Yoko Hasegawa
University of California, Berkeley
Journal of Pragmatics, vol. 25, 763-90, 1996.
Japanese TE, like English and, can convey a diverse range of semantic relations -- e.g. temporal sequence, cause-effect, means-end, contrastive, concession, conditional. Whenever such a relation is understood, however, it is always inferable solely from the conjuncts themselves. Moreover, these relations are cancellable and thus can be regarded as conversational implicatures. Most researchers, therefore, have considered TE-linkage to be primarily a syntactic device that in itself conveys little semantic information: the semantic relations associated with TE-linked sentences are worked out from the meanings of the conjuncts alone (Alfonso 1966, Morita 1980, Teramura 1981, Endo 1982, Himeno 1984, Ogoshi 1988, inter alia). Let us call this approach the implicature-only reductionist analysis.2
In this article I will argue that the implicature-only reductionist analysis is untenable for the following reason. Although all semantic relations associated with TE-linkage can be inferred from the conjuncts alone, the contrary does not hold: not all semantic relations that can be implicated by two paratactic clauses are possible when the clauses are linked by TE. For example, if the clauses equivalent to I sat down and Joan came into the room are presented paratactically in Japanese, the interpreter naturally reads in a temporal sequence relation, just as in English. This merely temporal relation, however, is not an available reading when the clauses are linked by TE. That is, among the relations potentially implicated by two co-present clauses, some are filtered out by TE-linkage. Therefore, TE-linkage cannot be a purely syntactic device; it must have some meaning that excludes the reading temporal sequence from the set of possible interpretations in the case of this example.
The fact that some 'implicated' meanings must be regarded as properties of TE-linkage, and thus should properly be described in its semantics, challenges the conventional dichotomy of semantics and pragmatics. I will argue that TE-linkage indeed has its own inherent meanings, and demonstrate that these meanings cannot be stated in terms of traditional semantic relations but can only be understood in cognitive terms.
The organization of the article is as follows. Section 2 lays out the morphological characteristics of TE and the conventional taxonomy of TE-linkage. Section 3 explains the notion of meaning which will be utilized in this study. Section 4 examines TE-linkage and the temporal sequence relation in detail, and demonstrates that TE-linkage is incompatible with an incidental temporal sequence (i.e. pure temporal sequence proper). Section 5 is devoted to cases where the second conjunct refers to a human action. Section 6 discusses additive and contrastive relations when expressed with TE-linkage. Section 7 addresses certain theoretical implications of the findings. The conclusion follows in Section 8.
As with the past-tense/perfective suffix -ta, TE participates in a number of assimilatory morphophonemic processes that respond to the final consonant of a consonant-final verb stem: when the verb stem ends in a voiced obstruent, TE is voiced, e.g. nug- 'take off' + TE > nui-de, as in (1a) below. The copula + TE is realized as de. These morphophonemic details are relevant in this study only insofar as they may help the reader to recognize the presence of TE in any given example.
Although TE-linked sentences are frequently translated into English with a present participle, as shown in (1), TE-linkage is significantly different both from free adjuncts, e.g. Inflating her lungs, Mary screamed, and from absolutes, e.g. The coach being crowded, Fred had to stand (both from Kortmann 1991:5), in that TE-linkage is iterable, as shown in (1c).
(1)a. jon wa [uwagi o nuide], [hangaa ni kaketa].
'John, taking off his jacket, hung it on a hanger.'
(Kuno 1973: 200; transcription modified)
b.jon wa [terebi o mite] [benkyooshita].
(Harasawa 1994:182; slightly modified)
'John studied, watching TV.'
'Having watched TV, John studied.'
c. [hi ni kazashite] [mizu o joohatsu-sasete] [futatabi omosa o hakaru].
'By holding (it) over the flame, evaporate the water and weigh (it) again.'
This syntactic property of iterability is another reason why TE should be treated as a connective suffix rather than as an element forming a gerund or participle. Semantically and pragmatically, however, TE-linkage exhibits many similarities with free adjuncts and absolutes in English.
2.2. Conventional categorization of TE-linkage. Traditionally, TE-linkage has been divided into three categories according to the function of TE: (i) as a nonproductive derivational suffix, as in (2a); (ii) as a linker connecting a main verb with a so-called auxiliary to form a complex predicate, as in (2b); and (iii) as a linker connecting two phrases or clauses, as in (2c).
(2)a. myoonichi aratamete [another occasion (=renovate-TE)] ukagaimasu.
'I'll visit (you) again tomorrow.'
b. otooto wa ima hon o yonde iru.
'(My) brother is reading a book now.'
c. [mina kawaki to nemuke ni taete] [same no oyogu ara-umi o hyooryuu-shita].
'Enduring thirst and sleeplessness, they drifted on the rough seas where sharks (sometimes) swam.'
In the first category, TE functions as a derivational suffix, forming an adverb from a verb. Morphologically, aratamete 'on another occasion' in (2a) could be analyzed as the verb aratame- 'renovate' + TE. However, aratamete in this usage does not have any valence of its own, i.e., it lacks a subject and object. In general, in this derivational category verbs lose part of their verbal nature when TE is attached. Furthermore, the meaning of the derived adverbial is not always predictable from the meaning of the base verb, and only certain verbs can form such an adverbial.5 Forms like aratamete must therefore be listed as such in the lexicon. (If aratamete were to take overt or covert arguments, on the other hand, it would belong to the second or third category.) Because the derivational process associated with TE in this function is non-productive and its semantic import is irregular, and in particular because TE does not function here as a true connective, this first category will not be considered further in the present study.
