Commentary on Lowe and Mazaudon 
Steven Lee Hartman 1 
(Southern Illinois University) 
The Reconstruction Engine (RE) described by Lowe and Mazaudon takes the linguist's 
hypotheses about sound correspondences between (on one hand) several presumably 
related modern languages and (on the other) a reconstruction of their presumed com- 
mon ancestor language, and it tests these hypotheses against a major portion of the 
lexicons of the languages. 
Five advantages of the program especially are worthy of note: (1) It can tabulate 
results for several languages at once. (2) It can project correspondences both "down- 
stream" and "upstream." (3) Its initial formation of cognate sets is not encumbered by 
semantic information (although word glosses are kept available for the user eventually 
to judge the plausibility of matches)--this ensures that the process does not overlook 
viable matches simply because of "small" semantic differences ('shoulder'/'thigh') or 
nonuniform rendering of glosses ('soak'/'wet,' 'snow'/'neige'). (4) It overcomes some 
of the dangers of losing potential matches to nonuniform parsing of syllable structure, 
by considering a variety of possible segmentations, based on partially overlapping 
categories (rhyme, vowel, etc.). (5) It incorporates "flexibility toward irregularity" (as 
in the use of "fuzzy" matching) as a kind of wider deployment of the net to catch 
cognates and as a way of keeping the process open to new discoveries. The authors 
rightly imply that the difference between regular and irregular sound change is not 
marked by a clearly drawn line based on simple observation, but rather depends on 
theoretical judgments made by the linguist. 
Meanwhile, the applicability of the RE in its present form is perhaps limited in a 
fundamental way by looking only at sound correspondences between two stages ("an- 
cestor" and "descendant"), rather than dealing with the undoubtedly multi-layered 
series of changes (not to mention the possibility that those changes may overlap in 
history, compete for input, and/or apply in different orders for different words). This 
may be merely the quibble of one who has worked mainly with a documented ancestor 
language (Latin), having had less experience with the problems of reconstructing an 
ancestor language. From this perspective, the limitations of a two-level table of corre- 
spondences can be brought into focus by considering for example the table that would 
be necessary to account for just the segment/o/in Modern French (derived from dif- 
ferent Latin ancestor sequences in haut, eau, h6tel, chevaux, chateau, etc.). The authors 
intend, rightly, to deal with the "chronologization of changes" in future refinements 
of the RE. 
The notation system and parsing algorithms of the RE seem well suited to the 
phonotactics of "monosyllabic" languages such as the Tamang of the examples pre- 
sented. With data from other language families, with polysyllabic strings and more 
complex consonant clusters, it might seem that the processing time to consider all pos- 
sible correspondences could increase geometrically; but other researchers (e.g., John 
Hewson \[this issue\]--working with the very different morpheme structure of Algo- 
nquian languages) have developed techniques for resolving that problem. 
1 Department of Foreign Languages, Southern Illinois University, Carbondale, IL 62901, 
418 
