You are on page 1of 59

Regularizing the Canaanite Shift

Masters Thesis for Research MA Linguistics: Structure and Variation in the Languages of the World Leiden University, spring semester 2012 Supervisor: Prof. Dr. H. Gzella Benjamin D. Suchard, student #0737178

Cover image: Amarna letter EA 161.

Table of Contents 0. Introduction 0.1 How reliable is the Masoretic vocalization? 0.2 Transcription 0.3 Methodology 1. Previous formulations of the Canaanite Shift 1.1 Only stressed * > * 1.2 All * > * 1.3 Conclusion 2. Relevant material 2.1 Biblical Hebrew 2.1.1 * > 2.1.1.1 Nominals 2.1.1.2 Verbs 2.1.1.3 Other 2.1.2 * > 2.1.2.1 Nominals 2.1.2.2 Verbs 2.1.2.3 Other 2.1.3 *a$> $ 2.1.3.1 Nominals 2.1.3.2 Verbs 2.1.3.3 Other 2.1.4 *a$ > $ 2.1.4.1 Nominals 2.1.4.2 Verbs 2.1.4.3 Other 2.2 Phoenician 2.2.1 * > /o/ 2.2.2 * > /u/ 2.2.3 *a$ > /o$/ (or /$/) 7 7 9 11 12 12 13 13 14 14 14 14 17 18 19 19 20 20 20 21 21 21 21 21 21 22 22 22 22 23

2.3 Amarna Canaanite 2.3.1 * > // 2.3.2 *a$ > // 2.4 Ammonite, Moabite and Edomite 2.5 Preliminary evaluation of the evidence 3. Reconstruction of the relevant material 3.1 Nominals 3.2 Verbs 3.3 Other 4. Analogy and contamination 4.1 Analogy 4.2 Contamination 5. Phonetics of the Canaanite Shift 5.1 Stress 5.2 Phonetic environment 6. * > in other languages 6.1 Germanic 6.2 Welsh 6.3 Rural Maltese 6.4 Summary 7. Conclusions and discussion 7.1 *a$ > * 7.2 Conditions of the Canaanite Shift 7.3 Dating the Canaanite Shift 7.4 Canaanite, Northwest Semitic and the family tree model 7.5 Topics for future research Acknowledgements References

23 23 23 23 24 24 24 35 38 40 40 43 43 44 44 45 45 45 46 46 47 47 47 48 50 52 54 54

0. Introduction In the history of linguistics, few concepts have had such a revolutionary effect on the discipline as that of the sound law, introduced by the nineteenth-century scholars known as the Junggrammatiker or Neogrammarians. In brief, the idea is that sound change is both purely phonetically motivated and regular. To quote their own words:
When we speak of systematic effect of sound laws we can only mean that given the same sound change within the same dialect every individual case in which the same phonetic conditions are present will be handled the same. Therefore either wherever earlier the same sound stood, also in the later stages the same sound is found or, where a split into different sounds has taken place, then a specific cause a cause of a purely phonetic nature like the effects of surrounding sounds, accent, syllabic position, etc. should be provided to account for why in the one case this sound, in the other that one has come into being. (Paul 1880:69, cited in Joseph-Janda 2003:343)

The Neogrammarians reached this idea based on the study of modern European languages, a relatively new pursuit (Jankowsky 1972). In contrast, Biblical Hebrew, the language of the Hebrew Bible, had been studied by European scholars for many centuries by then. But whereas the sound laws leading to other classical languages such as Greek, Latin, Sanskrit, or any number of relatives from reconstructed Proto-Indo-European have been known for the better part of two centuries, the sound laws for Biblical Hebrew (BH) have not yet been formulated in a way that accurately describes all the available data. In this paper, I will attempt to refine the description of one of the most important sound changes that have influenced BH: the shift from earlier * to *. As this development is also attested in Phoenician (Friedrich-Rllig 1999, 79), Amarna Canaanite (Sivan 1984:29ff.) and not contradicted by the evidence from the other Canaanite languages, this sound change is commonly known as the Canaanite Shift (CS). In the remainder of this introduction, I will discuss some necessary preliminaries about the status of the data that is to be used, the transcription of BH I will employ and the overall methodology of this research. 0.1 How reliable is the Masoretic vocalization? As our primary data will be the text of the Hebrew Bible vocalized according to the Masoretic tradition, ascertaining its reliability is an important first step. By Masoretic, we mean the pronunciation tradition (Hebrew: msr) that was followed in 8th-century CE Tiberias, modern-day Israel (EJ 13:613) hence the alternative appellation, Tiberian. It is around this time and place that the so-called Masoretes started recording the traditions surrounding the reading from the Hebrew Bible, such as the vowels and punctuation, which were not previously written down. This Tiberian or Masoretic vocalization is currently used by all Jewish communities in the world, although most (perhaps all) communities use pronunciations that are not derived from the Tiberian tradition (EJ 16:549). As it is the most detailed of three roughly contemporary vocalization traditions (EJ 13:617), the Masoretic Text (i.e. the text of the Hebrew Bible with Tiberian vocalization and punctuation) is the form in which the Hebrew Bible is most commonly taught and studied. The oldest extant manuscript that originally covered the entire Masoretic Text, the Aleppo codex, dates from the early 10th century CE (Goshen-Gottstein 1979). Hebrew 7

originally died out as a spoken language in the second century CE (Beyer 2004 I:49), and the oldest parts of the Bible are sometimes dated to the late second millennium BCE (Cross-Freedman 1975). Are we to assume that the correct vocalization to the consonantal text was transmitted faithfully for almost two thousand years? And if not, how can we be sure that the Masoretic Text bears any relation to the once-living language that underlies Biblical Hebrew? Ullendorff (1977) rightly notes that the vocalization of the Masoretic Text is remarkably uniform for a corpus that is generally assumed to span the greater part of a millennium. Like other scholars, he suggests that Biblical Hebrew as we know it is an amalgam, Tiberian Hebrew vowels inserted into a mostly unchanged consonantal text. Others (such as Goshen-Gottstein 1979) emphasize the care the Masoretes took to ensure the correct pronunciation of Scripture, arguing that straying from the original vocalization is not compatible with this reverent attitude. I will presently consider two extreme viewpoints on the matter. The Masoretic vocalization bears no relation to the original vocalization of Biblical Hebrew. One could imagine that the Masoretes either had no knowledge of the correct vocalization of BH, or that they did not find it important enough to preserve it. Perhaps they developed their own system to vocalize the consonantal text of the Hebrew Bible, or maybe they based it on their everyday language, Aramaic. This scenario seems unlikely to me, simply because of the uniquely Hebrew character of the Masoretic vocalization. Very many grammatical forms are vocalized in a way that is not exactly the same as in any other Semitic language (i.e. not copied from one), and yet is clearly related to it (i.e. not random). For instance, the 3rd person singular masculine of the sound qal perfect is vocalized as ktab. This is different from Aramaic ktab or Arabic kataba, to name two languages the Masoretes could have been inspired by; yet although the consonantal skeleton {ktb} could have been vocalized in any number of different ways, we find one that is not at all surprising from a comparative point of view. The Masoretes were therefore not just making things up: their vocalization does bear some relation to BH as it was spoken. The Masoretic vocalization is 100% faithful to the original vocalization of Biblical Hebrew. This point of view raises some a priori questions and difficulties. To which original vocalization should the Masoretic one correspond? As stated above, the Hebrew Bible is a diverse corpus, that is generally considered to incorporate texts that vary greatly in age, and yet the Masoretic Text shows a fairly uniform vocalization for texts from different periods. Surely we are not to assume that BH underwent no changes in its vowels whatsoever during the entire period from which Biblical texts are assumed to date. Besides this, there are some indications in the text itself that in some cases, the Masoretes pronounced words differently than the author(s) probably did. For instance, the second person feminine singular personal pronoun, normally at < *atti, is sometimes spelled <ty> in the consonantal text, as in Jd 17:2. This seems to indicate that atti was still the current pronunciation at the time the text was written. In these cases, however, the Masoretes indicated that the form is to be pronounced at, disregarding the final <y>. From this and other cases we can infer that some amount of leveling took place in the Masoretic pronunciation of BH, which was therefore not a fully accurate reflection of the original vocalization.