In the second category, exemplified by (2b), the verb preceding TE is semantically the main predicate of the clause, and the verb or adjective that follows TE is a so-called auxiliary.6 For example, 'verb-TE i-' in (2b) is the grammatical means for expressing the imperfective or perfect aspect, with the choice normally depending on the Aktionsart of the first verb. The semantic relations between the linked constituents in this second category are relatively fixed compared with the third category, and are in large part determined by the second constituent. Although TE does function here as a connective suffix, this category too will be excluded from the present investigation.
The semantic relations between the linked constituents in the third category, on the other hand, are so diverse that no single relation can be considered central. In (2c), the first clause holds a circumstance relation to the second; however, as shown in (3-9), many other relations can also be expressed by TE-linked constituents, e.g. additive, temporal sequence, cause-effect, means-end, contrastive, concession, and conditional.7 (Note that temporal sequence is included here only provisionally; later in this article it will be shown that the pure temporal sequence relation proper is in fact incompatible with TE-linkage.)
(3) Additive
kono uchuu no soodai-na sungeki wa [setsunakute] [shimpiteki da].
'This grand drama of the universe [i.e. an eclipse] is touching and mysterious.'
(4) Temporal sequence
[furasuko ni kitai o irete] [futatabi omosa o hakaru].
'Put the gas into the flask and weigh (it) again.'
(5) Cause-effect
[tomodachi o ijimete] [sensee ni shikarareta]. (Endo 1982)
'(I) was scolded by the teacher because (I) bothered (my) friend.'
(6) Means-end
[hi ni kazashite] [mizu o joohatsu-saseru].
'By holding (it) over the flame, evaporate the water.'
(7) Contrastive
[maki wa gookaku shite] [hiro wa fugookaku datta].
'Maki passed (the exam), but Hiro was disqualified.'
(8) Concession
kare wa [sono koto o shitte-ite] [iwanai].
'Although he knows the subject matter, he won't say it.' (Morita 1980: 318)
(9) Conditional
[zenbu tabete] 20-doru desu.
'If (you) eat everything, (it) is $20.'
It is largely because of this diversity of semantic relations that many researchers advocate what I call the implicature-only reductionist analysis of TE-linkage.
As shown in the above examples, TE-linkage exhibits an extreme degree of semantic unspecificness, and probably for this very reason is particularly common in actual usage8 -- without causing problems in communication. This leads to questions about how much of the meaning is attributable to the TE-linkage itself, how much to the properties of the conjuncts, and how much to the interpreter's extralinguistic knowledge of the described situations. Before proceeding, let me clarify the notion of meaning to be used in this study.
3.1. Independent and dependent semantic aspects. Following the methodology of Reichling, Dik (1968:257-58) divides linguistic information into semantic information and grammatical (i.e. syntactic/morphological) information. All expressions have grammatical information associated with them by virtue of being usable in larger syntagms.
Semantic information is further divided into independent and dependent semantic aspects. The independent semantic aspects are immediately obtainable from the expression without further linguistic context. By contrast, the dependent semantic aspects of the expression can be obtained only within a larger whole of which the expression is a part. For example, speakers of English know the semantics of table with no further context, whereas they do need some context, e.g. table-, to identify the semantics of the plural suffix -s; plurality, as a relational notion, cannot be defined without essential reference to some noun. Thus table is said to have an independent semantic aspect of its own, whereas -s has only a dependent one.
Henceforth I will use the expression meaning of the connective X to refer to X's dependent semantic aspects. Connectives have grammatical information associated with them; they also indicate certain relationships between the semantic information conveyed by the conjuncts. Crucially, however, connectives do not carry independent semantic aspects of their own. Even with a semantically loaded connective, such as before, it is necessary to mention the clauses which before links in order to describe the semantic information it conveys -- namely, that the occurrence of the situation described by the clause to which before is attached must temporally follow the occurrence of the situation described by the other clause.
Viewed in this light, the implicature-only reductionist analysis is justified only if meaning is restricted to independent semantic aspects, since indeed no semantic description of te is possible without recourse to the larger constituent of which TE is a part. But advocates of this analysis appear to contend that TE lacks even dependent semantic aspects: they contend that the contingent semantic relations associated with TE-linkage are so diverse that the interpreter only infers the specific sense intended by the speaker. In order to discuss this issue further, it is important to clarify the distinction between what is asserted (including what Grice calls conventional implicatures) and what is conversationally implicated.
3.2. Implicature. One of the basic requirements for understanding discourse is recognizing how each clause coheres with its predecessor. Our linguistic and pragmatic competence enables us to read in conceivable relation(s) even when two clauses are simply juxtaposed in parataxis (recall They had a baby and got married, in Section 1). Thus certain aspects of interpretation, e.g. temporal sequence with and-linkage, are not part of the conventional force of the uttered sentence, but rather what Grice (1975) has named conversational implicature.
In the Gricean theory of linguistic pragmatics, the cause relation observed between conjuncts linked by because and the precedence relation between conjuncts linked by before are considered conventional (not conversational) implicature. Conventional implicature involves the non-truth conditional lexical meaning of some element and is attached to a particular expression by convention, not by pragmatic principles. For example, the conjunctions and and but are truth-conditionally equivalent: the 'additional' meaning of contrast that but conveys is imparted by conventional implicature (Grice 1961). As Levinson (1983:128) points out, however, 'conventional implicature is not a very interesting concept -- it is rather an admission of the failure of truth-conditional semantics to capture all the conventional content or meaning of natural language words and expressions.' In this study, conventional implicatures will be considered as falling under the heading of asserted meaning. Implicature will thus be restricted to conversational implicature.
