There are more stems per item in Athabascan, which expresses the fact that the Athabascan languages have undergone somewhat more change in diverging from proto-Athabascan than the Yokuts languages from proto-Yokuts.
This may be because the Athabascan divergence began earlier;
or again because the Athabascan languages spread over a very much larger territory (including three wholly separated areas);
or both.
The differentiation, however, is not very much greater, as shown by the fact that Athabascan shows 3.46 stems per meaning slot as against 2.75 for Yokuts, with a slightly greater number of languages represented in our sample: 24 as against 21.
(On deduction of one-eighth from 3.46, the stem / item rate becomes 3.03 against 2.75 in equivalent number of languages.
) These general facts are mentioned to make clear that the total situation in the two families is similar enough to warrant comparison.

The greatest difference in the two sets of figures is due to differences in the two sets of lists used.
These differences in turn result from the fact that my Yokuts vocabularies were built up of terms selected mainly to insure unambiguity of English meaning between illiterate informants and myself, within a compact and uniform territorial area, but that Hoijer's vocabulary is based on Swadesh's second glottochronological list which aims at eliminating all items which might be culturally or geographically determined.
Swadesh in short was trying to develop a basic list that was universal;
I, one that was specifically adapted to the San Joaquin Valley.
The result is that I included 70 animal names, but Swadesh only 4;
and somewhat similarly for plants, 16 as against 4.
Swadesh, and therefore Hoijer, felt compelled to omit all terms denoting species or even genera (ox, vulture, salmon, yellow pine, manzanita);
their classes of animal and plant terms are restricted to generalizations or recurrent parts (fish, bird, tree, grass, horn, tail, bark, root).
The groups are therefore really non-comparable in content as well as in size.

Other classes are included only by myself (interrogatives, adverbs) or only by Swadesh and Hoijer (pronouns, demonstratives).

What we have left as reasonably comparable are four classes: (1) body parts and products, which with a proportionally nearly even representation (51 terms out of 253, 25 out of 100) come out with nearly even ratios;
2.6 and 2.7;
(2) Nature (29 terms against 17), ratios 3.3 versus 4.1;
(3) adjectives (16, 15 terms), ratios 3.9 versus 4.7;
(4) verbs (9, 22 terms), ratios 4.0 versus 3.4.

It will be seen that where the scope is similar, the Athabascan ratios come out somewhat higher (as indeed they ought to with a total ratio of 2.8 as against 3.5 or 4: 5) except for verbs, where alone the Athabascan ratio is lower.
This exception may be connected with Hoijer's use of a much higher percentage of verbs: 22% of his total list as against 3.5% in mine.
Or the exception may be due to a particular durability peculiar to the Athabascan verb.
More word class ratios determined in more languages will no doubt ultimately answer the question.

5.

If word classes differ in their resistance or liability to stem replacement within meaning slot, it is conceivable that individual meanings also differ with fair consistence trans-lingually.
Hoijer's Athabascan and my Yokuts share 71 identical meanings (with allowance for several near-synonyms like stomach-belly, big-large, long-far, many-much, die-dead, say-speak).
For Yokuts, I tabulated these 71 items in five columns, according as they were expressed by 1, 2, 3, 4, and more than 4 stems.
The totals for these five categories are not too uneven, namely 20, 15, 11, 16, 9 respectively.
For Athabascan, with a greater range of stems, the first two of five corresponding columns were identical, 1 and 2 stems;
the three others had to be spread somewhat, and are headed respectively Af;
Af;
and Af stems.
While the particular limits of these groupings may seem artificially arbitrary, they do fairly express a corresponding grouping of more variable material, and they eventuate also in five classes, along a similar scale, containing approximately equal numbers of cases, namely 19, 14, 15, 11, 12 in Athabascan.

When now we count the frequency of the 71 items in the two language families appearing in the same column or grade, or one column or grade apart, or two or three or four, we find these differences: Af

.
This distribution can be summarized by averaging the distance in grades apart: Af;
which, divided by Af gives a mean of 1.07 grades apart.
If the distribution of the 71 items were wholly concordant in the two families, the distance would of course be 0.
If it were wholly random and unrelated, it would be 2.0, assuming the five classes were equal in n, which approximately they are.
The actual mean of 1.07 being about halfway between 0 of complete correlation and 2.0 of no correlation, it is evident that there is a pretty fair degree of similarity in the behavior even of particular individual items of meaning as regards long-term stem displacement.

6.

In 1960, David D. Thomas published Basic Vocabulary In some Mon-Khmer Languages (AL 2, No. 3, pp. 7 - 11), which compares 8 Mon-Khmer languages with the I-E language data on which Swadesh based the revised retention rate (Af) in place of original (Af), and his revised 100 word basic glottochronological list in Towards Greater Accuracy (IJAL 21::
121 - 137).
Thomas' findings are, first, "that the individual items vary greatly and unpredictably in their persistence";
but, second, "that the semantic groups are surprisingly unvarying in their average persistence" (as between M-K and I-E).
His first conclusion, on behavior of individual items, is negative, whereas mine (on Ath. and Yok.) was partially positive.
His second conclusion, on semantic word classes, agrees with mine.
This second conclusion, independently arrived at by independent study of material from two pairs of language families as different and remote from one another as these four are, cannot be ignored.