In conclusion, it seems that the Masoretic vocalization certainly preserves features of the original vowels of BH, although it is not completely faithful to the way the text was pronounced at the time of composition. As vowels were not written in Hebrew prior to the Masoretes, texts were probably vocalized and read according to the current state of the language while Hebrew was still spoken. It seems plausible that the Masoretic tradition is based on how these texts were read just before Hebrew died out as a spoken language. This moderate stance on the reliability of the Masoretic Text is corroborated by comparison with other vocalization traditions, such as the comparison with the 3rd century CE or earlier Second Column (Secunda) of Origenes Hexapla conducted by Brnno (1943). In his conclusion, Brnno states:
Die groe Bedeutung der SEC[unda] fr die hebrische Sprachwissenshaft liegt u. a. darin, da diese alte berlieferung deutlich zeigt, da das tiberische Formensystem in seinen wesentlichten Hauptzgen eine alte Tradition hinter sich hat, wenn auch die SEC an der einzelnen Stelle keineswegs immer die Form aufweist, die morphologisch der Form des MT an der betreffenden Stelle entspricht (...). Hinsichtlich der Quantitt stimmt die SEC durchgehends genau mit der MT berein (dagegen nicht mit dem Quantittssystem der spteren jdischen Grammatiker, die bekanntlich berall das Sere und das Cholem fr lang hielten). Die SEC zeugt deutlich davon, da die tib. berlieferung rcksichtlich des Aufbaus des Formensystems viel zuverlssiger ist, als einige Forscher nach der Entdeckung der bab. und der pal. Vokalisationssysteme zu glauben geneigt waren. Da zwischen der Entstehungszeit des Textes der SEC und der des MT eine Entwicklung stattgefunden hat, ist ja kein Wunder. (p. 463, emphasis his.)

The most important structural differences between the Secunda and the Masoretic Text that Brnno (1943) identifies are the non-occurrence of segolisation and Philippis Law and a much more limited distribution of the law of attenuation, none of which are directly relevant to the present paper. So it seems that we are dealing with two sets of vowel changes between the way any given Bible text was originally meant to be read and the way it is read based on the Masoretic Text: the changes from the time of composition to the demise of spoken Hebrew and the changes that took place in the reading tradition itself. This is something to keep in mind during the investigation of the Canaanite Shift. 0.2 Transcription (Blau 2010:76 ff. for pronunciation) To make this paper more accessible to non-Hebraists, I will present the Hebrew material in Latin transcription. I would like to draw special attention to two features of the transcription I will be using, both related to the vowels. Firstly, I will follow the convention of transcribing some BH vowels with a macron (, , ). This macron does not mark vowel length; rather, it marks a difference in quality with the macron-less vowels, which are pronounced with less closure. This should be contrasted with vowels bearing macrons in reconstructions of Proto-(Northwest)Semitic and transcriptions of Arabic, where the macrons do mark length. Conversely, although i and u are always transcribed without a macron, they are only historically short when followed by two consonants; in all other cases, they derive from long * and *.

Secondly, I will also follow the convention of transcribing so-called qmes gdl as <>, even though it is quite clear that the Tiberian tradition pronounced all instances of qmes as []. Considering that this sound usually comes from PS *a or *, however, and that the Palestinian tradition (reflected in the Sefardi pronunciation; EJ 16:549ff.) pronounced it as [a], it seems safe to reconstruct this sound as * for living BH; and since this paper is mainly concerned with the appearance of where we expect , I fear transcribing qmes gdl as {o} might cause some confusion. The reader is invited to pronounce any instances of BH as [], should he or she be concerned with an accurate approximation of the Masoretic pronunciation. 0.2.1 Vowels Symbol Name patah hatep patah ** *** sgl hatep sgl sr hireq qmes

Pronunciation (IPA) Transcription [a] a, a* a [a] e [] [] [e] [i] []


e

i o
o

qmes htup [] hatep qmes [] hlem ureq qibbus w

[o] [u] u [u] u (none) [a], [] (none)**** *a is so-called patah furtivum, a vowel inserted between vowels other than a and and the guttural consonants h, h and when these are word-final. ** In stressed or open syllables; mainly derives from *a and sometimes *. *** In unstressed, closed syllables; derives from *u. **** Indicates the lack of a phonemic vowel; pronounced as [a] after word-initial consonants, two consonants, geminated consonants and , , plus a consonant; not pronounced elsewhere. 0.2.2 Consonants Letter Name lep bt gimel dlet h ww

Pronunciation (IPA) [] [b], [v]* [g], []* [d], []* [h] [w]

Transcription b, b g, g d, d h w

10

*** *** *** *** ***

zayin ht t t yd kp lmed mm nun smek ayin p sd qp r in

[z] [] [t]** [j] [k], [x]* [l] [m] [n] [s] [] [p], [f]* [s]** [q]** [r]

z h t y k, k l m n s p, p s q r

[] in [s] tw t, t [t], []* * Fricative after vowels (except when geminated), word-initial consonants, two consonants and geminated consonants; plosive elsewhere. ** So-called emphatics; these are usually assumed to have been pronounced with some glottalic element (ejectivity or pharyngealization; Blau 2010:68). From around 250 BCE onwards, the distinction was probably one of aspiration (non-emphatic plosives) versus non-aspiration (emphatic plosives) (Beyer 2004:I.125). *** Word-finally. I will indicate gemination by writing the consonant twice, e.g. dayyn. Consonants that are not pronounced (matres lectionis and cases such as the <y> of <ty> at above) will not be transcribed. 0.3 Methodology Before starting my own research proper, I will examine and evaluate various previous formulations of the Canaanite Shift (1). I will then present the relevant material, i.e. primarily BH words that are generally taken to have had an *-vowel prior to the operation of the Canaanite Shift (2.1), whether this vowel shifted to or appears as some other Masoretic vowel. I will also discuss the material from Phoenician (2.2), the Amarna letters (2.3) and Ammonite, Moabite and Edomite (2.4), as the Canaanite Shift is supposed to have affected all Canaanite languages (by definition). Having presented the examples of and exceptions to the Canaanite Shift, I will reconsider their reconstruction to ensure which of these words and vowel patterns should actually have been targeted by the sound change (3). Furthermore, I will consider the role of analogy in the spread or lack thereof of the CS between different vowel patterns (4). I will then discuss possible phonetic motivations for the CS (5) and similar developments in other languages (6), and finally, I will formulate what I deem to be the most accurate description of the CS in the conclusion (7).

11

1. Previous formulations of the Canaanite Shift There is some disagreement about how stress affected the operation of the Canaanite Shift. The majority opinion is that only stressed * was affected, but there is a significant minority that holds that all instances of * underwent the change. I shall discuss several suggested formulations from both camps. 1.1 Only stressed * > * 1.1.1 Brockelmann Brockelmann (1908) was an early advocate for the version of the CS which only affected stressed vowels. He states that all originally stressed instances of * shifted to *, while all unstressed instances of * stayed the same. This change must have taken place at a time when Hebrew stress still resembled that of Classical Arabic, i.e. stress is on the last long or closed syllable, excluding the ultimate. Hence, although the BH qal active participle, qtl, is stressed on the ultimate syllable, it would have been stressed on the first syllable when the change *qtilu > *qtilu occurred. This stress-based conditioning would have led to a lot of alternation between * and * within paradigms, which might still be preserved in rare cases such as r head ~ rim heads. In most other cases, paradigmatic leveling would either have eliminated the remaining forms with *, giving rise to < unstressed * as in the plural qal active participle, qtlim, or have restored the original * in stressed syllables, leading to the occurrence of apparent exceptions to the CS like dayyn judge, qinyn property. As word-final vowels (in polysyllabic words) were always unstressed, they did not shift to , cf. for instance att; the vowel must probably have been long, as short word-final vowels dropped off (but see 3.3.1). Another category of exceptions is provided by syllables that did not carry full stress, such as verbs and particles. A recent endorsement of this formulation of the CS can be found in Blau (2010:136). One of the problems of this approach is that it cannot explain why some paradigms generalized the , while others generalized the . 1.1.2 Bergstrsser Additional categories of exceptions are adduced by Bergstrsser (1918). Again, the CS is held to have affected only stressed instances of *, but Bergstrsser notes that there are cases where it does not occur after back vowels, as in ulhn table and qorbn offering. Words such as qrb combat are, in my opinion rightly, explained as loanwords from Aramaic, which is supported by the absence of a vowel after the first consonant, a rare phonotactic pattern in Hebrew which is normal in Aramaic. Bergstrsser admits that forms such as qm he stood remain unexplained. 1.1.3 Bauer-Leander Bauer-Leander (1922) uses the same formulation of the CS as Brockelmann (1908) and Bergstrsser (1918), but has a radically different explanation for the attested exceptions. The framework that underlies this entire work is the idea that BH is a mixed language, based on an old, natively Canaanite stratum, and a young, paradoxically more archaic stratum that was imported by the invading Israelites during their conquest of Canaan. The exceptions to the CS would then be taken from the young language of the Israelites, which had not undergone this sound change. Bauer and Leander also agree with