The difference in meaning between and-linkage (implicated) and because- or before-linkage (asserted) emerges sharply in the following pairs.
(10)a. One plus one is two, and I'm sad.
b. Because one plus one is two, I'm sad.
(11)a. John eats apples, and six men can fit in the back seat of a Ford.
b. John eats apples before six men can fit in the back seat of a Ford.
If the b-sentences were uttered, the interpreter would at least try to make sense out of them in such a way that a cause (10b), or a precedence (11b), holds between the conjuncts; the connectives because and before force these interpretations. Success or failure in interpreting anomalous sentences like (10b, 11b) will depend on one's deductive abilities.9 One might interpret (11b), for example, as describing John dieting so that he will be thinner and take up less space.
With the a-sentences, on the other hand, the word and does not demand some particular interpretation. Indeed, the most likely interpretation of and here is simply as a signal that the speaker has something more to say, i.e., intends to keep the floor. Halliday and Hasan (1976:233), who draw a strict line between structural and cohesive (semantic) relationships, note that 'the "and" relation is felt to be structural and not cohesive, at least by mature speakers; this is why we feel a little uncomfortable at finding a sentence in written English beginning with And, and why we tend not to consider that a child's composition having and as its dominant sentence linker can really be said to form a cohesive whole.' They contend that and has a syntactic function, but that it provides little information about the semantic relation between the conjuncts. In this respect, TE-linkage is quite similar to and-linkage in the sense that te does not specify which semantic relation is intended by the speaker.
3.3. Cancellability test. Grice (1975) proposes several diagnostic tests for conversational implicature, of which the so-called cancellability test is the most prominent. Conversational implicatures can be cancelled without yielding contradiction, as with and in (12a). By contrast, if something is asserted, denying (part of) it will result in contradiction, as with before in (12b).
(12)a. They had a baby and got married, but not necessarily in that order.
b. #They had a baby before they got married, but not necessarily in that order.
(# indicates that the sentence is deviant.)
Similarly, the cause relation associated with a TE-construction is cancellable and hence can be taken as a conversational implicature.
(13) kaze o hiiteatama ga itai. atama ga itai no wa itsumo no koto dakedo.
'(I) caught a cold, and (my) head aches. I always have a headache, though.'
If only the first sentence were supplied, it would naturally be implicated that the cold is the cause of the speaker's headache. Here, however, this implicature is cancelled by the second sentence, indicating that the speaker always has a headache anyway. In a typical scenario the speaker, after uttering the first sentence, realizes the potential implicature and cancels it explicitly.
The temporal sequence relation is likewise cancellable, and hence it, too, can be regarded as a conversational implicature.
(14) maki wa oosaka e itte hiro wa oosaka kara
kaette-kuru. hiro ga kaette-kuru no ga saki dakedo.
'Maki will go to Osaka, and Hiro will come back from Osaka. Hiro's return comes first, though.'
It might appear, therefore, that the semantic relations of cause and temporal sequence are not part of the conventional meaning of TE-linkage at all, but are derived by means of pragmatic principles.
4.1. Temporal sequence relation. It is frequently claimed in the literature that one of the major uses of TE-linkage is to express temporal sequence or consecutiveness (Matsuo 1936, NLRI 1951, Negishi 1970, Kuno 1973, Takahashi 1975, Morita 1980, Endo 1982, Konoshima 1983, Narita 1983, Hamada 1985, Matsuda 1985). In this section, it is argued to the contrary that temporal sequence per se cannot be expressed by TE-linkage.
Given appropriate pairs of clauses, temporal sequence can always be implicated when two clauses are in parataxis, as in (15).
(15)a. maki ga tachiagatta. mado ga aita.
'Maki stood up. The window opened.'
b. maki ga kaijoo ni tsuita. kooen ga hajimatta.
'Maki arrived at the meeting place. The lecture began.'
However, the same temporal sequence cannot be implicated when such pairs of clauses are linked by te, as illustrated in (16).10
(16)a. #maki ga tachiagatte mado ga aita.
'Maki stood up, and the window opened.'
b. #kodomo ga kaijoo ni tsuite kooen ga hajimatta.
'A child arrived at the meeting place, and the lecture began.'
Significantly, there would be no unnaturalness here if the connective to (with a necessary alteration to the inflection of the preceding predicate) were used instead of te, as shown in (17).
(17)a. maki ga tachiagaru to mado ga aita.
'Maki stood up, and the window opened.'
b. Kodomo ga kaijoo ni tsuku to kooen ga hajimatta.
'A child arrived at the meeting place, and the lecture began.'
The sentences in (17) now permit temporal sequence interpretations. There is thus nothing inherently anomalous about conjoining the two clauses in each pair in (16) — i.e., the anomaly is not purely pragmatic, as it would be in Joan ate sushi, and the tower collapsed.
Observe that a small alteration in (16a-b) enhances the acceptability (18a-b):
(18)a. maki ga oogoe o dashite mado ga aita.
'Maki screamed, and the window opened.'
b. kooshi ga kaijoo ni tsuite kooen ga hajimatta.
'The lecturer arrived at the meeting place, and the lecture began.'
Changing tachiagar- 'stand up' in (16a) to oogoe o das- 'scream' in (18a) improves the naturalness somewhat because an extremely loud sound can, in principle, cause windows to open. In (18b), replacement of the subject kodomo with kooshi 'lecturer' makes the sentence perfectly natural because it is precisely the arrival of the lecturer that enables the lecture to begin. The key in both cases is the notion of causation. If native speakers of Japanese are forced to interpret (16), they read in some sort of cause relation above and beyond mere temporal sequence -- e.g., Maki has the magical power to open windows by standing up.
