As Father B. A. Erdland's dictionary and grammar of the Marshall dialect, which were published in 1906 as the fourth volume of the Archiv für Deutsche Kolonialsprachen, are based on a far more extended acquaintance with the language than the author could hope to attain, it was decided to devote the available time to mechanical experiments, which do not appear to have been made for any Micronesian or in fact any Malayo-Polynesian language. The apparatus consisted of needles attached to rubber diaphragms at the end of tubes, inscribing on smoked paper covering a revolving drum. About four hundred tracings were secured simultaneously from mouth and throat, nose and throat or mouth and nose, or from the mouth alone. A selection from these is reproduced in the appended plates.

Father Erdland's orthography, while perhaps not above criticism on theoretical grounds, seems to be practically satisfactory, as he appears to have distinguished all the sounds of the language and to have represented them consistently. His choice of characters has therefore been followed here. That his description of the sounds is somewhat incomplete, certain of the most remarkable phenomena not being mentioned, is not surprising, as many of these traits are so unusual as to be possible of detection only by a practiced phonetician or through laboratory facilities.

VOWELS

Tracings of vowels do not lend themselves readily to analysis. A few aural impressions may therefore be in place. Erdland gives the vowels as i, e, a, o, u, with the additional umlaut qualities ä, ö, ü. Two mixed vowels of ö and ü type certainly occur; but they are considerably different from close and open ö and ü of German. The sound ä is of another character, being essentially an open e. O has two qualities, the more open of which is distinguished by Erdland as ô. Thus tô, rope, nearly like English taw. The other o is closer, but probably not so much so as in English or German "so." Its quality is probably between that of open and close o in most west European languages. The relation of ä and e seems to be parallel to that of ô and o. For i and u a similar distinction was not observed: each seemed to be, analogous to

o and e, midway between the close and open qualities of the same sound as spoken by Europeans.

All initial vowels were generally heard with a preceding aspiration. This is weaker than English h, and has not been recorded, or at least not written, by Erdland. As it is the normal approach of every initial vowel, its orthographic designation is perhaps unnecessary. The natives accept either hemen or emen as correct renderings of their word for four; probably 'emen comes nearest a true representation. In the tracings the aspiration usually shows as a slight rise preceding the voice vibrations of the vowel. That at times it does not thus appear is probably due to the weakness of the sound.

It will be seen that initial voiced consonants in Marshall normally begin with a flow of surd breath; that is to say, sonancy commences only some time after the sound is under way. The aspiration of initial vowels is evidently only part of the same tendency.

W, pronounced as in English, according to Erdland, did not appear in the words observed by the writer.

STOPS

Stopped consonants are formed in three positions, corresponding to p, t, and k. T is dental. K is more forward than in most American Indian languages, which on the whole form the sound farther to the rear than English, German, or French.

The character of the stopped consonants differs radically according to their position in the word, at least as much as has been noted in any American language.

In initial position it was found that sonancy regularly commenced a very short time, averaging probably less than a twenty-fifth of a second, after the beginning of the explosion. This is exactly the condition ascertained for most of the languages of California. In short, the Marshall initial stops are intermediate between true surds and true sonants, the occlusion and beginning of the explosion being unvoiced, the greater part of the explosion voiced. After its first inception, the voicing increases very rapidly, reaching a maximum, which surpasses the strength of the voicing of the fol

lowing vowel, within a very few hundredths of a second. In most American languages, the commencement of the vibrations of the vocal chords in intermediate stops presents a different character in tracings: the needle connected with the throat drops a short distance, indicating a slight retraction of the surface of the larynx, and at the same time the vibrations begin to appear, increasing in strength more gradually than in the Marshall records. Otherwise the sounds seem identical in the two groups of languages.

As regards strength of explosion, the Marshall initial stops resemble European surd stops, though they lack any sustained aspiration or rush of breath.

Erdland, while denying a p, writes both t and d, k and g. After a little familiarity with native words, these sounds are usually recognizable as distinct, though they are far less different than in English. An examination of tracings of words commencing with t and k as compared with d and g however fails to show any difference between the two classes as regards either duration or strength of sonancy, force of explosion, or continuance of aspiration. But, strange to say, it was discovered that in words written by Erdland with an initial surd stop, the occlusion is normally somewhat longer than in those of which the first sound is represented by a character for a sonant stop. This difference is clearest for the dentals. In "d" the occlusion is almost always less than a fifth of a second; in "t" it lasts from a fourth to a third of a second. For "g" and "k" the observed difference is not so pronounced, but the average duration of the occlusion in the former sound runs somewhat below and of the latter above a fifth of a second, corresponding to tracing lengths of respectively less and more than a centimeter. It is remarkable that two sounds that appear to be identical except for this difference in duration of occlusion should to the European ear give somewhat the effect of sonant and surd stops.

