Scientific generalizations must be reproducible (or replicable) to be accepted as valid, whether they are based on simple observations or controlled observations, i.e. experiments. In linguistics, issues of replicability arise at various levels: (i) replicability of speaker behavior under identical experimental conditions (acceptability judgments, elicitation of narratives); (ii) replicability of descriptions of the same language (either the same corpus: will two linguists arrive at the same description of Gothic on the basis of the Wulfila corpus?; or the same speech community: will two fieldworkers doing fieldwork in the same village arrive at the same grammars?); (iii) replicability of typological coding based on identical descriptions (will two typologists extract the same information from the same grammars?); (iv) replicability of typological generalizations based on different samples (drawn from the same universe of languages). At all these levels, research results should in principle be replicable. For linguistic typology, (iii) and (iv) are the most relevant. In addition to replicating typological generalizations based on a different sample, one could also ask for typological replication based on the same sample (as suggested by a reviewer). But the issues arising in such an exercise would not be distinct from (ii) and (iii). We feel that (iv) is a much more interesting issue for typologists, and we restrict our attention to it in this note. Full-scale replication of an earlier researcher’s results is rarely done in linguistics, perhaps because linguists value novelty more than reliability: Just confirming observations made by others is simply not particularly prestigious. Fieldworkers don’t generally write another grammar or another dictionary of a language spoken by a small community if a comprehensive grammar and a large dictionary have been published within the last decade. With so many languages disappearing, we cannot afford that. In language typology, the situation is similar: typologists do not seem to regard the replication of known facts as sufficiently prestigious to be worth the effort. This is so despite the fact that the empirical basis of typological generalizations is often far from ideal: Language samples tend to be biased, it is even unclear what an unbiased sample would be, grammars show widely varying degrees of thoroughness and explicitness, they adopt different theoretical outlooks, and so on. Every practitioner knows that doing typological surveys is not easy, and some critics have voiced skepticism about the entire typological enterprise as a result of these difficulties (Newmeyer 1998: ch. 6).
[1]
Joseph H. Greenberg,et al.
Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements
,
1990,
On Language.
[2]
Does sampling matter? A test in replicability, concerning numerals.
,
2006
.
[3]
John A. Hawkins,et al.
Word order universals
,
1983
.
[4]
Revere D. Perkins,et al.
The Evolution of Grammar
,
1994
.
[5]
Matthew S. Dryer,et al.
Large Linguistic Areas and Language Sampling
,
1989
.
[6]
David Gil,et al.
The World Atlas of Language Structures
,
2005
.
[7]
M. Dryer.
The Greenbergian word order correlations
,
1992
.
[8]
Matthew S. Dryer,et al.
Object-verb order and adjective-noun order: Dispelling a myth
,
1988
.
[9]
D. Bakker,et al.
Language sampling
,
2019,
The SAGE Encyclopedia of Human Communication Sciences and Disorders.
[10]
F. Newmeyer.
Language Form And Language Function
,
1998
.
[11]
M. Ruhlen.
A Guide to the World’s Languages
,
1987
.