If TE-linkage were in fact able to express a temporal sequence relation, then all naturally occurring event sequences should be compatible with TE-linkage. However, as shown above, this is not the case. From the anomalies observed in such sentences as (16), I therefore conclude that a mere incidental sequence of events -- i.e. pure temporal sequence proper -- cannot be expressed by the use of TE-linkage. The question, then, becomes 'What makes sequences of situations nonincidental?' As suggested above, the notion of causation is one factor that plays a central role.
It has been claimed that TE links two constituents more tightly than does to (Kuno 1973, Matsuda 1985). Of course, many interpretations could be given to the word tightly, and the authors just cited in fact have several senses in mind. But if we choose to interpret it as the involvement of some semantic notion of causation, this characterization provides a partial account of the inappropriateness of TE in the sentences in (16), in which the pairs of clauses fail to show any obvious cause relations. The next section will discuss what is generally meant by the term causation, and how these considerations contribute to our understanding of TE-linkage.
4.2. Causation. Humans do not perceive the physical world as a constantly changing stream of disconnected and arbitrary happenings, but rather as situations occurring in organized patterns over specific spans of time (Minsky 1975, Schank and Abelson 1977, Bullock et al. 1982, Shultz 1982). Bullock et al. (1982:209) claim that the fundamental basis on which humans assign boundaries to discrete situations is constituted by our tendency to perceive or infer cause-effect relations. In the act of cutting bread, for example, we regard the parting of the bread as being caused by the knife's action, rather than taking the scene as involving two simultaneous but disconnected sequences of knife movements and bread movements. As Bullock et al. note, 'First, by imposing a causal connection, we efficiently collapse a series of temporally successive motions into a single event. Second, by this bracketing into causal events, we not only separate meaningful, coherent patterns from all that goes on around us, but also impart structure to the world. When we attribute the parting of the bread to the knife's action, we relate actions to results, transformations to outcomes, and thus construct our own physical reality' (ibid., 210).
In analyzing TE-linkage, we need to keep in mind that the semantic relation cause is fundamentally interpretive: it signifies the speaker's interpretation of a succession of events and, in turn, the hearer's confirmation of such an interpretation. Humans bracket sequences of discrete situations in certain ways, which reflect our innate perception of physical and psychological reality. The fundamental use of TE-linkage is to express such bracketed situations.
4.3. Abductive interpretation of reality. Comparing the usages of the connectives TE and to, Hamada (1985:177) proposes an interesting generalization regarding TE-linkage. Although her formulation is rather vague, it seems possible to interpret it as follows: while to is utilized when the speaker reports two successive situations from a mere observer's point of view, TE is utilized when the speaker has internalized ('digested', as Hamada puts it) the situations. In this section I will attempt to elaborate on this generalization, which captures native speakers' intuitions about TE-linkage.
TE-linkage indicates that the speaker has abductively determined the principle which governs the two situations, and expresses them in the light of his/her own interpretation.11 The abductive mode of inference differs significantly from traditional deduction and induction. Deduction applies a principle (law) to an observed case and predicts a result, e.g. (19); induction proceeds from observed cases to establish a principle, e.g. (20).
(19) Principle: All linguists are sarcastic.
Observation: Ali's wife is a linguist.
Inference: Therefore, she must be sarcastic.
(20) Observation: Beth is a linguist and sarcastic.
Observation: Chris is a linguist and sarcastic.
Observation: Doris is a linguist and sarcastic.
...
Principle: Therefore, all/most linguists are sarcastic.
By contrast, 'abduction proceeds from an observed result, invokes a law, and infers that something may be the case' (Andersen 1973:775). The reasoning in (21), for example, involves an abductive inference.
(21) Observation: This article is nasty.
Invoked Principle: All/Most linguists are nasty.
Inference: Therefore, this article might well have been written by a linguist.
Note that a given situation (result) can in general evoke many different principles. One might, for example, have invoked the principle that people usually become nasty when they are hungry; then the inference would be that the writer might have been hungry when s/he wrote the article.
The invoked principle might not be something that is already known, but could be something that is conjectured on the spot. One may infer the principle from which the observation makes sense. The crucial step lying at the heart of all abductive reasoning is the choice of some particular principle, a choice which is inevitably subjective and context-dependent.
With TE-linkage, the speaker observes two situations which evoke some principle. S/he then conjoins the corresponding clauses with TE, assuming that the same principle will be evoked in the addressee's mind. For example, in (22), the speaker has observed a bad economic situation and an increase in the unemployment rate; these two states of affairs have evoked in his/her mind the principle that bad economic situations cause the unemployment rate to increase; the speaker now presents the two situations with TE-linkage, assuming that the addressee will interpret the clauses as standing in a cause relation to one another.
(22)keeki ga warukuteshitsugyooritsu ga agatta.
'Because the economic situation was bad, the unemployment rate increased.'
Had the speaker failed to recognize a cause relation between the bad economic situation and the increase in the unemployment rate, s/he would simply report the co-occurrence as such, using not TE but the conjunction to, as shown in (23).
(23) keeki ga warui to shitsugyooritsu ga agatta.
'When the economic situation was bad, the unemployment rate increased.'
(24) jon wa shukudai o shite ofuro ni haitta.
'John did his homework and took a bath.'
The linked clauses in (24) are normally interpreted as having a temporal sequence relation. Crucially, however, we do not consider this to be an incidental temporal sequence. The temporal alignment is nonincidental because the two actions have both been brought about through the intention of the same individual. Indeed, the very fact that the same human being is involved (agentively) in both clauses sets up an overwhelming expectation in the hearer that the two actions will be intentionally related in some way. The following subsections explore the licensing of TE-linkage through the notion of human intention.