12

Brockelmann (1908) that words that did not carry phrasal stress were unaffected by the CS. There is no further evidence to support this idea of Hebrew as a mixed language and therefore, it cannot serve as a satisfactory explanation. The notion is generally rejected, although Kutscher (1982) still considers it a possibility.

1.2 All * > * 1.2.1 Birkeland Contrary to the authors we have seen above, Birkeland (1940) claims that all instances of * underwent the CS. All apparent counterexamples did not yet have an * vowel at the time the CS was operative. Birkeland is very concise about this, simply stating that mli left (adj.), which is adduced by proponents of a stress-conditioned sound change due to its contrast with ml left hand (side), must have been a late introduction to Hebrew, without explaining any further. Birkeland does not offer a good explanation for the various exceptions to the Canaanite Shift, so the expansion of the CS to all instances of * seems a bit hasty. 1.2.2 Dolgopolsky A slightly more thorough defense of this version of the CS is put up by Dolgopolsky (1999). The two most-cited examples of non-occurrence of the CS in formerly unstressed syllables, rim heads and mli left (adj.), are derived from different reconstructed forms that usual, i.e. *raama and *imaaliyu1, respectively. The that is reflected in the Tiberian form is then a reflex of the *a following the *, not the one preceding it, which Dolgopolsky holds to have dropped out in this antepretonic position. The in r head and ml left hand (side) is the result of a different pattern of vowel reduction, stemming from different stress; Dolgopolsky reconstructs these words as *raau and *imaalu2, respectively; in these words, the *a after the * elided, as it was the first of two unstressed short vowels. To explain the presence of the glottal stop in Samaritan Hebrew forms like ro head, Dolgopolsky introduces a new sound law, similar to the CS, which states that the sequence *a became a long, glottalized * in closed syllables, which then shifted to * while preserving its glottalization. This glottalization must then have been lost in later Tiberian Hebrew, while it reintroduced the in the Samaritan Hebrew forms of these words. Dolgopolsky explains the in forms like qm he stood by stating that it derives from an older triphthong *awa, which only contracted to * after the CS had ceased operating. A major weakness of Dolgopolskys explanation is that he does not even mention nominal forms which show for expected **. 1.3 Conclusion All of the previously suggested formulations of the Canaanite Shift that we have just seen suffer from inadequacies. Of the versions that posit a sound change that was conditioned
1 2

Or *umaaliyu. Or *umaalu.

13

by stress, Bergstrsser (1918)s seems to fit the evidence best, although it lacks a good explanation for nominal forms with that are not obviously Aramaic loanwords. Dolgopolsky (1999) offers an interesting alternative, which highlights the fact that the strongest examples of presumably unstressed * > are actually cases of unstressed *a > , a problematic category in its own right (see 3.1.17). Here too, however, forms like dayyn judge and qinyn property remain unexplained. 2. Relevant material In this section, I will give the reflexes in various languages of all attested words that are commonly reconstructed with Proto-Canaanite * or *a. I will exclude proper nouns (names, place names) as they are often more resistant to change and more susceptible to borrowing. 2.1 Biblical Hebrew (mainly from Bauer-Leander 1922) Please note that I am not sure that I have covered all instances of productive categories (2.1.1.1i, 2.1.1.1j, 2.1.1.1k, 2.1.1.1n, 2.1.1.1p, 2.1.2.1f). I am fairly certain that all the exceptions to the regular developments of these categories are listed. I list all words mentioned by Bauer-Leander (1922) for a given category; those that can confidently be derived from a form with * (or *a, below) are printed bold, those that cannot, are not. See 3 for the motivation behind this decision for each particular word. 2.1.1 * > 2.1.1.1 Nominals 2.1.1.1a *qlu, *qltu and *qlatu ln tongue r light rd wild ass dd dear, fathers brother lm wellbeing hah thorn, hook l three hl sand hp shore qtl, the qal absolute infinitive tb good ks cup mne eight ql voice 2.1.1.1c *qitlu and *qitlatu tb good (noun) elah god qet watering-trough zr loincloth pd ephod (cultic garment) 2.1.1.1b *qatlu, *qatliyu hadm footstool of (construct state) dn lord zra arm ht sister3 hagr belt *arn ark (of the covenant) (attested hamr donkey with article or as construct state) abt rope t sign srr purse r leaven tn she-ass ml left hand side4 grn throat
3

Strictly speaking not *qatlu but *qaltu from a biradical root H, as in h brother, with a feminine suffix.

Written <mwl> and <mwl>, i.e. m()l. Bauer-Leander (1922) suggest a confusion

14

rk sandal strap of (construct state) thm ocean h gr belt abd work, service 2.1.1.1d *qutlu, *qutlatu, *qutltu, *qutlma en human *bly rag (attested as bl, construct state plural) br cypress *ptt morsel (attested as plural) rhb open place lbn incense nhet copper, bronze nret tow thrim hemorrhoids 2.1.1.1e *qtalu lm eternity 2.1.1.1f *qtilu, *qtilatu, *qtiltu yb enemy bqr herdsman ybl ram khn priest krm vinedresser rb raven qr (type of bird) r porter qtl, the qal active participle (m.) hm city wall l burnt offering qtl, the qal active participle (f.) kteret capital (of pillar)
a

qtelet, the qal active participle (f.) 2.1.1.1g *qattlu, *qittlu, *qattltu gibbr hero yissr reprover qann jealous qippz arrow-snake rattq chain ikkr drunk kinnr zither kappret cover (of the Ark of the Covenant) prket temple curtain 2.1.1.1h *quttlu sippren nail (of finger or toe) rimmn pomegranate 2.1.1.1i *maqlu, *maqlatu, *maqlma, *miqlu mb entrance mgr fear, exile mdn strife mhl dance mln place to spend the night mnah rest mns refuge mn dwelling msr stress, siege mqm place mqr spring mrm height *mgr horror (attested in construct state and plural with suffixes) mdk mortar mhl round dance mkn rack mnr candelabra mn dwelling mmt death of (construct state)

between cognates of Arabic amal north wind and iml left hand (side); see 3.1.3.

mrs race

15

2.1.1.1j *maqtlu, *miqtlu (not easily distinguishable), *maqtltu, *miqtltu, *maqtlma mibhr exquisite mizmr Psalm matmn treasure *makb pain (attested with suffixes and in the plural) mikll perfection *makmr net (attested as plural with suffixes) mikl stumbling-block malqah spoil, jaw malq spring rain miqsa corner mar saw *mit oar (attested as plural with suffix) milah sending miql weight maqp upper threshold mir level mahagret girdle of (construct state) miqlet plumb line of (construct state) mammrrim bitter things 2.1.1.1k *qatalnu, > *qattalnu, *qatalnima, maqtalnu abaddn perdition hmn commotion, multitude zdn insolence hzn sight halln window harbn dry heat hrn burning, anger lsn mockery wn sin przn dwellers of open country spn north rzn emaciation rbn hunger n joy

bitthn trust drn disgust gillyn tablet, hand mirror higgyn muttering hryn conception zikkrn memory hizzayn sight hippzn hurried flight yrqn paleness, mildew niqqyn purity iwwrn blindness issbn complaint rbn pledge irn tenth part pidyn5 ransom of (construct state) piqqdn care pitrn5 interpretation of (construct state) simmn thirsty ground rayn5 pursuit of (construct state) ripyn5 slackness of (construct state) ibbrn break abbtn sabbatical iggyn dirge iddpn scorching (of grain) immmn horror znunim fornication < *znnim man deception 2.1.1.1l *qitlnu, qutlnu (not easily distinguishable) hesrn lack hebn account yitrn profit kirn skill 2.1.1.1m Derived adjectives in *-nu ebyn needy agmn reed aharn last ln big tree of (construct state) alln big tree aqalltn tortuous
5

Might belong to 2.1.1.1l instead.

