<?xml version="1.0" standalone="yes"?>
<Paper uid="E85-1039">
  <Title>TOWARDS AN AUTOMATIC IDENTI~ICATION 0F TOPIC AND FOCUS</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
TOWARDS AN AUTOMATIC IDENTI~ICATION 0F TOPIC AND FOCUS
</SectionTitle>
    <Paragraph position="0"/>
  </Section>
  <Section position="2" start_page="0" end_page="266" type="abstr">
    <SectionTitle>
ABSTRACT
</SectionTitle>
    <Paragraph position="0"> The purpose of the paper is (i) to substantiate the claim that the output of an automatic analysis should represent among other things also the hierarchy of toplc-focus articulation, and (ii) to present a general procedure for determining the toplc-focus articulation in Czech  and English.</Paragraph>
    <Paragraph position="1"> (i) The following requirements on the output of an automatic analysis are significant: (a) in the output of the analysis it should be marked which elements of the analyzed sentence belong to its topic and which to the focus$ (b) the scale of communicative dynamism (CD) should also be identified for every  representation of a meaning of the -n-lyzed sentence, since the degrees of CD correspond to the unmarked distribution of quantifier scopes in the semantic interpretation of the sentences (c) the analysis should also distinguish toplcless sentences from those hav~ng a topic, which is relevant for the scope of negation.</Paragraph>
    <Paragraph position="2"> (ii) For an automatic recognition of topic, focus and the degrees of CD, two ~ oints are crucial: a) either the input language has (a considerable degree of) the so-called free word order (as in Czech, Russian), or its word order is determined mainly by the grammatical relations (as in English, Prench); (b) either the input is spoken discourse (and the recognition procedure includes an acoustic analysis), or written (printed) texts are analyzed.</Paragraph>
    <Paragraph position="3"> In accordance with these points, a general procedure for determining topic, focus and the degrees of CD is formulated for Czech and English, with some hints how the preceding context can be taken into account.</Paragraph>
    <Paragraph position="4"> le We distinguish between the l@vel of l~uistic,meaning (de Saussure s and HJe~mslev s &amp;quot;form of content&amp;quot;, Cosieru s &amp;quot;Bedeutung&amp;quot;, others &amp;quot;literal meaning&amp;quot;) and its interpretation in the sense of truth-conditional, intensional logic (see Materna and Sgall, 1980~ Sgall, 1983).</Paragraph>
    <Paragraph position="5"> Por some purposes of automatic treatment of natural l~n~uage (including machine translation) it is sufficient if the output of the procedure of analysis is more or less identical with the representation of the (linguistic) meaning of the sentence. For other purposes, such as that of full natural language comprehension, it is necessary to go as far as the semantic (truth-conditional) i~terpretation, using a notation that includes variables, operators, parentheses and similar means. The topic-focus articulation (TFA) is understood as one of the hierarchies of the level of meaning, whose other two hierarchies are that of dependency syntax (close to case grammar) and that of coordination (and apposition) relations. The basic task of a description of TPA is to handle the differences between such sentences as John gave Mary a BOOK. John ~ave a book to ~t%RY. It was ~RY John nave a book. It was a&amp;quot;~P~-~'~--John gave to Mary. It was JOHN who gave Mary a book. It was JOHN who gave a book to ~-~V- (Capitals denotethe position of the intonation center.) In Fig. I we present a simplified representation of one of the meanings shared by the sentences in which the intonation center is placed on BOOK; Fig. 2 corresponds to one of the meanings shared by the sentences with the intonation center on ~LARY.</Paragraph>
    <Paragraph position="6"> The following requirements on the output of an automatic analysis are significant, whatever approach to the description of the structure of the sentence has been chosen: a) In the output of the analysis it should be marked which elements of the analyzed sentence belong to its topic  (= one of the books) to MARY. It was MARY (to Whom) John zave a book. and which to its focus, since this is relevant as for which questions can be answered by the sentences thus e.g. (I) can answer (2) and (3), while (4) can be answered by (5) rather than by (I).</Paragraph>
    <Paragraph position="7">  (I) John talked to few girls about many PROBLEMS.</Paragraph>
    <Paragraph position="8"> (2) How did John behave? (What about John?) (3) About what did John talk to whom? (4) To whom did John talk about many problems? (5) John talked about many problems  to few GIRLS.</Paragraph>