5.1. Perceived intention. We have intuitive ideas about our own actions as directed toward achieving various goals or bringing about various states of affairs, and under normal circumstances we perceive other people's actions in the same way. In other words, we perceive the other's actions by analogical reasoning from our own actions. This is possible because humans have an innate awareness of the similarities between themselves and others (Miller and Johnson-Laird 1976:101-2). Human infants display a special interest in the human face and the human voice. Indeed, every species has some mechanism for recognizing its own members, for obvious biological reasons (ibid.).
Intentions are 'pro-attitudes', or conduct-controllers (Bratman 1987:7), and in concert with belief they move us to act. Human intentional actions, both the speaker's and everyone else's, are (and are perceived as) plan-based and goal-oriented. Human beings typically plan to perform their actions in a certain sequence, as in (24) above, and intentions play a motivational role in such planning. A particularly salient example of goal-orientedness is actions which involve a means-end relation, e.g. (25); with such sentences, the intention behind performing the first act is to achieve the second act.
(25)a. haha ni denwa-shite okane o karita.
'(I) called (my) mother and borrowed (some) money.'
b. renga o katte ie o tateta.
'(I) bought bricks and built a house.'
These sentences show that when an event sequence is perceived as a succession of intentional acts of a single individual, the corresponding clauses can naturally be linked by te. The reason is not difficult to see. Humans experience, and in turn describe, sequences of events that involve voluntary actions differently from event sequences that do not; we seldom consider a series of acts by a rational being to be random coincidence, but as something with some degree of intention (and rationale) behind it. This has immediate consequences at the more concrete level of syntax and semantics. If the conjuncts share an agentive subject, the temporal sequence relation indeed appears to be compatible with te-linkage, but only because some degree of intention is automatically read into the sequence. It is the perceived intention, and not the temporal sequence proper, which licenses the use of te-linkage.
5.2. Simultaneous relation and te-linkage. Because TE-linkage involves a non-finite first clause, identification of the temporal alignment between the two denoted situations is an issue that deserves further discussion. Harasawa (1994: 182) contends that (26) is ambiguous: on his account, the two situations can be considered either in a simultaneous or consecutive (i.e. temporal sequence) relation.
(26) jon wa terebi o mi-te benkyoo-shita.(=(1b))
'John studied, watching TV.'
(simultaneous reading)
'Having watched TV, John studied.'
(consecutive reading)
Sentence (26) is indeed ambiguous, but not between simultaneous and consecutive readings as Harasawa claims. As will be demonstrated in a moment, the sentence cannot be used to express a simultaneous relation proper, i.e. John watched TV and studied simultaneously, or While watching TV, John studied. Rather, the ambiguity is between a means-end reading, By watching TV, John studied, and a temporal sequence reading, John watched TV, and then studied. (Note that the agentive subject jon is shared by both conjuncts, thereby making the temporal sequence reading possible.)
In fact, mere simultaneity, like mere temporal sequence, is one of the few semantic relations incompatible with TE-linkage. The key point is that a simultaneity relation is reciprocal -- i.e., if we ignore the focus shift, John studied while watching TV and John watched TV while studying depict exactly the same scene. Thus the possibility of exchanging two clauses is a good diagnostic for whether the sentence truly permits a simultaneous reading. Consider (27), where the clauses of (26) are reversed.
(27) jon wa benkyoo-shite terebi o mita.
#'John watched TV, studying.'
'Having studied, John watched TV.'
Here only a temporal sequence reading is possible. Simultaneity is excluded for (27), and hence (as argued) for (26) as well. Indeed, in (27) even a means-end reading is impossible. The means-end reading is factored out because one can study, say a foreign language, by watching TV, but watching TV by studying lacks any sensible interpretation.
Consider finally (28).
(28) #hon o yonde-ite basu ga kita.
'When (I) was reading a book, the bus came.' (Intended)
Unlike (27), in which TE-linkage is licensed by the shared agentive subject, sentence (28) is totally anomalous. Because the conjuncts have distinct subjects, no human intention to connect the two situations can be inferred. There is only abstract simultaneity, and that is insufficient to license TE-linkage.
These considerations lead us to the important conclusion that the expression of extrinsic, abstract temporal alignment per se -- e.g. temporal sequence or simultaneity proper, divorced from human concerns -- is not a component of the meaning of TE-linkage.12
5.3. Controllability constraint. Kuno (1973:196-97) observes that in TE-linkage with identical subjects, both clauses must be either self-controllable (agentive) or non-self-controllable (nonagentive) — the controllability constraint. In his view, violation of this constraint leads to anomalous sentences. In this section I will examine Kuno's examples and demonstrate how the analysis proposed in this article can account for the anomalies in his examples.
Kuno considers that (29a-c) are anomalous because of the violation of the controllability constraint, i.e., jon is the (nonagentive) patient/theme in the first clause but the agent in the second.13
(29)a. jon wa asa me o samashite kao o aratta.
'John woke up in the morning and washed his face.'
b. jon wa marii ni guuzen deatte sono hanashi o shita.
'John ran into Mary accidentally and talked about it (i.e. some matter or other).'
c. jon wa hikoojoo ni tsuite ie ni denwa shita.
'John arrived at the airport and called home.'
When there is agreement in controllability, the sentences sound quite a bit more natural: in (30a,b) both clauses are controllable, and in (30c) both are noncontrollable. ((30a,c) are Kuno's examples, while (30b) is mine.)