16

*gaayn proud (attested as plural) zdnim running high (of water; plural) hisn outer *izbn merchandise (attested as plural with suffixes) elyn highest paamn bell (on high priests robe) *qadmn eastern (attested as feminine) *qisn last (attested as feminine) rin first pipn horned snake tikn middle 2.1.1.1n Diminutives in *-nu in pupil aharnim lunulae (type of collar) *sawwrnim necklace (attested with 2f.sg suffix attached)

2.1.1.1o Derived adjectives in *-niyyu admni red yiddni soothsayer, familiar spirit sipni viper qadmni eastern 2.1.1.1p The feminine plural ending *tu -t 2.1.1.1q Miscellaneous ekl grapes < *itklu *nahall watering-place (attested as plural) < *nahllu pqah-qah opening (of eyes) < *paqahqhu qlql contemptible food < *qalqilu qiqln disgrace < *qilqalnu

2.1.1.2 Verbs 2.1.1.2a All types of verbs We have already seen the most common cases of * > that are linked to the verbal system: the qal active participle (2.1.1.1f) and absolute infinitive (2.1.1.1b). The other cases are rare forms of stems that might be cognates of the Arabic fala stem (stem III), although Gzella (2010) rightly notes the lack of semantic correspondence. Based on their vocalism, these verbs are said to be in the pl or the pal, but I agree with Gzella (2010:76) that the one example of the pal is probably just the pausal form of the pl, as it has a no more passive meaning than its corresponding pl form has. The forms, all of which are only attested once, are: Pl ysr is blown mloni6 my slanderer (active participle) mpti my judge (active participle) r has taken root ru they took root (pausal form. Formally, this verb could also be a pual, but in the context, it cannot be the passive counterpart of the pil r to uproot, and as it means the same as r, positing a separate pal category is unnecessary) ti I plundered (from the weak root -s-y to plunder, spelled with <> for **<s>) There are also two cases of the reflexive counterpart to the pl, the hitpl, both from the same verb and occurring in Jeremiah. The forms are hitgau shake back and forth and yitgau they will shake back and forth.
6

Thus qr; ktib <mlwny> suggests *mlni.

17

2.1.1.2b Verbs with a weak middle radical (hollow roots) In the qal, the Canaanite Shift may have occurred in verbs that are reconstructed with a * as root vowel, i.e. perfect *qla, imperfect *yaqlu. The only non-derived verb that is often considered to be in this category (as in Bauer-Leander 1922) is b (<< *b, still attested in imperfect yb etc.) to come; additionally, there are the denominative verbs b to be ashamed, tb to be good and r to be light. It also seems to have affected the nipal perfect of hollow roots after the contraction of the triphthong that resulted from the presence of their weak middle radical, e.g. nkn it is established (< *nakna?) < *nakawana. Bauer-Leander (1922) also state that the in the imperfect, like yikkn it will be established, originated in verbs with * as their root vowel and spread from there. In the nipal and hipil perfect, an extra -- is inserted between the last radical and endings that begin with a consonant, e.g. haqimti I have erected (this also occurs in verbs with a geminate second radical). If this -- is the reflex of the long * that is attested in Akkadian stative forms like parsku I am x-ed, as is suggested by JoonMuraoka (2009:199), this too is an instance of the Canaanite Shift. Instead of the pil, most verbs with a weak middle radical form a stem called the 7 pll , which also occurs with roots with a geminate second radical; see 2.1.1.2c below. Attested forms of hollow roots that show an where the paradigm regularly has : bsim treading down (qal active participle) qmim standing (qal active participle) 2.1.1.2c Verbs with a geminate second radical As was mentioned in 2.1.1.2b, these verbs usually do not follow the pattern of the strong pil; instead, they have forms like perfect qll, imperfect yqll. This stem is known as the pll and seems to have the same range of meanings as the pil. Bauer-Leander (1922) derive it from the pil: *yuqallilu > *yuqlilu, with simplification of the geminate as a form of dissimilation and compensatory lengthening of the preceding *a accompanying it. The next step, *yuqlilu > *yuqlilu, shows the working of the Canaanite Shift. This stem was analogically extended to verbs with a weak middle radical. Verbs with a geminate second radical insert an -- before endings that start with a consonant in the qal, nipal and hipil perfect, much like verbs with a weak middle radical (2.1.1.2c above). 2.1.1.3 Other The first person singular personal pronoun nki < *anku seems to have undergone the Canaanite Shift, as may k thus and z this have done.

And, accordingly, a hitpll instead of the hitpal.

18

2.1.2 * > 2.1.2.1 Nominals 2.1.2.1a *qlu, *qlma *y or *ye (<*yiyu) shovel (attested as plural yim) ss moth rim cities 2.1.2.1b *qawilu n raw b cloud ql, the qal active participle of hollow roots 2.1.2.1c *qatlu8 nn cloud 2.1.2.1d *qitlu, *qitliyyu issr oath of abstinence, esrh with suffix nissb handle (of knife) mli left-hand (adjective, <mly>; see note 3) written

*hatt sinful (attested as feminine and plural) hall weakling hr artisan tabbh butcher, cook, bodyguard *keh lying (adj.) (attested as plural) naggh addicted to goring sabbl burden-bearer sallh forgiving awwl evildoer pr rider sawwr10 neck *sayyd hunter (attested as plural) qann jealous *qab attentive (attested as feminine) rakkb charioteer, rider *raqqh ointment-mixer (attested as plural) behl sudden terror baqqrat care of (construct state) baqq desire *nes contumely (attested as plural) *nehm comfort (attested with suffix) 2.1.2.1g *qitlnu, qutlnu binyn building kibn kiln inyn task qinyn posession qorbn sacrifice ulhn table 2.1.2.1h Derived adjectives in *-nu (used substantively) hammn (possibly) incense-stand liwytn Leviathan, serpent nhutn serpent idol tn adversary, Satan 2.1.2.1g Miscellaneous tb resident alien < *tawtbu

2.1.2.1e *qtilatu *dlit foliage (attested as plural with suffix dliyytw) *zwit corner (attested as plural zwiyyt) 2.1.2.1f *qattlu, *qattlatu gannb thief *dawwg fisherman (attested as plural)9 dawwy ill *dayyg fisherman (attested as plural)9 dayyn judge

I have excluded several words with the pattern qtl, which are fairly certainly Aramaic loanwords. 9 There is also one case of a ktib <dwgym>, qr dayygim.

Spelled <swr>; Bauer-Leander (1922) list it as a *qatladu form, *sawaru.