    <Paragraph position="9"> Note that in (I) the verb as well as the Addressee belong to the focus in some of the readings, while in others they are included in the topic; in (5) the Addressee belongs to the focus and Objective to the topic in all readiags, only the position of the verb here being ambiguous (this can be checked by tests including negation and by the question test, see Sgall etal., 1973).</Paragraph>
    <Paragraph position="10"> b) The scale of communicative dynamism or CD (ibid.) should also be identified for every representation of a meaning (underlying structure, etc.) of the anRlyzed sentence, since the degrees of CD correspond to the unmarked distribution of quantifier scopes in the semantic interpretation of the sentence; thus in (I) and (6)the Addressee includes a quantifier with a wider scope than that of the Objective (on the primary reading), while in (5) and (7) the quantifier of the Ob-Jective includes the Addressee in its  scopes (6) It is JOHN, who talked to few girls about many problems.</Paragraph>
    <Paragraph position="11"> (7) It is JOHN, who talked about many problems to few girls.</Paragraph>
    <Paragraph position="12"> c) The analysis should also distingu null ish topicless sentences (corresponding, ~n the prototypical %ases, to Kuno s neutral description or to the thetic Judgements of classical logic) from those having a topic; this difference is relevant for the scope of negation: only (8)(b) is semantically equivalent to It is not true that fo~ is falling, wher~s in (9)(b) the subject, included in the topic, is outside the scone of negation; a more appropriate paraphrase is= Shout Pather it is not true that he xs coming.  (8)(a) POG is falling.</Paragraph>
    <Paragraph position="13"> (b) No FOG is falling.</Paragraph>
    <Paragraph position="14"> (9)(a) Father is COMING.</Paragraph>
    <Paragraph position="15"> (b) Pather is not COMING.</Paragraph>
    <Paragraph position="16"> 2. It has been found (HajiEov~ and  Sgall, 1980 and the writings quoted there) that the scale of CD coincides to a great part with Chomsky (1971) calls &amp;quot;the range of permissible focus&amp;quot;. With the elements belonging to the focus the scale of CD is determined by the kinds of complementation (deep cases~ the order always being in accordance with what we call systemic ordering; for the main participan%s of the verb in  English this ordering is Time - Actor - null - Addressee - Objective - Origin - Effect - Manner - Instrument - Locative (see Seidlov~, 1983). The scale of CD differs from this ordering only if at least one of the elements in question belongs to the topic (this is true about the Addressee in (10)(b),about the Origin in (11)(b), and about the Effect in 412) (b) below): (10)(a) I gave several children a few APPLES.</Paragraph>
    <Paragraph position="17"> (b) I gave a few apples to several CHILDREN.</Paragraph>
    <Paragraph position="18"> (11)(a) John made a canoe out of a LOG.</Paragraph>
    <Paragraph position="19"> (b) John made a CANOE out of a log.</Paragraph>
    <Paragraph position="20"> (12)(a) John made a log into a CANOE. (b) It was a LOG John made into  a canoe.</Paragraph>
    <Paragraph position="21"> Thus, in the (b) sentences a few apples, ~ and a canoe are contextually bound, standing close to a few of those one of the lo~s, the canoe we bout, respectively. In the (a) examples the rightmost complementations belong to the focus (they carry the intonation center), while the complementations standing between them and the verb are ambiguous in this respect: in some meanings of the sentence they are contextually bound and belong to the focus; a similar ambiguity concerns also the verbs in all the examples.</Paragraph>
    <Paragraph position="22"> Systemic ordering is language specific; Czech differs from English e.g. in that it has Time after Actor and Object after Instrument; for more details, see Sgall, Haji~ov~ and Panevov~ (in press, Chapter 3).</Paragraph>
    <Section position="1" start_page="264" end_page="264" type="sub_section">
      <SectionTitle>
3.1 Pot an automatic recognition of
</SectionTitle>
      <Paragraph position="0"> topic, focus and the degrees of CD, two points are crucial: (A) Either the input is spoken discourse (and the recognition procedure incl~des an acoustic analysis), or written (printed) texts are analyzed.</Paragraph>
      <Paragraph position="1"> (B) Either the input language has (a considerable degree of the so-called free word order (as in Czech, Russian), or its word order is determined ma~n!y by the grammatical relations (as in English, Prench).</Paragraph>
      <Paragraph position="2"> Since written texts do not indicate the position of the intonation center and since the &amp;quot;free&amp;quot; word order is determined first of all by the scale of CD, it is ,! ,, evident that the either cases in (A) and (B) do not present so many difficulties for the recognition procedure as the &amp;quot;or&amp;quot; cases do.</Paragraph>