(30)a. jon wa asa okite kao o aratta.
'John got up in the morning and washed his face.'
b. jon wa marii ni denwa-shite sono hanashi o shita.
'John telephoned Mary and talked about it (some matter).'
c. jon wa hikoojoo ni tsuite nimotsu no kensa o uketa.
'John arrived at the airport and underwent the inspection of his luggage.'
As discussed in Section 5.1, te is appropriate in (30a,b) because the two events are related in a principled way, viz. via the perceived intention of the single individual. Sentence (30c), on the other hand, is also acceptable because it evokes a cause relation, although not a prototypical one. John's arrival at the airport and being subjected to the inspection are not just an accidental sequence. The naturalness of the sentences in (30), therefore, can readily be explained by the analysis proposed in this study.
Let us now compare the sentences in (29) and (31), all of which violate the controllability constraint. In all these examples the first clause is non-self-controllable, whereas the second is self-controllable. The sentences in (29) are indeed awkward; but those in (31) are perfectly natural.
(31)a. saifu o nakushite tomodachi ni okane o karita.
'(I) lost (my) purse and borrowed money from a friend.'
b. kaze o hiite yasumimashita. (Minami 1974: 122)
'(I) caught a cold and took a day-off.'
c. furarete yakezake o nonda.
'(I) got jilted and drank out of desperation.'
What, then, accounts for the difference in naturalness between (29) and (31)? I contend that the anomaly of (29) boils down to what does and does not constitute an acceptable reason explanation. In (31) the first clauses supply a normally acceptable reason for the action denoted by the corresponding second clause, whereas in (29) they fail to do so. In general, when the second conjunct indicates an action and the first conjunct indicates a non-action or an action with a distinct agent (i.e., the first situation is not controllable by the agent of the second clause), native speakers are inclined to consider the first conjunct a reason, rather than a cause in its ordinary sense. In the next section the differences between causes and reasons are discussed.
5.4. Causes and reasons. Prototypically, causation applies to the world of physical entities and natural laws, whereas reasons concern human beings and their intentions. On the other hand, it has often been argued that reasons are themselves causal in nature (cf. e.g. Davidson 1980). Ordinary language sometimes employs a word corresponding to cause, and frequently a word corresponding to because, even where reasons and not causes are involved. Further, there are clear regularities obtaining between reasons and actions, regularities similar to those that lie at the heart of the cause relation in the Humean conception of causation. Donnellan, however, argues that while appeal to such facts 'may shift the burden of proof to the other side, it does little to establish that reason explanations are straightforward causal explanations. The word "because" may have a different use in these circumstances, or it might only be a way of emphasizing, somewhat metaphorically, the "compelling" nature of the reason' (1967:86). He also argues that 'while regularity is the core of the causal relation for Hume, it must be a regularity of a certain kind: an empirical regularity. Whether the connection between reasons and actions is merely empirical has been strongly questioned' (ibid.).
Donnellan points out the following differences between reasons and causes. First, the agent seems to have a privileged and self-sufficient position concerning the reasons (though not the causes) underlying his/her own actions. In the normal case, the agent need not appeal to evidence and empirical investigation to establish what the reasons are. Second, humans seem willing to accept a reason explanation without demanding any generalization (to a larger class of analogous cases) of the relationship between the particular action and the particular reason. These two characteristics are foreign to causal explanation. Third, while causal explanation depends on the empirical and contingent nature of the causal connection, there is a more than contingent relation between reasons and actions. An action is performed because the actor desires the outcome of the action. To give a reason is to indicate, explicitly or implicitly, such wanting. However, wanting is nothing but a tendency to act;14 to want to do something is to be prepared under certain circumstances to take the necessary steps.15 If wanting is conceived as a tendency to action, then by that very token there is a logical (analytic) connection between wanting and action; thus it would be odd to count wanting as a cause of an action and thus to construe its relation to the action as merely contingent.16
From a linguistic point of view, it is apparent that when we describe or explain a situation involving a person, we make a distinction between those situations in which the person acts intentionally (i.e., the person is an agent) and those in which s/he has no control over what happens. Language provides a rich vocabulary for distinguishing the agentive and nonagentive roles of the person as a participant in the described situation (cf. Fillmore 1968; Lyons 1968, 1977; Talmy 1976).
When we conceive or perceive a person as an agent, we expect his/her behavior to be a succession of rational, purposeful actions. For example, if someone with whom I am walking suddenly stops, I will automatically think that there is a reason for her action. If I find that she is looking at something, I will understand that she has stopped because she wanted to see that object. If the reason is not obvious, I will ask her what happened in her mind (asking for a reason explanation). If she does not explain and says 'Nothing,' then I will think the reason was trivial and ignore it. However, if she keeps on stopping, I will then start to worry and will ask her if she is sick (asking for a causal explanation). If she still does not explain, I will think that she is bizarre and that my rational expectations do not work in her case.
These considerations are relevant to the semantics and pragmatics of TE-linkage. TE-linkage is used to express a nonincidental sequence of situations; and such sequences include (i) those in which humans normally perceive the first situation as a cause of the second, (ii) those in which both situations involve an action performed by the same individual, and (iii) those in which the first situation is to be regarded as the reason for the second. These principles reflect human strategies of bracketing surrounding situations.
Let us now reconsider Kuno's examples (29). In all the sentences in (29), the first conjunct is a non-action, while the second is an action; furthermore, no obvious causation is involved. According to the analysis proposed above, then, the first conjunct can only be construed as the reason for the action, i.e. as an instance of (iii); (i) and (ii) are excluded. The anomaly in (29) arises simply because the first conjunct does not provide an acceptable reason.