10

2.1.2.2 Verbs 2.1.2.2a All types of verbs Both the cohortative (eqtl let me x, niqtl let us x) and the long imperative (qotl x! (polite)) end in -. As PNWS word-final short vowels underwent apocopation, these endings should come from a long *-. If that is the case, then these are two examples of * not undergoing the Canaanite shift. 2.1.2.2b Verbs with a weak middle radical (hollow roots) The in the qal perfect and active participle of verbs with a weak middle radical is generally held to derive from an earlier long *. This explains why it is not reduced when it is more than a syllable away from the stress; compare qm standing, qm standing ones of (construct state plural) < *qmu, *qmay to yd hand, yd hands of < *yadu, *yaday. This * did not undergo the Canaanite Shift. It is probably the result of the contraction of the triphthong, i.e. *qawam(a) > *qm(a). 2.1.2.2c Verbs with a weak third radical The perfect in all stems of roots with a weak third radical ends in - in the third person masculine singular (and in -t in the archaic third person feminine singular forms wt so that it shall yield and helt it has frustrated (hipil)). As this is probably the result of a contracted triphthong, it is held to derive from a long * as well. None of the attested Hebrew forms show any trace of the Canaanite Shift. 2.1.2.3 Other There are some examples of occurring word-finally in endings, suffixes and particles. As discussed in 2.1.2.2a, this - cannot easily be explained from a short *-a and might therefore be an example of * that did not undergo the CS. The forms in question are - (the h locale or locative ending on nouns), att you (m.sg), -t (second person masculine singular ending of the perfect), -k (second person masculine singular object suffix on verbs or possessive suffix on nouns), -h (third person feminine singular object or possessive suffix), m what, kk thus and k how. 2.1.3 *a$> $11 This development is often explained by positing a change *a$ > *$, which would make the words that were affected liable to participation in the Canaanite Shift. See 1.2.2 for Dolgopolsky (1999)s explanation. Interestingly, these words are still spelled with <> in most cases. Exceptions to this tendency are br cistern and msr bands, which are always spelled without <>, mainly with a mater lectionis <w> (br sometimes defectively <br>); but compare br well and sar to tie with <> written.

11

$ marks a syllable boundary, so *a$ means *a before a consonant or word-finally.

2.1.3.1 Nominals 2.1.3.1a *qalu br cistern nd skin-bottle sn small cattle

r head 2.1.3.1b *matilu msr bands

2.1.3.2 Verbs In the verbal system, the change *a$ > $ is attested in the so-called p-lep verbs, a small group of frequent words that have as their first radical (note that consonantal has been restored in almost all other verbs that have it as their first radical). In the qal imperfect, these verbs have an -vowel after the prefix consonant, e.g. ymar he will say from mar to say. Verbs that regularly show this development are: bad to be lost b to want zal to leave har to stay behind kal to eat (also shows *a$ > $ in hipil: ykil he will cause to eat etc.) mar to say p to bake Verbs that show this development as well as strong forms with consonantal : hab to love (only in consecutive imperfect whab and I loved) haz to take (additionally, *a$ > $ twice in the nipal form nhazu they were taken) sap to gather *a$ > $ may also have occurred in the forms tbn you (f.pl)/they (f.) will come and wayyb and he came, if these are from *taban and *wa-yaba, respectively. 2.1.3.3 Other The feminine singular demonstrative pronoun zt this, written <zt>, seems to have come from *zat. Similarly, the negative particle l no is written <l> and is probably reconstructible as *la. 2.1.4 *a$ > $ 2.1.4.1 Nominals ms exit < *mawsau or *mawsu mr fear < *mawrau rim heads shows this development if it comes from *rama. For mli, see 2.1.2.1c; for sawwr, see 2.1.2.1e. 2.1.4.2 Verbs Verbs that have as their third radical drop it in the pronunciation, if not in writing (although it is sometimes replaced by a <h> due to confusion with verbs with other weak

third radicals). This results in forms like ms he found, yims he will find, spelled <ms> and <yms>, respectively. The verb sal to take away has the third person masculine singular consecutive imperfect form wayysel and he took away; contrast this with the usual behaviour of I- verbs (2.1.3.2). 2.1.4.3 Other The post-verbal precative particle -n and its longer form nn are spelled <()n> and might therefore have come from *-na. 2.2 Phoenician (from Friedrich-Rllig 1999 and Krahmalkov 2000) As there is no vocalization or reading tradition for the native script of Phoenician, we are dependent on transcriptions of Phoenician in other scripts that do write vowels. I will indicate in which script a form is attested by its initial; the scripts in question are A(kkadian cuneiform), G(reek) and L(atin). Please note that very few of the attested forms unambiguously evidence the operation of the Canaanite Shift, as Phoenician later underwent the so-called Phoenician Shift, which changed stressed *a to /o/ (or //, quantity of the vowel is unknown) in nondoubly closed syllables (e.g. *adamu man > /adom/, but *balu Lord (DN) > /bal/). As so few forms are attested, I list nominal and verbal forms together. 2.2.1 * > /o/ <alonuth> (L) goddesses < *ilntu (see also 2.2.2) <(h)avo> (L) lived, may he live < *haw? < *hawaya? (cf. 2.1.2.2c) <chon> (L) he was < *kna? < *kawana? (cf. 2.1.2.2b) <dobrim> (L) saying < *dbirma <donni> (L) my lord < *adniya <macom> (L) place < *maqmu <ozerbalos> (G) Bal is helper (PN) < *dir-balu 2.2.2 * > /u/ This /u/ (or /) is generally held to be a later development of * < *. <adoun> (G) Lord < *adnu <a-hu-ut-mi-il-ki>12 (A) Sister of Milk (PN) < *aktu-milki <alonuth> (L) goddesses < *ilntu (see also 2.2.1) <iusim> (L) exiting < *ysima <koul> (G) his voice < *qlahu <lasounalph> (G) cow tongue (name of plant) < *lanu-alpi <Oulmos> (G) Eternity < *lamu <rufe> (L) doctor < *rpiu <salus> (L) three < *taltu suf(f)es, -etis (Latin loanword) Suffete < *pitu

12

As cuneiform has no signs for syllables with an [o], this form could also belong in 2.2.1.

22

2.2.3 *a$ > /o$/ (or /$/) <corathi>13 (L) I called < *qaratu <nasot> (L) I lifted < *naatu 2.3 Amarna Canaanite (from Sivan 1984) Amarna Canaanite is a cover name for the West-Semitic linguistic elements that pop up in the Akkadian of the so-called Amarna Letters, a collection of clay tablets that were found at the former Egyptian capital at modern Al-Amrna. The Amarna Letters date from the 14th century BCE. All forms are attested in Akkadian cuneiform. Although Amarna Canaanite still marks case on most nouns, I will not indicate the case in the translation, as it is not very relevant. 2.3.1 * > // (or another back vowel) As Akkadian lacks signs for [o], these forms are written with signs for [u], but they are reconstructed with // based on Hebrew and Phoenician. The forms are: <LUa-bu-ti-nu> our fathers < *abtin <LUa-bu-tu-nu> our fathers < *abtun <ah-ru-un-> after him < *akrnahu <a-nu-ki> I < *anku <URUbe-ru-ta> Beirut (LN), probably wells < *birta <hu!-mi-tu> wall < *hmiy(a)tu <ma-u4-un-nu> dwelling < *manu (reading is problematic) <s-ki-ni> officer < *skini <u-s-mi> despoilers < *iymi <-bi-il> bearer of (construct state) < *ybilu <-bi-li-mi> bearers < *ybilmi <-[zi]-ri> helper < *diri <zu-ru-uh> arm of (construct state) < *diru 2.3.2 *a$ > // (or another back vowel) <ru-u-nu> our head < *raun <s--nu> small cattle < *anu 2.4 Ammonite, Moabite and Edomite (from Garr 1985) There is very little conclusive evidence of the CS available in the other attested Canaanite languages. In an Akkadian text, the Ammonite personal name <pu-du-il> is attested, which shows a development *aya > * > // if it is indeed from *padaya-ilu, meaning El has ransomed. The name Ammon itself, spelled <m(w)n>, is another example, as it probably comes from *amm- kinsman with the *-nu suffix. There is an attestation of <am-ma-na-aia> Ammonites in Akkadian which does not show the CS, but this might be an adaptation to Akkadian. Garr (1985:30) also suggests the name of the Ammonite
13

Probably a scribal error for **<carothi>.

