File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/03/p03-2010_metho.xml
Size: 17,312 bytes
Last Modified: 2025-10-06 14:08:20
<?xml version="1.0" standalone="yes"?> <Paper uid="P03-2010"> <Title>A Computational Treatment of Korean Temporal Markers, OE and DONGAN</Title> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> (1) ach'im ilgopshiOE </SectionTitle> <Paragraph position="0"> morning/seven o'clock-OE at seven o'clock in the morning (2) han shigan DONGAN one/hour/DONGAN for an hour As it is widely known, time adverbials play important roles in sentence meaning processing. Meanwhile, there is a significant divergence in opinions whether time adverbials or tense/aspect is a more efficient indicator leading to a correct temporal representation of sentences. To some [Kim, 1981], [Jo, 2000], [Vet, 1980], [Verkyul, 1989], tense or aspect is the only credible index to consult in establishing temporal interpretation, and the time adverbials are complementary. To others [Renaud, 1996], [Vlach, 1993], time adverbials are regarded as much more reliable than tense/aspect which is too ambiguous to provide coherent instructions about how to locate the event in time. We agree with the second point of view, as we observed that Korean tense markers fail to provide a solid and coherent way to capture the relevant time span. For example, the verbal infix '-at-', generally considered as a typical past tense marker in Korean, brings about several time interpretation possibilities such as simple past (3), completion (4), resultant state (5) and progressiveness (6). (3) shiwidaega oje hanshiOE shich'ongul dulrossatta demonstrators-NOM / yesterday /one o'clock-OE / the city hall-ACC/ surround-PA-DEC1 The demonstrators surrounded the city hall at one o'clock yesterday.</Paragraph> <Paragraph position="1"> (4) shiwidaega mach'imnae shich'ongul</Paragraph> <Paragraph position="3"> At last, the demonstrators surrounded (succeeded in surrounding) the city hall.</Paragraph> <Paragraph position="4"> (5) shiwidaega harudongan shich'ongul dulrossatta demonstrators-NOM/one day-DUR/the city hallACC/surround-PA-DEC null The demonstrators have surrounded the city hall for one day.</Paragraph> <Paragraph position="5"> 1 We used the McCune-Reischauer system to transcribe the Korean data. For glossing grammatical morphemes, we use the following abbreviations: ACC: accusative, AS: attributive suffix, CIRCUM: circumstantial, CL: classifier, DEC: declarative, DUR: durative, INT: interrogative, LOC: locative, NOM: nominative, NS: nominal suffix, PA: past, TOP:topic. (6) onjebuto shiwidaega shich'ongul dulrossatssumnikka? since when/ demonstrators-NOM/ the city hall-</Paragraph> </Section> <Section position="4" start_page="0" end_page="0" type="metho"> <SectionTitle> ACC/ surround-PA-INT </SectionTitle> <Paragraph position="0"> Since when have the demonstrators been surrounding the city hall? Moreover, what triggers these interpretation possibilities is still being discussed among Korean linguists2.</Paragraph> <Paragraph position="1"> In the following, we attempt to show how time adverbials can remedy this shortcoming and specify the event time.</Paragraph> </Section> <Section position="5" start_page="0" end_page="0" type="metho"> <SectionTitle> 2 Semantic description </SectionTitle> <Paragraph position="0"> The assumption underlying our temporal description is that the linguistic time is ordered, discrete, infinite and consisting of instants corresponding to the natural numbers. The linguistic time can be expressed with one of these three notions: instant, extended interval and duration. Instants are unitary constituents of linguistic time and noted by a quintuplet of natural numbers [x1,x2,x3,x4,x5] of which x1 stands for year, x2 for month, x3 for day, x4 for hour and x5 for minute.</Paragraph> <Paragraph position="1"> (ex) at 3 o'clock on April 5th 2003: instant [2003,4,5,3,0] An extended interval is a set of consecutive instants determined by a beginning instant and an ending instant.</Paragraph> <Paragraph position="2"> (ex) on April 5th, 2003: interval [[2003, 4,5,0,0], [2003,4,5,23,59]] A duration refers to a temporal distance between two distinct instants.