Consider the following conversations, which are English constructs similar to (29):
(32)a. A: Why did you wash your face?
B: Because I woke up.
b. A: Why did you talk to Mary about it?
B: Because I accidentally ran into her.
c. A: Why did you call home?
B: Because I arrived at the airport.
While all three conversations sound somewhat strange to me, (32a) sounds the worst and (32c) the best, and the differences reflect the varying plausibility of the reasons presented in the three sequences. Interestingly, there seems to be a correlation between these judgments regarding acceptable reasoning and the grammaticality judgments of (29). I feel that (29c) is slightly better than the others, and that (29a) is the worst. (When the asserted reason is blatantly inappropriate, it can even create a comical effect, e.g. to answer the question Why are you going to divorce? with Because we got married.) On this view, if the first conjunct counts as an acceptable reason (explicitly or implicitly) for performing the action referred to by the second conjunct, a disagreement in controllability should not affect the acceptability of the sentence. The sentences in (31) confirm this prediction.
Kuno's controllability constraint, accordingly, cannot be upheld as a syntactic principle. Rather, the awkwardness of his examples is due to the conflict between the principles of interpretation found with te-linkage, on the one hand, and the interpreter's standards regarding what can plausibly count as a reason, on the other.
(33) Additive Relation
(kyonen wa hidoi toshi datta.) [jishin ga atte], [kaji ga
atte], ...
'(It was a terrible year last year.) There was an earthquake and a fire ...'
(34) Contrastive Relation
[minami no kuni wa atsukute], [kita no kuni wa suzushii].
(Morita 1980: 315)
'It's hot in southern countries, but it's cool in northern countries.'
In both (33) and (34) no cause or reason relation is inferable, and there is no human intention to connect two situations; yet the sentences are natural. What licenses TE-linkage in these cases is the fact that the collocated situations are not arbitrarily chosen by the speaker. Here, again, TE-linkage indicates the speaker's conceptualization of two aspects of the surrounding reality as being related in some principled way.
Van Dijk (1977: 41) points out that in order to make intelligible sense out of merely juxtaposed clauses, it is usually necessary to supply a context which specifies the 'when, where, or why' of the conjuncts. For example, in order to interpret Mary knitted, and the fire was burning, the interpreter needs to presuppose some proposition which specifies a general topological identity for the two conjuncts, such as 'I came into the room'. This is precisely the case for (33). In (33), although both conjuncts refer to events, the temporal alignment is not what is focused on. Rather, the two events are presented as supporting evidence of the first sentence. The speaker has bracketed the two situations together because they are events of the same type vis-à-vis the current purpose of the discourse.
In (34), the two situations are naturally understood as contrastive because of the lexical properties of the predicates atsu- 'be hot' and suzushi- 'be cool'. The contrastive relation may require more reasoning than that involved in a simple lexical contrast, however. In fact, discovering a contrastive relation frequently involves an abductive reasoning process, as in (35).
(35) jimintoo wa antee-tasuu o kakuho shite shakaitoo wa teeraku ga tsuzuita.
'The Liberal Democratic Party secured a firm majority, but the Japan Socialist Party continued its decline.'
In (35) the interpreter must invoke the principles that for a political party to secure a firm majority requires gaining sufficient votes, and that for it to decline indicates its loss of votes. When this much of the inference has been made, then recognizing a contrastive relation becomes a matter of lexical contrast.
To sum up, whenever te links conjuncts -- temporal (e.g. cause, reason) or atemporal (e.g. additive, contrastive) -- the use of TE-linkage guarantees that the speaker has recognized some principle binding together the described situations.
7.1. Semantical hypothesis. Cohen (1971) analyzes cancellable meanings differently from Grice -- an analysis he calls the semantical hypothesis. Under the Semantical Hypothesis, the meaning of and is much richer than that of the truth-functional connective &. 'In addition to expressing the conjunction of two truths it also indicates that the second truth to be mentioned is a further item of the same kind, or in the same sequence, or of a kind belonging to the same set of commonly associated kinds of item, or etc. etc., as the first truth to be mentioned' (ibid., 55). Consider (36), in which the semantic relations temporal sequence and cause are cancelled.
(36) The old king has died of a heart attack and a republic has been declared, but I don't know which of these two events preceded the other nor do I wish to suggest some connection tends to exist between two such events. (ibid., 54)
For Cohen, what is cancelled in (36) is not conversational implicature, but is 'a feature that is one of those features which should be listed in any adequate dictionary entry for the word' (ibid., 55) -- in this instance, presumably, the dictionary entry for the word and. A more concrete example will help clarify what is involved in this approach. The conversational-implicature approach would posit that the word flower in the sentence This is a flower implicates that the object in question forms part of a plant, and that this implicature is in turn cancelled by This is a plastic flower. Cohen, to the contrary, argues that in this example some part of the normal meaning of the noun flower, rather than any implicature, is what is cancelled by the adjective plastic. Such a view of 'meaning', it should be noted, is quite different from Grice's. For Grice, a word's semantic 'meaning' is the invariant part of its meaning -- a minimalist view. Cohen, on the other hand, takes a maximalist position, whereby a word's meaning encompasses all of its (apparent) subsenses in a singly whole -- with the proviso that some of these meaning components can be cancelled.