      <Paragraph position="3"> A written &amp;quot;sentence&amp;quot; corresponds, in general, to several spoken sentences which differ in the placement of their intonation center. Thus, I/ an adverbial of time or of place stands at the end of the sentence, as in (13), then at least two sentences may be present, see (14)(a) and (b), where the intonation center is marked by the capitals;the TFA clearly differs:  (13) We were swimming in the pool in the afternoon.</Paragraph>
      <Paragraph position="4"> (14)(a) We were swimming in the pool in the AFTERNOON.</Paragraph>
      <Paragraph position="5"> (b) We were swimming in the POOL  in the afternoon.</Paragraph>
      <Paragraph position="6"> In languages with the so-called free word order this fact does not bring about serious complications with technical texts, since there is a strong tendency to arrA-ge the words so that the i~tonation center falls on the last word of the sentence (if this is not enclitical).</Paragraph>
    </Section>
    <Section position="2" start_page="264" end_page="266" type="sub_section">
      <SectionTitle>
3.2 A procedure for the identification
</SectionTitle>
      <Paragraph position="0"> of topic and focus in Czech written (first of all technical) texts can then I be formulated as follows: (i)(a) If the verb is the last word of the surface shape of the sentence (SS), it always belongs to the focus.</Paragraph>
      <Paragraph position="1"> (b) If the verb is not the last  word of the SS, it belongs either to the topic, or to the fOCUS.</Paragraph>
      <Paragraph position="2"> Note: The ambiguity accounted for by the rule (i)(b) can be partially resolved (asp. f?r the purposes of practical systems) on the basis of the features of the preceding sentence: if the verb Y= of the analyzed sentence is identical with the verb of the preceding sentence, or if a relation of synonymy or inclusion holds between the two verbs, then V belongs to the topic. Also, a semantically weak, general verb, such as to be, to become, to carry out, can-Y~derstood as belonging to the topic. In other cases the primary position of the verb is in the focus.</Paragraph>
      <Paragraph position="3"> (ii) The complementations preceding the verb are included in the topic.</Paragraph>
      <Paragraph position="4"> (iii) As for the complementations following the verb, the boundary IThe term oomplementation (sentence part) is used in the sense of a subtree occupying the position of a deep case or an adverbial. - We disregard the so-called subjective order here, since Gin contrast to English) in a Czech written technical text such sentences as The SWITCH was off are extremely rare.  between topic (to the left) and focus (to the right) may be drawn between any two complemenrations, provided that those belonging to the focus are arranged in the surface word order in accordance with the systemic order~.</Paragraph>
      <Paragraph position="5"> (iv) If the sentence contains a rhematizer (such as even, also, only), then in tHeprimary ~-~ the ~omplementation following the rhematizer belongs to the focus and the rest of the sentence belongs to the topic. 2 3.22 Similar regularities hold for the analysis of spoken sentences with normal intonation. However, if a non-flnal complementation carries the intonation center (IC), then (a) the bearer of the IC belongs to the focus and all the complementations standing after IC belong to the topic;  (b) rules (ii) and (iii) apply for the elements stand~-g before the bearer of the intonation center~ (c) the rule (i)(b) is applied to the verb (if it does not carry the IC). 3 3.31 As for the identification of topic and focus in an English written  sentence, the situation is more complicated due to the fact that the surface word order is to a great extent determined by rules of grammar, so that intonation plays a more substantial role and the written form of the sentence displays much richer ambiguity. Por English texts from polytecbn~cal and scientific domains the rules stated for Czech in 3.21 should be modified in the following ways: (i)(a) holds the surface subject of the sentence is a definite NP; if the subject noun has | 2This concerns such sentences as Here even a device of the first type can e~ used; in a secondary case the rhematizer may occur in the topic, e.g. if the sentence part in its scope is repeated from the preceding co-text.</Paragraph>
      <Paragraph position="6"> 3However, an analysis of spoken discourse should pay due respect not only to the preceding co-text, but also to the situation of the utterance, i.e. to its context in the broader sense.</Paragraph>
