraph>
    <Paragraph position="5"> 8Non intersective adjuncts cannot be easily treated, particularly those that contribute a new entity involved in the relation denoted by the predicate (for some of them quite complex proposals have been developed within HPSG: e.g., Kaspers, 1994).</Paragraph>
  </Section>
  <Section position="6" start_page="0" end_page="7" type="metho">
    <SectionTitle>
4 Proposed treatment
</SectionTitle>
    <Paragraph position="0"> In the following we attempt to account for some optional complements of predicative nouns in HPSG, by enlarging the semantic information and representing these complements in an information level distinct from the valence lists. With regards to our semantic representation proposal, we maintain the referential information level in HPSG, expressed by means of the INDEX attribute. However we follow Pustejovsky (1995) in enriching the argument structure information level (expressed in HPSG as the RESTRICTION attribute and adapted here as ARGUMENT-STRUCTURE (ARGSTR)), SO that Dand S-Args can be represented (in addition to the strictly obligatory ones). Furthermore we also introduce the EVENT-STRUCTURE (EVSTR) and QUALIA levels of information estipulated in Pustejovsky (1995). The latter describes the lexical semantics of the entity and its related arguments, whereas the former displays the event properties of the expressed eventuality. As will be seen below, the argument structure level acts as interface between the mechanisms that allow for the syntactic realisation of the complements and the rich lexical semantics -constituted by the EVSTR and QUALIA levels. 9 Let us now consider the syntactic treatment of D- and S-Args. Due to their optionality, the HPSG standard treatment of obligatory complements by means of valence lists is not adequate for them. Nonetheless a recent proposal by Sanfilippo (1998) gives a good insight into how they can be treated. For independent reasons he proposes that some complements are treated as real adjuncts from a syntactic point of view, even if they are thematically bound to the relation denoted by the head. We adopt this mechanism and represent D- and S-Args as thematically bound adjuncts, introduced as set members at the nonlocal (NONLOC) information level. Thus the basic structure of the sign is as follows: ldeg 9In fact our implementation follows the lines developed by Copestake (1992), Johnston (1996), among others. Particularly we follow the formalisation of Pustejovsky's semantics in feature structures developed in Johnston (1996). In his treatment the selection of the appropriate reading in aspectually-based verbal polysemies is expressed by means of a boolean value in the head attribute of each of the relevant qualia (see figures below). ldegFor reasons of space we do not include the LOC at- null nominal such as construccid ('building'). It is a nominalization from the verb construir ('build'), which subcategorises for two obligatory complements: the agent and the result of the building process.</Paragraph>
    <Paragraph position="1"> As indicated in Pustejovsky (1995) this verb has also a third argument expressing the material, a D-Are, which is syntactically optional but participates in the logical expression of the event (see example (7a) above). In the representation for that verb (figure 1), the first two complements are expressed in the valence lists, whereas the latter is treated according to Sanfilippo's proposal for thematically bound adjuncts; that is, as a member at the nonlocal information level. In figure (2) we show the process nominalisation of construir where not only the D-Are but also ARG1 and ARG2 are not declared in the valence lists but included in the nonlocal set. Recall that we treat them like DAres. null For our implementation we use LKB (Copestake, 1993). We first derive the process reading of the nominal from the verbal lexical entry: the nominal representation for the process reading (in figure 2) is obtained by (non-locally) inheriting from the already existing lexical entry of the verb (in figure 1). 11 And secondly the lexical entry for the result sense of construcci6 is created by means of a lexical rule that replaces the EVSTR head value and removes the ARG2 attribute, the value of which is now declared as the index value. ARG2 has to be removed since, as it expresses the entity resulting from tribute in the rest of the structures in the paper. nThis is expressable in LKB by using the verbal lexical entry as a pseudo-sort (psort) (see Copestake  the process of building, it cannot appear as an argument of the result noun itself, n Thus the treatment of optional complements of nominalizations is identical to the treatment assigned to D-Ares of verbs. Both kinds of argument are logically implied by the semantics of the predicate (verb or noun) but are syntactically optional. They are distinguished from Pustejovsky's True-Arguments (T-Arg) -that is, the syntactically realized parameters of the lexical items (Pustejovsky, 1995:63)- by means of the syntactic level at which they are declared: in the valence list or in the set of nonlocal elements. null In order to show that other types of nominals can also be treated in this way, consider for innRecall that creation nominals do not allow the result nominalisation with the thematic argument: *la construcci6 de la casa ds molt sblida ('the building of the house is very robust').</Paragraph>
    <Paragraph position="2">  stance redescription nominalizations. They differ from standard creation predicates in that their process reading cannot express syntactically the argument denoted by the result nominalization. null Consider the noun decoracig ('decoration'), derived from the verb decorar ('decorate'). We start by assuming three arguments to that predicate: a first one corresponding to the agent, a second one that expresses the theme (i.e., the object being decorated), and finally a D-Arg referring to the material. The agent and the theme arguments are subcategorized as T-Args by the verb, but realized as thematically bound adjuncts when appearing in a nominalization.</Paragraph>
