T HE RELATIONSHIP BETWEEN SINGLE EXPERIMENTS AND HYPOTHESES / THEORIES : TYPES OF

For decades, linguistic data have been viewed as relatively unproblematic entities which pro-vide rich information about linguistic behaviour directly. Epro-vidence was interpreted as a special subset of data, whose certainty is guaranteed by experience (“observation”) and intersubjective testability. Thus, evidence was supposed to provide a firm base for the testing of theories and deciding among rival theories. However, this view is untenable for several reasons. First, it has become generally acknowledged in the philosophy of science that experience cannot guarantee the truth of a statement. Second, intersubjective testability does not eliminate the subjectivity immanent in individual experience as a source of knowledge completely. If n persons evaluate a phenomenon in the same way, it does not follow that the n+1^th person will also agree. That is, the criterion of intersubjectivity is built on induction (that is, on a type of plausible infer-ence), since it infers from a finite number of cases to an infinite number of cases. Third, expe-riences have to be interpreted: the object of one’s experience has to be described with the help of a category system, that is, a theory. Consequently, it is not directly the experience to which one compares the hypotheses of the theory, but its processed and interpreted version.

Therefore, we need a concept of ‘data’ and ‘evidence’ which takes into consideration their uncertainty and complexity.

14.1. The p-model’s concept of ‘data’

The p-model by Kertész & Rákosi does not identify data with “linguistic examples” such as the following Hungarian sentences:

Ennek a tyúknak már megint agymenése van.

[‘This chick appears to have suffered a brainstorm again.’]

Ez a bikapiac sem tart örökké!

[‘This bull market won’t last forever, either!’]

Rather, their structure consists of two components: a statement with an information content and a plausibility value supported by a direct source (see Section 4.2.2):

(D) A datum is a statement with a positive plausibility value originating from some direct source.

This means that data are not inferred from other statements but constitute the starting points of a theory (plausible argumentation process). The above “examples” can be transformed into data as follows:

0 < |The Hungarian sentences Ennek a tyúknak már megint agymenése volt. and Ez a bikapiac sem tart örökké! contain conventional metaphors.|S < 1

where S is a compound of the linguist’s linguistic intuition and linguistic knowledge (i.e., some theory of metaphors).

Accordingly, data are in most cases not claimed to be true with certainty, but they are usually more or less plausible “truth candidates”. Their plausibility is usually supported by the sources to some extent, but the sources are not able to make them certainly true. Nevertheless, a datum must possess a certain degree of initial plausibility, that is, it has to receive a plausibility value from some reliable source. Statements which are of neutral plausibility or implausible accord-ing to any sources in the p-context, are not data in this sense.

Experimental data can be regarded as data in the sense of (D). Although it is possible to provide a partial reconstruction of the argumentation process leading to the creation of the experimental data, the amount of information which cannot be found in the experimental report and additional materials is too large. Therefore, while the plausibility of experimental data originates from an indirect source from the point of view of the person(s) who conducted the experiment, it is more appropriate to treat them as ‘data’ in the sense of (D) from the point of view of the reader of the experimental report. Nonetheless, the reliability of their source is strongly influenced by pieces of information pertaining to the components of the experimental process.

14.2. The p-model’s concept of ‘evidence’

Within the p-model, it is possible to define three types of evidence in order to grasp the rela-tionship between data and other hypotheses of the theory.

Weak evidence for a hypothesis H simply means that we can build inference(s) on the given datum that make(s) h plausible (in the extreme case true with certainty). Weak evidence against a hypothesis H means a datum on which we can build inference(s) that make(s) h im-plausible (in the extreme case false with certainty):

(EW) (a) A datum D is weak evidence for hypothesis H, if the p-context contains state-ments that extend D into an indirect source on the basis of which a positive plau-sibility value can be assigned to H.

(b) A datum D is weak evidence against hypothesis H, if the p-context contains statements that extend D into an indirect source on the basis of which a positive plausibility value can be assigned to ~H.

From this definition it follows that a datum can be weak evidence for a statement and for its rival simultaneously, although it may support them to different extents. The strength of the support it is capable of providing is determined by the peculiarities of the plausible inference connecting the datum and the hypothesis at issue: the plausibility of the datum and the plausi-bility of the necessary latent background assumptions.⁹⁰ That is, the more plausible the datum is and the stronger the link between the datum and the hypothesis at issue is, the stronger is the

90 On latent background assumptions, see Sections 3.2.7 and 4.2.2.

support this piece of evidence provides to the hypothesis. The strength between the datum and the hypothesis is influenced by the directness of their relationship and the plausibility values of the latent background assumptions. Thus, a great distance between the datum and the hy-pothesis tested means that a higher amount of latent background assumptions are needed, whose plausibility might be dubious or at least, hard or even impossible to be checked.

Relative evidence for a hypothesis H also requires that the datum provides stronger support to H than to its rivals:

(ER) (a) A datum D is relative evidence for hypothesis H, if (i) D is weak evidence for hypothesis H;

(ii) the inference(s) connecting the premises and H provide(s) H with a higher plausibility value than the plausibility values of H’s rivals assigned to them by the inferences also using D as a premise.

(b) A datum D is relative evidence against hypothesis H, if (i) D is weak evidence against hypothesis H;

(ii) the plausible inference(s) connecting the premises and ~H provide(s) ~H with a higher plausibility value than the plausibility value of H assigned to it by the inferences also using D as a premise.

The third type is strong evidence which means that the datum makes only hypothesis H plau-sible and does not provide any support to its rivals:

(ES) (a) A datum D is strong evidence for hypothesis H, if (i) D is weak evidence for hypothesis H;

(ii) D is not weak evidence for any of H’s rivals.

(b) A datum D is strong evidence against hypothesis H, if (i) D is weak evidence against hypothesis H;

(ii) D is not weak evidence against any of H’s rivals.

Evidence is interpreted by the p-model not as a special subset of data but as a datum with a special function relative to some hypothesis of the theory. From this it follows that evidence is not objective, immediately given, theory-independent and completely reliable but source- and theory-dependent and reliable only to a certain extent. Further, data which meet the criteria laid down in (EW)-(ES) in most cases do not perfectly support or refute the given hypothesis. The connection between the datum and the hypothesis is established by plausible inferences relying on plausible premises. A third important corollary of these definitions is that the function of evidence is not restricted to the testing of hypotheses, that is, to the justification of theories, but data and evidence play a role in every stage of the process of linguistic theorising.