</Paragraph> <Paragraph position="3"> (ex) for 5 years: duration [5,_,_,_,_] For the purpose of temporal description of a sentential event, we defined the following types and functional terms on the basis of typed lambda 2 See [Jo, 2000], [Lee, Ch., 1987], [Lee, H., 1993] and [Lee, J., 1982] for more detailed discussion. calculus3. The symbol l stands for abstraction and * stands for application4.</Paragraph> <Paragraph position="4"> Definitions of types i : type symbol denoting the type of individuals p : type symbol denoting the type of propositions e : type symbol denoting the type of events ent: type symbol denoting the type of natural numbers inst : type symbol denoting the type of instants inter : type symbol denoting the type of extended intervals dur: type symbol denoting the type of durations Type symbols may be omitted when no ambiguity is introduced.</Paragraph> <Paragraph position="5"> Definitions of functional terms (le. moment*e): e-inst Applying this function to any argument of type e, we obtain the moment of e of type inst.</Paragraph> <Paragraph position="6"> (le. interv*e): e-inter Applying this function to any argument of type e, we obtain the interval of e of type inter.</Paragraph> <Paragraph position="7"> (lx. beginning*x): e-inst (lx. ending*x): e-inst Applying these functions to any argument x of type e, we obtain the beginning/ending instant of x of type inst.</Paragraph> <Paragraph position="8"> (lx. duration*x): e-dur Applying this function to any argument x of type e, we obtain the duration of x of type dur. (lx. beg*x): inter-inst (lx. end*x):inter-inst Applying this function to any argument x of type inter, we obtain the beginning/ending instant of x of type inst. By definition, beg*[A,B] = A and</Paragraph> <Paragraph position="10"> 3 [Andrews, 1986 ; 2002], [Hindley et al., 1986] and [Renaud, 1996].</Paragraph> <Paragraph position="11"> 4 If M and N are lambda-terms, then M*N is a lambda null term.</Paragraph> <Paragraph position="12"> (lx. length*x): inter-dur Applying this function to any argument x of type inter, we obtain the length of x of type dur. By definition, length*[A,B]= |B-A| (lxly. x <<<t>> y): inst-inst-p It denotes that x of type inst is anterior to y of the same type. When no ambiguity is introduced, <<t>> will be omitted.</Paragraph> <Paragraph position="13"> (lxly. x =<<t>> y): inst-inst-p It denotes that x and y of type inst are simultaneous.</Paragraph> <Paragraph position="14"> (lxly. x [?]<<t>> y): inst-inst-p It denotes that lxly. (x <<<t>>y [?] x =<<t>>y). (lxly. x [?]<<t>> y): inst-inter-p It denotes that x of type inst is a member of y of type inter. By definition, lxly. (beg*y [?] x [?] end*y) (lxly. x [?]<<t >> y): inter-inter-p It denotes that x of type inter is included by y of the same type. By definition, lxly. (beg*y < beg*x [?] end*x < end*y).</Paragraph> <Paragraph position="15"> (lxly. x =<<t >> y): inter-inter-p x and y of type inter are simultaneous. By definition, lxly. (beg*x=beg*y [?] end*x=end*y). (lxly. a0 [?]<<t >> y): inter-inter-p It denotes that lxly. (beg*y [?] beg*x [?] end*x [?] end*y).</Paragraph> <Paragraph position="16"> The temporal adverbials with OE or DONGAN do not bring the same semantic constraints in all the sentences. It can be illustrated by the following examples of OE (7-10) and DONGAN (11-14). (7) ach'im ilgopshiOE nurongoiga chugotta. the morning /seven o'clock-OE / Nurongoi-NOM /die-PA-DEC At seven o'clock in the morning, Nurongoi died. The OE adverbial of this example indicates the moment when the event described by the nuclear</Paragraph> <Paragraph position="18"> But in (8) and (9), OE adverbials indicate an interval of which an instant is identified with the moment of the event.</Paragraph> <Paragraph position="19"> the last summer-OE/ Nurongoi-NOM/die-PA-DEC Last summer, Nurongoi died.</Paragraph> <Paragraph position="20"> (sr 9) [?]e[?]I die*e*nurongi [?] moment*e<pt_speech [?] interval*I [?] summer*I [?] moment*e[?]I Moreover, OE adverbials can introduce a period of recurrent events as in (10).</Paragraph> <Paragraph position="21"> (10) iljuirOE so dasot mariga chugotta a week-OE/cow/five/classifier-NOM/die-PA-DEC Five cows died every week.</Paragraph> <Paragraph position="22"> (sr 10) [?]I interval*I [?] length*I=(7 days) [?] [?]J interval*J [?] [?]P (equi-partition*I*P*J [?][?]K (P*K - |lx. cow*x [?] [?]e die*e*x [?] moment*e<pt_speech [?] interv*e[?]K|=5))7</Paragraph> <Paragraph position="24"> where [?]