Are we then to explicitly attribute temporal sequence, cause, and all other compatible semantic relations to the lexical entry for and in English and TE in Japanese? If so, does this not violate Ockham's Razor, the metatheoretical principle which states 'Do not multiply entities beyond necessity'? Does it not say more than is actually necessary? Cohen claims that under the Semantical Hypothesis there is in fact no such violation. Both his approach and Grice's, he asserts, involve only a single lexical meaning for the given word. Grice's 'meaning', in truth-conditional terms, says less and hence is weaker; Cohen's 'meaning', as formulated within the Semantical Hypothesis, is richer and hence stronger; but both approaches posit just one meaning as the lexical meaning of and. Hence neither theory has any advantage over the other regarding lexicographical simplicity.
Cohen's approach is plausible in the case of plastic flower, but difficult to envisage in the case of TE. The question is, the truth-conditional meaning being the weaker one, whether or not there is, correspondingly, some specific 'strong' definition of TE-linkage which can then be cancelled. Cohen does not supply candidates. Let us, therefore, turn to Kortmann (1991) to yield one possible approach to such a 'strong' meaning.
7.2. Semantic informativeness. Investigating English free adjuncts and absolutes, Kortmann (1991: 119-121) hypothesizes that identification of the semantic relation(s) between a given free adjunct or absolute and the matrix clause is essentially determined by a scale of semantic informativeness or specificness. Analyzing 1,681 examples of present-participial free adjuncts and absolutes in his corpus, Kortmann proposes the following scale (the line indicates the border between categories deemed 'more informative' and 'less informative' categories):
'More informative' semantic relations require more knowledge or evidence on the interpreter's part than do 'less informative' ones. For example, to interpret the semantic relation between the two propositions as one of adverbial modification, e.g. concession, condition, cause, rather than as simple accompanying circumstance, requires more knowledge of the world, as illustrated in (37).
(37) Serving the two portions ... Royce remained silent, ...
(Kortmann 1991: 122)
The semantic relation between the main and the participial clause in (37) can be a concession, i.e. Although serving the two portions, Royce remained silent, or a cause, i.e. Because he served the two portions, Royce remained silent, or finally a mere accompanying circumstance, While serving the two portions, Royce remained silent. 'Lack of substantiation of a "more informative" member [of the scale] will lead to the selection of a "less informative" one. On the other hand, if the information in the preceding context, for example, makes it appear justified, then a "more informative" interpretation may be chosen' (ibid., 129).17
7.3. Semantic informativeness and TE-linkage. We can now draw on Kortmann's framework in an attempt to apply Cohen's Semantical Hypothesis to the analysis of TE-linkage. Might one perhaps postulate a semantic informativeness scale similar to that proposed by Kortmann and then take the most informative relation as actually being the 'meaning' of TE-linkage?
The answer appears to be negative. It seems impossible to posit concession, the most 'informative' relation on Kortmann's scale, as the 'meaning' of TE-linkage and then cancel part of this meaning when the concessive reading is inappropriate.18 One might instead propose temporal sequence as a more realistic candidate for the 'meaning' than concession. However, as demonstrated above, the genuine temporal sequence relation cannot be expressed by the use of TE-linkage. Furthermore, under the Semantical Hypothesis a meaning 'stronger' than the one posited in the lexicon cannot be inferred: cause is certainly more informative (hence stronger) than temporal sequence, and TE-linkage is indeed compatible with cause; yet if we posit temporal sequence as the 'meaning' of te-linkage, there is no mechanism to account for the stronger cause readings in the Semantical Hypothesis. The final possibility is to consider cause itself to be the 'meaning' of TE-linkage. But we would then need to posit more meanings, e.g. means, because there are many TE-compatible semantic relations that cannot be deduced from cause.
I have argued that there is a meaning of TE-linkage that is stronger than that of the truth-conditional &; viz. the speaker presents the two situations referred to by the conjuncts as nonincidental. Still operating within Cohen's framework, might we posit this as the 'meaning' of TE-linkage? Here a central problem is that this meaning belongs to a different realm from that of the semantic relations which Cohen has discussed: it makes essential reference to the human world. Furthermore, nonincidentalness is not cancellable under any circumstances. For example:
(38) maki wa tookyoo e itte guuzen ni hiro ni deatta.
A: 'Maki went to Tokyo and accidentally ran into Hiro.'
B: 'Maki went to Tokyo and unexpectedly ran into Hiro.'
At first glance, (38) appears to describe an incidental sequence of events because of the phrase guuzen ni 'accidentally'. A careful examination reveals, however, that this sequence of events is nonincidental. It is Maki's going to Tokyo which enables her to run into Hiro.19 What guuzen ni signifies here is unexpectedness from the subject-referent's viewpoint, rather than any interpretation on the speaker's part that the two events are truly disconnected. Therefore, in (38) (B) is a more accurate translation than (A), and the sequence remains nonincidental.
I conclude that the approach suggested by Cohen, though insightful and deserving of serious consideration, is not very promising empirically when applied to the problem of Japanese TE. The analysis proposed in this article is not compatible with, and provides no support for, either Grice's or Cohen's analysis of cancellable meanings. Rather, I contend that in order to analyze the meanings of 'fuzzy' connectives like TE, it is necessary to look beyond the traditional semantic paradigm and invoke notions such as incidentalness.
In this paper, I have presented a different approach to the analysis of the connective TE, one which utilizes the vocabulary of human cognition. I contend that TEis not a mere syntactic device for conjoining two clauses, but has its own semantic value. Through the use of TE-linkage, the speaker presents the two situations as being related in some principled way -- e.g. causation, intention of a single individual, or reason for an action -- so that the presented situations are viewed as nonincidental. This very human principle, in fact, is the semantic value of TE-linkage.
Copyright (c) 1996 Yoko Hasegawa
Since 12/24/96