Page 216 - Encyclopedia of Philosophy of Language

Truth and Meaning
According to the 'about'-context test, a structurally unmarked sentence S<NP|> is paraphrased by extending the sentence in agreement with a scheme like He said aboutIof NP, that S<NPi>. Properly speaking, the extended sentence is not an adequate paraphrase of the original sentence. The added part forces the orig- inal sentence to be about what is represented by NP, or, perhaps more precisely, what the person in ques- tion says about it. Without this addition, the original sentence may be about a different topic, depending on the preceding context.
In the case of the 'about'-question test the topic- bearing sentence S<NP|> is not paraphrased but is placed in a context of a specific question What about NP, ?The test is based on the assumption that if sen- tence S<NP(> is about NP,, S<NP|> is an acceptable answer to the 'about'-question. According to this test accented and clefted NPs have no topic function since the sentences which contain them constitute no acceptable answer to the 'about'-question. Like all other tests, this test does not make manifest how sen- tence topics affect discourse coherence. The coherence in discourse does not become apparent either when every topic-bearing sentence is preceded by an 'about'- question, or when every topic-bearing sentence is replaced by one of the proposed paraphrases.
Of all identification tests that have been proposed the question test is probably the best known. This test has many variants, the most comprehensive of which is presented in Sgall, et al. (1986). According to this test the division of a topic-bearing sentence in a topic and a comment (focus) part is determined by the set of w/i-questions to which the sentence is an appropriate direct answer, both informationally and inton- ationally. Constituents of the sentence which appear in every question belong to the topic part and con- stituents which appear in no question belong to the comment part. The test fails to specify the status of the constituents that appear in only some of the questions.
When the question test is applied to, for example, the sentence John hit BILL, it determines that the constituents John and Bill belong to the topic and comment part respectively. Since the sentence con- stitutes an appropriate direct answer to both What did John do? and Who did John hitl the test leaves the status of the verb undecided. Although this sentence is also an appropriate answer to questions like What happened"! and What's newl, these questions are excluded from the set. It is assumed that sentences in discourse which answer such general questions are topicless (1986:212). A characteristic of this test is that different questions can determine the same topic-comment modulation, despite the fact that they may arise in different appropriate contexts. In addition to this test, an algorithm for topic-focus identification has been developed (HajiCova et al 1995).
3.2 Classification of Operational Characterizations
Context-independent operational characterizations of sentence topics can be divided into two types: either in terms of just a specific syntactic category (Chafe 1976) or in terms of word order (e.g., Chomsky 1965; Halliday 1967), with or without the requirement of a specific category. According to the former, the topic of a sentence is identified with the grammatical subject of the sentence. According to the latter, the topic (theme) of a sentence is, in principle, identified with the element in sentence-initial position. In Halliday (1967), which comes into the word order category, this characterization is meant to apply without any restriction as to the syntactic category of the sentence- initial element. Chomsky (1965) on the other hand (also of the word order class) explicitly states that sentence topics must have NP-status. The author characterizes a sentence topic as the leftmost NP immediately dominated by S in the surface structure.
Characterizations in terms of word order prevail. Three specific consequences of this type can be men- tioned. The first consequence (which is not generally accepted) is that a topic is defined for every sentence, and that, moreover, this is always linguistically ex- pressed. The second consequence is that the same topic is defined for succeeding utterances which have the same constituent in first position. Creider (1979) demonstrates that this is not always correct. He shows that especially a left-dislocated constituent cannot also serve as the topic of a succeeding utterance. The third consequence has to do with topic-comment phenomenathatarerelatedtoquestion-answer pairs. Characterizations in terms of word order have as a consequence that the results of topic identification are inconsistent with the widely accepted assumption that one single topic is defined for question-answer pairs. This assumption implies that in question-answer pairs the topic is constituted by the question, so that, in the answer, the topic constituent always represents given information. According to word-order type charac- terizations, the topic constituent in the answer can also have new status. This is typically the case when the sentence element in sentence initial position receives the primary accent, for example, the con- stituent Harry in 'Who has been arrested?—Harry has been arrested.'
As has been said, some characterizations of sen- tence topics are context-dependent. These can be div- ided into three types: characterizations in terms of informational status (Bolinger 1977; Hornby 1970; Sgall, et al. 1973 and many others), sometimes involv- ing the extra requirement of a status of contextually determined preference as is characteristic for the com- putational approach of Centering Theory (e.g., Grosz, et al. 1986, 1995; Joshi and Weinstein 1981; Walker, et al. 1997); characterizations in terms of alternatives as is suggested by the formal semantic approach known as Alternative Semantics (in particular Rooth
194