      <Paragraph position="7"> an indefinite article, then the subject mostly belongs te the focus, and the verb to the r topic; however, marginal cases with both subject and verb in the focus, or with subject (though indefinite) in the topic and the verb in the focus are not excluded} (i)(b) holds, including the rules of thumb contained in the note)  (ii) holds, only the surface sub~ect and a temporal adverbial can belong to the focus, if they do not.have the form of definite NP s~ (iii) holds, with the following modifications: (a) If the rlghtmost complementation is a local or temporal adverbial, then  it should be checked whether its lexical meaning is specific (its bein~ a proper name, a narrower term,or a term not a belongin~ to the sub~ect domain of the given text)or general (a pronoun, a broader term); in the former case it is probable that the adverbial bears IC and belongs to the focus, as in (15) and (16) while in the latter case it rather belongs to the topic, as in (17) or (18), where the word method probably carries  the IC~ (15) Several teams carried out experiments with this method during a single week.</Paragraph>
      <Paragraph position="8"> (16) Several teams carried out experiments with this method In Berkeley and Princeton.</Paragraph>
      <Paragraph position="9"> (17) Several teams ... method during the last decades.</Paragraph>
      <Paragraph position="10"> (18) Several teams ... method in this country.</Paragraph>
      <Paragraph position="11"> (b) If the verb is followed by more than one complementation and if the  sentence final position is occupied by a definite NP or a pronoun, this right-most complementation probably is not the bearer of IC and it thus belongs to the topic.</Paragraph>
      <Paragraph position="12">  (c) If (a) or (b) apply, then it is  also checked which pair of complementations disagreeing in their word order with their places under systemic ordering is closest (from the left) to IC i.e. to the end of the focus)i the boundary between the (left-hand part of the) topic and the focus can then be drawn between any two complementations beginning with the given pair.</Paragraph>
      <Paragraph position="13"> 3.32 If a spoken sentence of English is analyzed, the position of IC can be determined more safely, so that it is easier to identify the end of the focus than with written sentences and the  modifications to rule (lii) are no longer necessary. The procedure can be based on the regularities stated in 3.22.</Paragraph>
      <Paragraph position="14"> 4. Whenever the preceding context can be taken into account in the analysis, the salient (activated) items (see HaJi~ov~ and Vrbov~, 1982) should be registered. A &amp;quot;pushdown&amp;quot; principle can be used: the item mentioned as the (last part of the) focus of the last utterance is the most salient in the given time-point of the discourse, while the elements that were mentioned in other positions of this utterance get a lower status in the activated part of the stock of shared knowledge, and those that have not been mentioned in one or several subsequent utterances may fade away (if they do not have a specific position of a &amp;quot;hypertopic&amp;quot;, which concerns e.g. those mentioned in a heading). Thus it can be decided in some of the unclear cases (e.g. with temporal adverbials, see point (iii) in 3.31) whether a complementation belongs to the topic or to the focus. This method has its limits: the set of activated items should include not only items mentioned in the text, but also their parts, counterparts and other items connected with them by associative relations; on the other side, if a specific kind of contrast is involved, it is possible that also an item included in this set is mentioned as a part of the focus of the next utterance.</Paragraph>
      <Paragraph position="15"> 5. A procedure of automatic identification of topic and focus has been incorporated, to a certain degree, in the parser for Czech implemented in Colmerauer s Q-language on EC-I040 (compatible with IBM 360) and briefly described by Panevov~ and Sgall (1980), Panevov~ and Oliva (1982). The prototypical cases of topic-focus articulation are also covered by the English parser designed and implemented (using the same programming tools and hardware) by Kirschner (1982) as a part of the Engllsh-to-Czech machine translation project.</Paragraph>
      <Paragraph position="16"> Both parsers account also for more complex sentences than the examples quoted in this summary~ in some such cases there are embedded sentence parts that have (partial) topics and loci of their owD~ Both in translation and comprehension the topic-focus articulation should be paid due respect to, since - as we have seen in Sect. I - it is relevant not only for an appropriate use of a sentence in different contexts, but also for truth-conditional interpretation, especially of sentences with certain kinds of quantification and with negation.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>