    <Paragraph position="3"> As opposed to construir, the direct object of decorar does not denote the object resulting from the decoration act, but the object being decorated. As a consequence both nominalisation of decorar (process and result) allow the  complement that corresponds to the direct object of the verb (recall that this is not possible in the case of result readings of creation predicates). Therefore the presence of the objective complement of decoraci6 is not useful for disambiguating between the two readings. Thus nominalizations like construcci6 and decoraci6 with the objective complement differ in that the first one does not accept the two possible senses appearing together, while the second one does.</Paragraph>
    <Paragraph position="4"> Following the treatment of Pustejovsky's dotted types in Buitelaar (1997), this fact leads us to conceive construccid as an open dotted type, and decoracid as a closed dotted type.</Paragraph>
    <Paragraph position="5"> The treatment we propose for decoracid is  shown in figure (3). Note that it is adequate to deal with facts such as the ones exemplified in (9a). In this example the redescription nominal (decoraci6) expresses the process reading in the first clause, and in the second one it is referred to as denoting the object resulting from the process. In the representation of the nominal decoraci6 in figure (3) this is allowed by the third argument of the AGENTIVE qualia, which is not bound by any argument in the ARGSTR, just because it cannot be syntactically realised in any case as an argument. Note that argument structure acts as an interface between the rich semantic representation (i.e., the QUALIA and EVSTR structures) and the surface mechanisms that allow for complements of the predicate (valence lists and non-local set). Only semantic arguments that may be syntactically realised are present in the argument structure (either as T-Args or as D- or S-Args). On the other hand the  rich semantic structure allows to express semantically implied arguments, and thus provides a treatment for semantically motivated discourse factors (such as the ones shown in (9) above).</Paragraph>
    <Paragraph position="6"> The proposal outlined here provides an appropriate treatment for the traducci6 example (9b). Instead of the ad hoc solution adopted in Badia (1997) the treatment proposed here derives from a general and systematic approach to the semantic structure of predicates and their nominalisations. To make things concrete we present in figure (4) the full lexical entry of the process reading of the nominal traduccid. This representation is able to deal with anaphoric reference to the unexpressed result semantic argument of the process reading (9b). Note that a parallel verbal lexical entry also gives an account of the anaphoric relation allowed by the verb traduir in (9c), and that a similar treatment is applicable to the verbs menjar and amanir of examples (5) and (9e).</Paragraph>
    <Paragraph position="7"> Of course this treatment applies in a general way to other redescription predicates, like estudi ('study'), soluci6 ('solution') and avaluaci6 ('evaluation') in (8). Notice as well that the rich semantic information we use allows us to deal in a natural way with the distinction between subjective and objective complements in these data. In (Sa), for instance, estudi does not allow plantes ('plants') as the agent of the event since tha agent has to be an animate individual: 13 Other types of nouns with semantically implied (optional) complements can be similarly treated; for example non deverbal nouns that express a relation with another entity, like nouns denoting partition or sets (full 'sheet' in (2) and grup 'group' in (3)) or relational nouns (mare 'mother' in (9d)). The arguments of full or grup can be treated as D-Args because they are not strictly obligatory but their semantics is implied. However the semantics of the syntactically realised complement need not correspond exactly to the semantic implication. For example, the implication of .full realised without a complement is that it is a piece of paper; however the complement may denote any sim*aNote that in the figure for estudi the HEAD values of EVSTR and QUALIA levels are underspecified, similarly to the value of the INDEX attribute, thus letting the entry underspecified between the process and the result inter- null ilar class of material (card, cardboard, tissue paper...) (Sauri, 1998). We account for this difference by means of a lexical rule that overwrites the implied semantics of the complement when necessary.</Paragraph>
    <Paragraph position="8"> This treatment however is not adequate for relational nouns, like mare ('mother') or pare ('father'). These nouns do not allow the surface realisation of the semantic implication unless it is more specific: (10) a. *Ha vingut el pare d' un fill *Has come the father of a son b. Ha vingut el pare d' un nen Has come the father of a boy canadenc Canadian c. Ha vingut el pare de la Joana Has come the father of the Joana This kind of complement can be seen as shadow arguments (S-Arg) of Pustejovsky's. In our proposal they are treated as D-Args because they are optional: their syntactic realisation is controlled via the non-local set and the semantic implication is expressed in the FORMAL qualia.</Paragraph>
    <Paragraph position="9"> However a restriction has to be formulated upon the semantics of the syntactic element within the non-local set: namely that it be more specific than the semantic implication (see figure</Paragraph>
  </Section>
class="xml-element"></Paper>