*R [?] lx. [?]R (R*P [?] P*x) As for DONGAN adverbials, they present the maximal duration of the described event as in (11). (11) hanshigan DONGAN kwanghoe bihaenggiga naratta an hour /DONGAN /of Kwangho /airplane-NOM /fly-PA-DEC Kwangho's airplane flew for an hour.</Paragraph> <Paragraph position="26"> In (12), the interval denoted by the DONGAN adverbial is included by that of the sentential event. In other words, it is not clear, for the moment, whether the described event reached its end or not.</Paragraph> <Paragraph position="27"> The following example (14) denotes that fishing of Yunsok has been repeated in a regular way during the interval indicated by the DONGAN adverbial. (sr 14) lD. [?]P equi-partition*D*P*intref [?] [?]H H=(lJ. (P*J [?] [?]e fish*e*yunsok [?] in*e*the_far_ocean [?] ending*e<pt_speech [?]</Paragraph> <Paragraph position="29"> Such a distributional pattern of events disappears when the nuclear sentence is modified by quantification, which is illustrated by (15).</Paragraph> <Paragraph position="30"> (15) shimnyon DONGAN yunsokun mon badaeso kokijabirul se bon haetta</Paragraph> </Section> <Section position="6" start_page="0" end_page="0" type="metho"> <SectionTitle> 10 years /DONGAN /Yunsok-TOP /far ocean-LOC </SectionTitle> <Paragraph position="0"> /fishing-ACC/three times/do-PA-DEC For ten years, Yunsok had fished in the far ocean three times.</Paragraph> <Paragraph position="1"> (sr 15) [?]I interval*I [?] length*I=(10 years) [?] |le. fish*e*yunsok [?] in*e*the_far_ocean [?] ending*e<pt_speech [?] interv*e[?]I|=3 To find a strategy to solve such a multiple ambiguity, we investigated three thousand sentences for each temporal marker 9 and discovered the following facts: 1. The semantic and syntactic properties of the phrase accompanying the temporal markers play an important role to locate the event in time.</Paragraph> <Paragraph position="2"> 2. It is necessary to distinguish monooccurrent sentences concerning a single event from multi-occurrent sentences concerning a set of different events10. The multi-occurrent nature is very often 8 [Renaud, 2002] defines the function used in this formula as follows: intref [?] interval of reference [[?]*I] [?] [inferior*([?]*I), superior*([?]*I)] where the brackets denote an interval.</Paragraph> <Paragraph position="4"> 9 We took the sentences from Yonsei malmunchi corpus built by Yonsei Center for Linguistic Information.</Paragraph> <Paragraph position="5"> 10 [Renaud, 2002].</Paragraph> <Paragraph position="6"> signaled by bare plurals in nominal phrases, adverb like ch'arero 'in turn', and quantification modifiers.</Paragraph> <Paragraph position="7"> 3. When it comes to the multi-occurrent sentences, DONGAN adverbials impose constraints on the distribution of events in some cases (see (14)).</Paragraph> <Paragraph position="8"> 4. The quantification negates the distributional meaning brought by DONGAN adverbials and gets them to indicate the temporal scope of this semantic operation (see (14) and (15)). 5. As for the verbal infix '-at-', its common semantic value is to denote the fact that the beginning of the event is anterior to the point of speech 11 ; le. beginning*e < pt_speech.</Paragraph> <Paragraph position="9"> 6. The information relevant to the time interpretation is scattered over the whole sentence; in the verbal phrase, quantification modifiers ranging over individuals or events, determiners in the nominal phrases and time adverbials.</Paragraph> <Paragraph position="10"> Therefore, the temporal interpretation of a sentence should be constructed in a compositional way.</Paragraph> <Paragraph position="11"> 7. For the same reason, the aspectual value should be attributed to the nuclear sentence and not to the verbal phrase.</Paragraph> </Section> <Section position="7" start_page="0" end_page="0" type="metho"> <SectionTitle> 3 Computational implementation </SectionTitle> <Paragraph position="0"> We discovered that Renaud's G Grammar is suitable for the purpose of computational implementation of these facts. This grammar loads information on word definitions as little as possible and charges the rules with detailed description.