On the other hand, certain late texts show how fallible this criterion is. Anyone dating Maldon solely by "Lichtenheld's Test" would assuredly place it much earlier than 991.

It is easy to make a false use of grammatical statistics: and this test should only be applied with the greatest caution. But the difference between Beowulf and the works of Cynewulf is too striking to be overlooked. In Beowulf, to every five examples without the article (e.g. heaðo-stēapa helm) we have one with the article (e.g. se hearda helm): in Cynewulf to every five examples without the article we have forty with it.

A further test of antiquity is in the use of the weak adjective with the instrumental—a use which rapidly diminishes.

There are eighteen such instrumental phrases in Beowulf (3182 lines)[[235]]. In Exodus (589 lines) there are six examples[[236]]—proportionally more than in Beowulf. In Cynewulf's undoubted works (c. 2478 lines) there is one example only, beorhtan reorde[[237]].

This criterion of the absence of the definite article before the weak adjective is often referred to as Lichtenheld's Test (see article by him in Z.f.d.A. XVI, 325 etc.). It has been applied to the whole body of O.E. poetry by Barnouw (Textcritische Untersuchungen, 1902). The data collected by Barnouw are most valuable, but we must be cautious in the conclusions we draw, as is shown by Sarrazin (Eng. Stud. XXXVIII, 145 etc.), and Tupper (Pub. Mod. Lang. Assoc. XXVI, 274).

Exact enumeration of instances is difficult. For example, Lichtenheld gave 22 instances of definite article + weak adjective + noun in Beowulf[[238]]. But eight of these are not quite certain; se gōda mǣg Hygelāces may be not "the good kinsman of Hygelac," but "the good one—the kinsman of Hygelac," for there is the half line pause after gōda. These eight examples therefore should be deducted[[239]]. One instance, though practically certain, is the result of conjectural emendation[[240]]. Of the remaining thirteen[[241]] three are variations of the same phrase.

The statistics given above are those of Brandl (Sitzungsberichte d. k. Preuss. Akad. d. Wissenschaften, 1905, p. 719) which are based upon those of Barnouw.

"Morsbach's Test."

Sievers' theories as to O.E. metre have not been accepted by all scholars in their entirety. But the statistics which he

collected enable us to say, with absolute certainty, that some given types of verse were not acceptable to the ear of an Old English bard.

Sceptics may emphasize the fact that Old English texts are uncertain, that nearly all poems are extant in one MS only, that the MS in each case was written down long after the poems were composed, and that precise verbal accuracy is therefore not to be expected[[242]]. All the more remarkable then becomes the fact, for it is a fact, that there are certain types of line which never occur in Beowulf, and that there are other types which are exceedingly rare. Again, there are certain types of line which do occur in Beowulf as we have it, though they seem contrary to the principles of O.E. scansion. When we find that such lines consistently contain some word which had a different metrical value when our extant MS of Beowulf was transcribed, from that which it had at the earlier date when Beowulf was composed, and that the earlier value makes the line metrical, the conclusion is obvious. Beowulf must have been composed at a time or in a dialect when the earlier metrical values held good.