23

god Milkom as an example, but as its etymology is unclear (cf. Gzella 2012), I do not think it is a very good one. Moabite and Edomite, finally, offer very little evidence indeed. There is one Moabite personal name that seems to have undergone the shift attested in Akkadian, but the same nominal element is also attested twice with <a> (Garr 1985:31). All the Edomite evidence is from questionable personal names as well, and I will therefore not discuss it. 2.5 Preliminary evaluation of the evidence The vast majority of evidence for the Canaanite Shift is from Biblical Hebrew, and this language will bear the main focus of the rest of this paper. As noted above, many of the relevant forms in Phoenician could also be the result of the Phoenician Shift; forms that show a rounded back vowel (i.e. /o/ or /u/) for * that was unstressed when the Phoenician Shift occurred, are <alonuth>, <Oulmos>, <koul>, the construct states <lasounalph> and <a-hu-ut-mi-il-ki> (which probably predates the Phoenician Shift anyway) and the qal active participles <dobrim>, <ozer>(-<balos>), <iusim>, <rufe> and suf(f)et-. As the other Phoenician forms may be secondary, they cannot be used in a convincing argument. Like Phoenician, Amarna Canaanite does not show any good examples of the CSs non-occurrence. Due to the nature of the Akkadian script, we cannot be certain of the quality of the vowel that the CS resulted in in Amarna Canaanite. Finally, the other Canaanite languages offer too little and too poor evidence to advance the discussion in any meaningful way. 3. Reconstruction of the relevant material 3.1 Nominals 3.1.1 ql, ql, qlet, ql, qlim, Phoenician /qul/ (BL: *ql((a)t)u; for *qawilu, see 3.2.2) Most of these forms have cognates that firmly establish the presence of * in these forms14. As cognates of hl sand, Jewish Aramaic and Syriac have hl and Arabic has hl black fetid mud. Khne (1974:166ff.) mentions the Ugaritic word <hu-li>, occurring in several place names in Akkadian texts, which would point to *hawlu, but its meaning is uncertain. The Aramaic and Arabic evidence seems to outweigh the Ugaritic in this case; hl sand probably comes from *hlu. For hp shore, Arabic does have hfa edge, but the more convincing Ugaritic cognate <hp y[m]> seashore (KTU 1.3ii.7) shows that it is not related to the Hebrew (the regular correspondence should result in Arabic **kfa or similar). There is a syllabic transcription of a Ugaritic word /huppati/ or similar (Huehnergard 1987), but the meaning is uncertain; either way, without the Arabic, there is no cognate of hp that reflects *. An * in ks cup is supported by Aramaic ks (various dialects), but Arabic kas suggests *kasu; Akkadian /ksu/ could be the result of either form. Since the

14

r: Ug. <ar> (Akk. /urru/ day(light) is probably a different word); dd: Akk. /ddu/ beloved one, lust,

Arab. dd foster-father; hah: Akk. /hhu/; tb(): Akk. /tbu/, Aram. tb; *y(e): Aram. y.

24

Hebrew spelling (<kws>, never **<ks>) does not point to an original form with *, however, the Arabic form of the word may well be secondary. Aramaic ql supports * for BH ql and Phoenician <koul> (G) (his) voice; Arabic qawl speech does not, but this may well be an analogical form based on the verb (e.g. mta he died : mawt death = qla he said : x; x = qawl). Bauer-Leander (1922:458) adduce Akkadian /qlu/, but the CAD lists its meaning as silence, a word that must almost certainly be unrelated. qet watering-trough must seemingly be reconstructed with * based on Arabic sqiya, but this is simply an active G-stem (qal) participle, a different noun pattern. qet can simply be reconstructed as *uqtu. While ss moth < *ssu is supported by Akkadian /ssu/ and Aramaic ss, Arabic ss moth-worm and Geez e moth, worm show that a straightforward Proto-Semitic form is hard to reconstruct. Perhaps the word is a Wanderwort, a word that is borrowed into many different languages without a clear source language; the fact that it referst to an animal species would not make this implausible. The word may have been borrowed into BH after the operation of the CS or as pre-Hebrew *sas. rim cities, mentioned by Blau (2010:48), has no attested cognates that show it must have had *; rather, it could be the result of a contracted broken plural *ayarma, cf. Mishnaic Hebrew ayrt cities (Koehler-Baumgartner 1985). Ugaritic <rm> cities is either a post-contraction form or a dual, which would have the same base as the singular. 3.1.2 qtl, qtl, mne, Phoenician /adn()/, /aht/, /lan/, /al/ (BL: *qatlu) See footnote15 for unproblematic forms. For *rn ark (of the covenant), Akkadian has /arnu/, but also /arunnu/ and /erinnu/. Aramaic has rn in various dialects, first attested as <rwn> in a text from around 500 BCE (Beyer 2004:I 137); although this might reflect *arawnu, it could also be an example of Aramaic * > after dental continuants (n, r, l; Ibid.). t sign is probably related to Syriac t, Arabic ya sign and Akkadian /awatu/ speech, what is said, but the exact relationship is unclear, and so therefore is the reconstruction. The final vowel in mne eight suggests the word be reconstructed as *tamniyu, cf. Arabic tamn(in) with -in < *-iyun. Aramaic tmn confirms the quantity and quality of the *. nn cloud might be reconstructed with * based on Aramaic ann etc. and Arabic ann clouds. The construct state anan, however, seems to show *ananu for BH (the one instance of annk your cloud is comparable to dbrk your word from *dabar-); Meyer (1966:III 25), for one, lists nn as a *qatalu form without comment. Also note that the Arabic word is a collective (the nomen unitatis, cloud, being anna), while the BH word has a singular meaning and might therefore be of a different noun pattern.
dn: syll. Ug. <a-da-nu> father, att. alphabetically as <adn> father, lord (also <ad>; so perhaps *ad-nu); ht: Akk. /ahtu/, Aram. aht; tn: Akk. /atnu/, Arab. atn; grn: Arab. jirn (*girnu probably > *garnu in PNWS); ln: Akk. /linu/, Arab. lisn, B.Aram. lin etc., syll. Ug. <la-a-nu> (*linu > PNWS *lanu like *girnu); rd: B.Aram. ard; lm: Akk. /almu/, Arab. salm, Aram. lm etc.; l: Akk. /alu/, Arab. talt, Aram. tlt (Old South Arabian evidence points to *altu, with * > *t in most other languages); qtl: Akk. /parsu/.
15

25

3.1.3 q(a)tl(), tl, qittl, mli, Amarna Canaanite /zr/ (BL: *qitl(iyy)u, qitlatu) A few words in this category can be reconstructed to *qitlu unproblematically16. zr loincloth shows the regular development of *iC > C before a long vowel, rather than dropping out as it does before a long vowel after other consonants (Fox 2003:226, there formulated as applying before masoretic high vowels rather than long vowels, which amounts to the same). Its reconstruction as *izru is confirmed by Arabic izr loins. The reconstruction of pd ephod is less clear. It might be related to Akkadian /epattum/ cloth, but as listed in the CAD, this word has short /a/ (contra KoehlerBaumgartner 1985). BH apudd suggests that pd is from *upudu, with the development of *u- > *i- > described for other patterns by Fox (2003:226). Even less clear is the etymology of hadm footstool of. It has a cognate with unknown vocalization in Ugaritic <hdm> and it is attested in Egyptian as <hdmw>, but Osing (1976:374) states that this Egyptian word is a loanword from a Semitic language. There are no known cognates of hagr and hagr belt; Bauer-Leander (1922) ascribes them to the *qitlu pattern based on semantics, but they might be *qutul(at)u forms like the reconstruction proposed for pd above. The same goes for r leaven. The etymology of abt rope is also unclear; Koehler-Baumgartner (1985) connects it with Akkadian /abuttu/ slave-mark, but the stable and ungeminated t in suffixed forms point to either *ibtu from -B-T to pervert (*to twist?) or *ibawtu, a *qitaltu form of -B-Y to be thick. It is hard to decide which etymology is correct. There are no cognates that prove srr purse had *; Arabic sirr is not a purse, but rather a string that is tied around a she-camels udder (Lane 1863; contra KoehlerBaumgartner 1985, Klein 1987). The case of ml left hand (side) is a very interesting one. For starters, it is spelled both <mwl> (31 times) and <ml> (23 times); contrast this with the *qalu words r head, sn small cattle and nd skin-bottle, which are always spelled <r> 17 , <sn> and <nd>, respectively (see 3.1.17, also for br cistern), and their Samaritan Hebrew versions ro and son (nd does not occur in the Samaritan Pentateuch) with retained or restored vs. ml without. When we look at the words cognates, Syriac semml and Arabic iml both point to *imlu, a form that should indeed result in Hebrew ml; Akkadian /umlu/ shows unexpected vowel qualities. Arabic, however, has another set of closely related words: iml (again), aml and amal (and several less frequent variations), all meaning north wind (i.e. the wind coming from your left when oriented towards the east; cf. *yamnu right and south). In this last form, we find the that is written in BH <m(w)l> and Old Aramaic <ml>. The situation is made even more complex by the vocalism of mli left-hand (adjective), which shows an for mls . As stated in 1.2.2, Dolgopolsky (1999) attempts to solve this problem by reconstructing the forms as Proto-Semitic *imaalu and *imaaliyu; *imaalu left

lah: Aram. elh etc., Arab. ilh; zra: Aram. dr etc., Arab. dir; hamr: Akk. /imru/, Aram. hamr, Arab. himr; rk: Arab. irk; thm: Akk. /timtu/, Arab. tihma (name of a coastal plain; Syr. thm is probably a BH loanword). 17 One case of <rw> with a secondary meaning poisonous plant.