</Paragraph> <Paragraph position="1"> This principle contributes to gathering the pieces of information scattered throughout a sentence and to establishing a semantic representation of the sentence in a compositional way. Moreover, it enables us to deal with all the other linguistic phenomena in the same way as with the temporal problems. This grammar has been applied to French [Renaud, 1996; 2000; 2002] and Japanese [Blin, 1997] as well.</Paragraph> <Paragraph position="2"> 11 [Reichenbach, 1966].</Paragraph> <Paragraph position="3"> This grammar is divided into word definitions called 'dico' and composition rules. Each of them consists of syntactic constraint, unification-based feature constraint and semantic constraint written in lambda-terms.</Paragraph> <Paragraph position="5"> In composition rules, the symbol '-' stands for syntactic rewriting and '<<=' stands for breduction. null We present here an example process establishing the temporal interpretation of a Korean sentence extended by a DONGAN adverbial.</Paragraph> <Paragraph position="6"> (11) hanshigan DONGAN kwanghoe bihaenggiga naratta an hour/ DONGAN/ of Kwangho/ airplane-NOM/ fly-PA-DEC Kwangho's airplane flew for an hour. We determine the semantic term of a DONGAN adverbial, according to the semantic and syntactic properties of the phrase preceding the temporal marker.</Paragraph> <Paragraph position="7"> The time adverbial hanshigan DONGAN in (11) includes a NP denoting duration and conforms to the syntactic condition E 12 . Thus, the semantic term of type t5 is assigned to this time adverbial. We also calculate the semantic term of the nuclear sentence relying on criteria such as quantification modification, mono/multi-occurrent and aspect 13 , which get involved in the feature constraint at the levels of both dico and of rules. Since the nuclear sentence of (11) is not modified by quantification, and since it concerns a single event of activity, it receives a semantic term of type c3 in the following figure.</Paragraph> <Paragraph position="8"> We excluded the last structure from our research because of its highly context dependent meaning. 13 Aspectual classification is done by the following method; first, we observed the compatibilities of nuclear sentences with linguistic expressions such as -go innun chungida, mane and dongan. And then we investigated whether mane indicates the preparatory stage of the concerned event and whether dongan marks the resultant state of the event. As a result, we obtained seven distinct combinations as follows.</Paragraph> <Paragraph position="9"> At last, the semantic term of the time adverbial and that of the nuclear sentence are joined together by the following rule to put the final semantic representation of (11)14: ph1- adv_DG, ph As we mentioned above, one of the most important advantages of G Grammar consists of its capacity to establish semantic interpretations in a compositional way. Even if we presented only the final step of semantic processing, our Korean parser constructs a semantic representation at each step15.</Paragraph> <Paragraph position="10"> 14 '/' stands for disjunction.</Paragraph> <Paragraph position="11"> 'lxly. proj*x*y' returns the member occurring in the xth place in the list y.</Paragraph> <Paragraph position="12"> 15 Our Korean parser is built in LPI Prolog. In Figure 6, 'lb' stands for l-abstraction and '*' stands for lapplication. null</Paragraph> <Paragraph position="14"> The sentences extended by an OE adverbial are represented in the same way as those by a DONGAN adverbial, as will be seen in the following.</Paragraph> <Paragraph position="15"> On the fifteenth of March, Nurongoi died. Relying on the semantic and syntactic constraints of the phrase preceding OE at the same time, we determine the semantic term of the OE adverbial. Since samwol shiboirOE of (8) denotes an extended interval and it conforms to the syntactic condition C, this adverbial is attributed the semantic term of type t3.</Paragraph> <Paragraph position="16"> The semantic representation of the nuclear sentence of (8) is established in the same way as explained above in Figure 3. At last, taking the semantic terms of the OE adverbial and of the nuclear sentence, the following rule serves to construct the final representation of the whole</Paragraph> </Section> class="xml-element"></Paper>