Thomas also presents a simple equation for deriving an index of persistence, which weights not only the number of stems ('roots') per meaning, but their relative frequency.
Thus his persistence values for some stem frequencies per meaning are: stem identical in 8 languages, 100%;
stem frequencies 7 and 1, 86%;
stem frequencies 4 and 4, 64%;
stem frequencies 4, 3, and 1, 57%.
His formula will have to be weighed, may be altered or improved, and it should be tested on additional bodies of material.
But consideration of the frequency of stems per constant meaning seems to be established as having significance in comparative situations with diachronic and classificatory relevance;
and Gleason presumably is on the way with a further contribution in this area.

As to relative frequencies of competing roots (7 - 1 vs. 4 - 4, etc.), Thomas with his 'weighting' seems to be the first to have considered the significance this might have.
The problem needs further exploration.
I was at least conscious of the distinction in my full Yokuts presentation that awaits publication, in which, in listing 'Two-Stem Meanings', I set off by asterisks those forms in which N of stem B was Af of stem A/3, the unasterisked ones standing for Af;
or under 'Four Stems', I set off by asterisks cases where the combined N of stems Af was Af.

7.

These findings, and others which will in time be developed, will affect the method of glottochronological inquiry.
If adjectival meanings show relatively low retentiveness of stems, as I am confident will prove to be the case in most languages of the world, why should our basic lists include 15 per cent of these unstable forms, but only 8 per cent of animals and plants which replace much more slowly?
Had Hoijer substituted for his 15 adjectival slots 15 good animal and plant items, his rate of stem replacement would have been lower and the age of Athabascan language separation smaller.
And irrespective of the outcome in centuries elapsed since splitting, calculations obviously carry more concordant and comparable meaning if they deal with the most stable units than with variously unstable ones.

It is evident that Swadesh has not only had much experience with basic vocabulary in many languages but has acquired great tact and feeling for the expectable behavior of lexical items.
Why then this urge to include unstable items in his basic list?
It is the urge to obtain a list as free of geographical and cultural conditioning as possible.
And why that insistence?
It is the hope of attaining a list of items of universal occurrence.
But it is becoming increasingly evident that such a hope is a snare.
Not that such a list cannot be constructed;
but the nearer it comes to attaining universality, the less significant will it be linguistically.
Its terms will tend to be labile or vague, and they will fit actual languages more and more badly.

The practical operational problem of lexicostatistics is the establishment of a basic list of items of meaning against which the particular forms or terms of languages can be matched as the medium of comparison.
The most important quality of the meanings is that they should be as definable as possible.
In proportion as meanings are concrete, we can better rely on their being insulated and distinctive.
An elephant or a fox or a swan or a cocopalm or a banana possess in unusually high degree this quality of obvious, common-sense, indubitable identity, as do an eye or tooth or nail.
They isolate out easily, naturally, and unambiguously from the continuum of nature and existence;
and they should be given priority in the basic list as long as they continue to show these qualities.

With the universal list as his weapon, Swadesh has extended his march of conquest farther and farther into the past, eight, ten, twelve millennia back.
And he has proclaimed greater or less affiliation between all Western hemisphere languages.
Some of this may prove to be true, or even considerable of it, whether by genetic ramification or by diffusion and coalescence.
But the farther out he moves, the thinner will be his hold on conclusive evidence, and the larger the speculative component in his inferences.
He has traversed provinces and kingdoms, but he has not consolidated them behind him, nor does he control them.
He has announced results on Hokan, Penutian, Uto-Aztecan, and almost all other American families and phyla, and has diagrammed their degree of interrelation;
but he has not worked out by lexicostatistics one comprehensively complete classification of even a single family other than Salish.
That is his privilege.
The remote, cloudy, possible has values of its own -- values of scope, stimulus, potential, and imagination.
But there is also a firm aspect to lexicostatistics: the aspect of learning the internal organization of obvious natural genetic groups of languages as well as their more remote and elusive external links;
of classification first, with elapsed age merely a by-product;
of acquiring evidential knowledge of what happened in Athabascan, in Yokuts, in Uto-Aztecan in the last few thousand years as well as forecasting what more anciently may have happened between them.
This involves step-by-step progress, and such will have to be the day-by-day work of lexicostatistics as a growing body of scientific inquiry.
If of the founders of glottochronology Swadesh has escaped our steady plodding, and Lees has repudiated his own share in the founding, that is no reason why we should swerve.

8.

There is no apparent reason why we should feel bound by Swadesh's rules and procedure since his predilections and aims have grown so vast.
It seems time to consider a revision of operational procedures for lexicostatistic studies on a more humble, solid, and limited basis.

I would propose, first, an abandonment of attempts at a universal lexical list, as intrinsically unachievable, and operationally inadequate in proportion as it is achieved.

I would propose, next, as the prime requirement for constitution of new basic lists, items whose forms show as high an empirical retention rate as possible.
There would be no conceivable sense in going to the opposite extreme of selecting items whose forms are the most unstable.
An attempted middle course might lead to devices like a 5000-word alphabetized dictionary from which every fiftieth word was selected.