16 e

26

hand (side) then undergoes the development > *imalu > *imlu > *imlu > ml, while *imaaliyu left-hand becomes *imaliyu > *imaliyu > mli. The difference in vowel quality would then be the result of a different stress pattern in Proto-Semitic. The differences in spelling and Samaritan Hebrew pronunciation between ml and other *a words discussed above, however, make this solution problematic. Proponents of the stress-based conditioning of the CS attribute the different vocalism directly to a different stress pattern, as Blau (1993:35), who derives mli < *iml. What happened to the * in ml < *iml remains unexplained. It seems, regardless of spelling, that the spoken form ml goes back to *imlu as reflected in Arabic iml. Of the three most common Arabic forms, amal < *amalu allows us to derive mli fairly unproblematically, if we assume * elided between a preceding consonant and a following unstressed *a, as in, for instance, mlk work, spelled <mlkh>, probably < *malakatu (cf. Barth 1906:794). Perhaps the <> in the spelling of ml was influenced by this still-pronounced * in what was to become mli. To be fair, mli could also come from *imliyyu or *amliyyu, corresponding to the other two Arabic words, but then the spellings with <> are inexplicable. To conclude and summarize this rather confusing discussion: the least problematic solution seems to be the existence of (at least) two versions of the ProtoSemitic word for left hand (side), *imlu (Arabic iml) and *amalu (Arabic amal). For whatever reason, by the time the Masoretic pronunciation was fixed, *imlu > ml had become the only remaining form for the noun, while *amaliyyu > mli (with elision of *) was the only remaining form for the adjective. The spelling <ml> may still reflect substantive forms deriving from *amalu, while <mwl> possibly reflects the outcome of *imlu with a spelling that has been influenced by the form with *. There are no direct cognates of abd work, service, but the stability of in an open, unstressed syllable shows that it is originally long, making *ibdatu the most likely reconstruction. Wagner (1966:28) lists issr oath of abstinence as an Aramaic loanword, but states that the form with suffixes is in accordance with sound laws. While this may be true for the development of the first vowel, the is unexpected if the word is truly cognate with Biblical and Syriac Aramaic esr. The form with suffixes is probably also borrowed from Aramaic, or, less probably, the Aramaic forms are borrowed from Hebrew, in which case there are no cognates showing the presence of an original * in the word, which could then be derived from *isaru. nissb handle (of knife) seems to be paralleled by Arabic nisb, but once again, the gemination of the second consonant is problematic. Klein (1987) sees the BH word as the nipal participle of N-S-B to stand, i.e. that which (the blade) is set in. Formally, this works quite nicely. 3.1.4 q(e)tl(im), qtl, qtlet (BL: *qutl((a)t)u, *qutlma) See footnote18 for unproblematic forms.

n: Aram. en etc. human, Arab. uns men; bly: Syr. bly worn-out (clothing); ptt: Arab. futt; lbn: Arab. lubn (without fem. afformative); thrim: Syr. thr straining at stool (Syr. thr idem perhaps Heb. loanword).

18 e

27

br cypress can fairly confidently be reconstructed as *burtu based on Akkadian /buru/ and Aramaic brt. There is, however, a puzzling form, attested once as brtim in BH and as brt in Syriac, which appears to show the effects of the CS, but not of the common Canaanite development *t > . Perhaps this is a loanword from a Canaanite dialect that preserved *t or shifted it to /t/ like Aramaic, or maybe it is the Aramaic version of the word that has been influenced by Canaanite vocalism. Alternatively, this could reflect the same shift of Aramaic * > after r seen in the spelling <rwn> (Beyer 2004:I 137). nhet copper, bronze should be reconstructed as *nuhutu, as is clear from suffixed forms; it is from a different noun pattern than Aramaic nh and Arabic nuhs. Bauer-Leander (1922) reconstructs nret tow as *nuratu based on its meaning, as *qutlu is a common pattern for waste products in Arabic and Ethiopic (cf. Fox 2003:229ff.), but without cognates, there is no need to reconstruct this form for this word, especially since it should result in a different BH reflex, **nr. 3.1.5 lm, Phoenician /lm/ (BL: *qtalu) If we accept the reconstruction *lamu for lm eternity, which is based on Aramaic, Arabic and Ethiopic lam, that would make this word the only Semitic word with a *qtalu pattern, or with any disyllabic pattern other than *qtilu (the G-stem (qal) active participle) with a long vowel in the first syllable (Fox 2003:289). See Fox (ibid. and f.) and Jenni (1953:3-6) for various alternative reconstructions. Considering the words etymology is so unclear, we cannot use lm as evidence of the CS. 3.1.6 qtl(), qtelet, qt, qtit, various forms in Phoenician and Amarna Canaanite (BL: *qtil((a)t)u) The qal (G-stem in other languages) active participle, BH qtl, almost certainly derives from *qtilu; compare Akkadian /prisu/, Aramaic qtel, Arabic fil, Geez qtl (not a productive participle, although it is in Tigria; Fox 2003:239). Some words with this pattern are not (or no longer) semantically participles, most notably khn priest without an associated qal verb; cf. Aramaic khen etc. priest and Arabic khin soothsayer. It is unclear whether this Arabic word is derived from the G-stem verb kahana to divine or vice versa. rb raven has a direct cognate in Akkadian /ribu/, while other languages show different vowels: Jewish Aramaic rb (may be borrowed from Hebrew), Syriac urb, Arabic gurb. It may be that Aramaic and Arabic changed the words vowels to more closely resemble the ravens cry, or Akkadian and BH reinterpreted the word as a participle, perhaps of cognates of the Arabic verb gariba to be black. qr, another type of bird, is probably cognate to Arabic qriya bee-eater, which has added or retained a feminine afformative. hm city wall (Amarna Canaanite: <hu!-mi-tu>) is probably originally a feminine participle *hmiy(a)tu of a cognate of Arabic ham to protect. *dlit foliage and *zwit have exact parallels in Aramaic, occur very rarely and, in the case of *zwit, show retention of *w, which is unusual for BH in this position. They are probably Aramaic loanwords. 3.1.7 qittl, qattl(et), prket, qattl(at)u (BL: qattl((a)t)u) 28