In medial position, at least between vowels, the sounds written as stops by Erdland were usually heard and always observed as fricatives, and generally voiced. The tracings present a marked dip, though without the definite vertical and horizontal contour of stops. The voice vibrations usually show through the whole period of the sound both in the mouth and the throat record, though

in some instances enough of the original stop character, with its surd occlusion, remains for at least part of the fricative to be unvoiced.

A similar tendency toward the substitution of fricatives for stops was at times noted in initial position, though less regularly than medially, and usually confined to the explosion. That is to say, such modified initial sounds usually show at least some remains of an occlusion, but the curve for the explosion is that of a fricative instead of a stop. They are probably not mere affricatives, as tracings of these normally present the character of stops. This fricative quality of initial stops was sometimes also perceived by the ear; and, in their imperfect rendering of English, the natives employ initial stops and fricatives almost indiscriminately. Thus "flenty" for plenty, but "pish" for fish; "thongue" for tongue; and either "shome" or "djome" for some.

Final stops possess a firm occlusion, but no perceptible explosion from the mouth. The lips or tongue make a closure which is either indefinitely sustained or concluded after the breath pressure necessary for speech has ceased. Hence final stops are very faint to the ear. Bād, smoke, is not quite bāt and almost bā; "Gilbert" is either kilua or kiluat. In some cases a distinct explosion through the nose was observed, and a tube from the nostril to a flame usually showed a puff when the explosion would be expected in an English stop. Many tracings from the nose however reveal no explosion whatever; so that it is possible that the nasal explosion is due merely to an occasional unintended opening of the naso-oral passage before the release of the mouth closure and cessation of breath pressure. If the sounds are essentially nasal stops, they differ from m, n, and ñ in that they possess an occlusion which is complete for nose as well as mouth. They certainly do not markedly resemble final nasal continuants, although these, like final stops, are entirely surd in the Marshall dialect.

Erdland writes both d and t, but only b and k, at the end of words. As in the case of initial stops, these show no difference in point of sonancy, which, as just stated, is completely lacking. It seems that there may be a longer occlusion for t and k than for d and b, just as when they are initial; but this is not certain, as

the end of the occlusion is discernible in but very few tracings. Stops closing syllables that precede a consonant are treated as final. In bödökdök, blood, the first k is of final quality, the following d initial, and only the first d of medial character. Erdland says that in such words the final surd stop of the first syllable often becomes sonant-and presumably fricative and is followed by a vowel. In conformity to this statement, nugenuk and bödögödök were sometimes heard instead of nuknuk and bödökdök; but the majority of tracings do not show the change.

NASALS

The nasal continuants, m, n, ñ, correspond to b, d, g in position of articulation. They also resemble the stops in being at least partly voiced when initial, formed with imperfect mouth closure when intervocalic, and entirely surd when final. Initial sonancy commences gradually and some cases were observed where it seemed to begin after mouth closure, so that the first part of the sound would be surd. Medial nasals are heavily voiced, and, as the vibrations show in mouth tracings, it appears that the mouth is closed imperfectly. The metal mouthpiece employed has been found by the writer to interfere at times with tight lip closure after a wide vowel, so that an effort at clear enunciation is often necessary to produce a true closure effect for medial m. In the Marshall tracings, n, to which this defect of the apparatus would not apply, is, however, recorded as voiced from the mouth as well as m, and medial m shows much heavier vibrations than those occasionally traced in English words: so that it is clear that the typical intervocalic nasals in Marshall are made with very loose contact of the mouth parts. The surdness of the final nasals is usually complete; sometimes the voice of the preceding vowel continues to die away in the first part of the succeeding nasal. This is a phenomenon that has been observed in several American languages.

It is obvious that the difference shown by stops and nasals, according as they are initial, medial, or final in Marshall, does not point to so many organically distinct classes of sounds, but to a single sound of each type which becomes radically modified according to its position in the word.