Commentary on Lowe and Mazaudon

What Lowe and Mazaudon have done in this article is demonstrate that, by intelligent exploitation of a simple strategy, computerized methods may be used in protolanguage reconstruction for any group of languages. This streamlining enables us to complete, in a matter of months, work that often occupied a scholar's lifetime. The simple strategy can be stated in a single sentence (Hewson 1993:iv): "From the data of the daughter languages generate all possible protoforms, then sort alphabetically, and examine all sets of identical protoforms collocated by the sort." When we used this strategy in the reconstruction of Proto-Algonkian we were fortunate to be dealing with polysyllabic words. By eliminating the vowels and using the consonant frameworks of these words, we bypassed all problems of segmentation: the consonants and consonant clusters remaining were the segments used to generate, by means of the known reflexes, all of the possible protoforms. Lowe and Mazaudon show how this strategy can be used for a language with monosyllables and tones, vastly different typologically from Algonkian. Here a strategy had to be devised for dealing with the various possibilities of segmentation. Such technical adjustments are required for every language family: for IE, for example, programming would need to accommodate ablaut, or strip inflections. They have also streamlined the decision-making process. The linguist is presented at one and the same time with all the different possibilities of reconstruction (see their Figure 3), so that the secondary and tertiary alternatives can very quickly be deleted. In our case a preliminary reconstruction was often based on two cognates, before a preferable reconstruction with three or more cognates appeared further down in the listings. Their engine is like a Rolls Royce when compared to our Model T experimentations in the 1970s, when printout was still restricted to uppercase letters. It is not surprising that polysyllabic Algonkian words generate on average over 20 proto-projections. Most of these would be singletons, and thus filtered out by the sort, which separates the gold nuggets from the tailings. Because of the complexities of segmentation, the Tamang monosyllables also generate large numbers of protoprojections. This fact points directly to the main challenge of the comparative method: it requires finding needles in haystacks, work for which the computer is the machine par excellence. With programs like RE it is now feasible to do the massive amount of low-level reconstruction that needs to be done for the world's language families. The data of the new protolanguage dictionaries may then be compared to create a further, deeper stage of reconstruction: we can compare Proto-Algonkian with Proto-Siouan and ProtoIroquoian, for example (or Proto-Germanic with Proto-Slavic, etc). This work should produce a surer insight into prehistory than the long-range guesswork (which, although limited, has its usefulness) that goes on at the moment.