More about causality and probability
July 26, 2007
Over the past few years, a few TS scholars have focused on difficulties in establishing causation in Translation phenomena and have suggested probabilistic rationale as a tool to deal with them. Some clarification may be useful to take these ideas further.
In a text posted in this section in November 2005, it was explained that causation could be hidden by random variation in extraneous variables and could not be detected regularly even if the link was deterministic. One cannot say that a causal link is probabilistic just because empirical observation does not make its effect detectable every time. What is often probabilistic is the detection of the effect, not the causal link itself. This distinction between the probabilistic or deterministic features of detection of a phenomenon and the probabilistic or deterministic nature of the phenomenon being investigated is important.
A second point is that causation may be magnitude-dependent, but it may not be. Some factors may have an effect just by their presence or absence. For instance, the performance of an interpreter may be different if s/he knows s/he is being recorded - there is no “more recording” or “less recording”.
Thirdly, in many cases, causation does involve magnitude, leading to laws such as: ‘the more causal factor A is present, the stronger the effect B will tend to be’. Even in such a case, this magnitude-related effect may hold only between a minimum threshold and a maximum threshold of the cause and/or of the effect. For instance, certain effects of time pressure on translation may be detectable only between say x minutes allotted to the translator per 100 words of text and y minutes per 100 words of texts. Think the therapeutic dose of medical drugs beneath which they will not have any effect on the medical condition being treated. Going beneath x or above y will not lead to significant differences from x or y respectively, or may have effects which follow a different law than within the x-to-y interval. This is one reason why piloting is important in experimental studies, and perhaps one reason why experiments conducted without previous piloting to check for such ‘floor effects’ and ‘ceiling effects’ did not produce convincing results in spite of the possible existence of the effect which was sought.
Fourthly, while detecting and/or measuring causality clearly falls within the scope of ESP (the down-to-earth, technical, empirical science paradigm – see the relevant texts in this Research Issues section), LAP (a more philosophical, reflection-oriented type of research) can have a valuable role to play in suggesting candidate factors for causality, because it tends to address phenomena from a higher vantage point than ESP and may detect potential relations that are more difficult to see for ESP researchers. Complementarity, not mistrust, is the effect to be sought in the co-existence of these two approaches to research in TS.