These words are commonly derived from *qattlu, a pattern that is attested in most Semitic languages, cf. Akkadian /dayynu/ judge, Aramaic kaddb lying, Arabic tabbk cook, Geez assb hireling. This pattern has an agentive, habitual meaning, and is often used for profession names. Interestingly, hardly any of the words held to be the BH reflexes of this pattern can be shown to have a long vowel in their final syllable. Of the words with , those that occur in the plural or with suffixes (where an from *u would normally reduce to zero or cause gemination of the following consonant) both have an r following the , neutralizing the difference between * and *u; i.e., gibbrim heroes could theoretically come from *gabbrma (assuming the i in the first syllable comes from *a), or from pseudo-PNWS *gabburrma with secondary gemination of the *r, followed by pre-Masoretic degemination of *r and compensatory lengthening of the reflex of *u. Huehnergard (1992:221-2) does indeed derive the qittl words from the *quttulu pattern, attested in Akkadian. The Samaritan Hebrew reading tradition, however, does allow geminate r (cf. rr Sarah (PN), Tiberian r), and the attested form aggibbrm the heroes (Tiberian: haggibbrim) shows that this word at least did not have a geminated third radical in the plural (Ben-Hayyim 1977). The direct Syriac cognate gabbr hero too strongly suggests that this word is a *qattlu form, but a derivation from *quttulu is not ruled out for the other words. kappret cover (of the Ark of the Covenant) and prkt temple curtain would then come from *qattultu forms. This explanation leaves two forms with an in the second syllable and an a in the first, which cannot directly derive from *quttulu or *qattlu if these patterns do indeed yield BH qittl. Of these, rattq is only attested once with , and this may be a scribal error for **battq slaughter (Koehler-Baumgartner 1985), a nominalized pil absolute infinitive. qann jealous, which is attested twice as an alternative to the more copious qann idem, may have taken over the first vowel of this more frequent synonym. On another note, kinnr zither looks like the other qittl (<*quttulu?) forms, but it is a loanword, ultimately from the ancestor of Telugu (Dravidian) kinnra or a related word. Be that as it may, it should be reconstructed with * for an earlier stage of (pre)Hebrew; cf. Syriac kennr, Arabic kannra and kinr. While the forms with may be derived from *u, those with could come from short *a. Once again, there are hardly any attested words where *qattlu with retained and *qattalu would result in different reflexes. This should happen in the construct state plural (or the plural with heavy suffixes), where the two forms should yield qattl and qattl, respectively. Yet here too, the possibility of secondary gemination (yielding an expected qattall) causes trouble; the two attested forms that should be decisive, hatt sinful ones of and sawwr 19 necks of could also reflect forms with secondary gemination, as they both have a guttural or r as their third radical. These two forms are not attested in the Samaritan Pentateuch or the preserved fragments of the Secunda, which might resolve the question20. The presence of a long can be seen in nstek your contumelies, but Bauer-Leander consider this and other qattl forms to be patterned after Aramaic D-stem infinitives and thus secondary, a plausible suggestion
19 20

But cf. sawwrtkem your necks with reduction of *a. There are attestations of Tiberian hattim sinners, i.e. Samaritan ttm and <attaeim> (G) in the Secunda; but Samaritan Hebrew does not geminate , and the Greek alphabet does not write it, so these forms are inconclusive.

29

given that *nes is only attested in Ezra and Nehemiah, books with strong Aramaic influences. Although *qattalu, the other pattern that BH qattl forms may trace back to, is rare or non-existent in almost all other Semitic languages, its existence is proved by a set of words in Akkadian, more specifically Old Assyrian. This dialect of Akkadian underwent a sound change that resulted in variation in the last short *a in *qattalu forms, which assimilated to the case ending (Fox 2003:254ff.). For instance, the word thieves is spelled <a-ru-q> in the nominative, but <a-re-q> in the genitive-accusative. *qattlu forms, which are also attested in Akkadian, did not undergo this change; cf. <napa-hi-im> smith (genitive), which must therefore be read /napphim/. As with the qittl words, it is counterintuitive to reject the *qattlu reconstruction while so many related languages have reflexes of this pattern with very comparable meanings; even Akkadian, which has both *qattalu and *qattlu, has the *qattlu form /dayynu/ as a cognate of BH dayyn judge. Still, it is more economical to assume the BH qattl words come from *qattalu, given its proven existence and similar semantics to those of *qattlu, than to either posit unnecessary sound laws or to resort to borrowing from Aramaic as an explanation. The etymology of sawwr, which is spelled <swr> but has no cognates in which is pronounced, remains uncertain. 3.1.8 sippren, rimmn (BL: *quttlu) Assuming the vocalism of the cognates 21 in related languages is original, these forms should indeed be reconstructed as *quttlu. Both words may have undergone dissimilation of *u > *i before the following labial consonants, cf. *ummu mother > *immu > m. sippren nail has added an afformative -en, possibly to avoid confusion with sippr bird of unsure etymology. 3.1.9 mq(l), mql, mmt, mrs, Phoenician /maqm/, Amarna Canaanite /man/ (BL: *maql(at)u, *maqlma, miqlu) This is a quite productive pattern, being *maqtalu nouns formed from hollow roots, roots with W or Y as their middle radical. As such, they mainly refer to places, tools and actions. Hollow roots in other Semitic languages also form their *maqtalu nouns like this, i.e. they have reflexes of Proto-Semitic *maqlu, cf. Aramaic mqm place, Arabic and Geez makn idem. The CAD lists Akkadian /mnahtu/ resting place thus, with a long first vowel and a short second one; perhaps this is some kind of metathesis quantitatis due to the following feminine afformative. Brockelmann (1908:I 186) attributes these forms to a general sound law that he holds to have operated in Proto-Semitic: *CWVC > *CV:C, i.e. *w and *y disappear between a preceding consonant and a following vowel, lengthening the vowel22. *maqlu is thus a regular development of *maqwalu and *maqyalu. Note that this development does not happen in doubly weak roots that have *w or *y as their second and third radical; cf. BH miqwe hope from Q-W-Y, Arabic miw grill from -W-Y, Geez
sippren: Akk. /suppru/; rimmn: Aram. rummn etc. (Arab. rummn possibly borrowed from Aram.) Actually, Brockelmann formulates it as *wa, *ya, > *, *wi, *yi > *i, *wu > *u (all after consonants and within words), not mentioning *yu.
22 21

30

mhyw place where one lives from H-Y-W (the Arabic form is not *maqtalu but *miqtalu, the other two may be as well). Some of the BH forms have direct cognates in other languages, e.g. mqm place ~ Arabic maqm, mn dwelling ~ Arabic man, mmt death of ~ Arabic mamt. mrs race is not the expected form; *mirtu should have yielded **mrs, cf. *diru arm > zra. This is probably due to analogical restoration of the *i based on *miqtalu forms from strong roots. 3.1.10 maqtl, miqtl, miqtlet, mahagret, mammrrim (BL: *maqtl(t)u, *miqtl(t)u, *maqtlma) *maqtlu and *miqtlu nouns are also attested in other Semitic languages, although they are rarer than the related *maqtalu and *miqtalu nouns; cf. Syriac makt blow, Arabic mifth key. Forms that are not attested in the plural or with suffixes (as matmn, *makb, makmr, mikl, malqah and miqsa are) might also come from *maqtulu; that this pattern exists in BH is proved by the correspondence of, for instance, maaklet food to Arabic makula, both from *makul(a)tu. The only actual cognate that is attested is thebisch (Koehler-Baumgartner 1985) mizmre (type of pipe) ~ BH mizmr psalm. The older form of this word, however, as reflected in the Secunda transcription <mazmr> (G), shows that this is not an exact correspondence. Given the existence of *maqtulu nouns in BH, it seems prudent to qualify the feminine nouns in this category as such. mammrrim has a dg dirimens in its second <m>, which is to say that its gemination is secondary; it comes from *mamrrma. Note that *maqtlu etc. forms with retained would have merged with *maqtalu in the singular and cannot therefore be easily identified. 3.1.11 q(t)ln, abaddn, qit(t)ln, rayn, abbtn, znunim, man (BL: qatalnu, qatalnma, maqtalnu) In Arabic, *qatalnu forms verbal nouns (masdars) of some verbs of repeated motion, e.g. ramaln walking. In Aramaic, they are closer to the meaning of the BH words, forming abstract nouns like Syriac abdn perdition. In the absolute state, all words from strong roots except for przn and rbn geminate their second radical (and many words from weak roots do so as well). BauerLeander (1922:498) attribute this to the effect of the secondary stress on the vowel preceding the second radical. Words in which the second radical is a sibilant carry this gemination over to the construct state. abaddn perdition is unique in geminating its third radical, rather than the second one. The motivation for this is unclear, but Syriac abdn shows it belongs in this category. Given the pan-Semitic frequency of the *-nu afformative and the universal lack of an *-unu suffix which might also result in -n, it seems quite safe to assume this pattern did originally end in *-nu. abbtn sabbatical has retained the a in the first syllable due to contamination (see 6) by the noun it is derived from, abbt sabbath.

31

Brockelmann (1908:I 52) gives an alternative explanation for znunim fornication. He sees it as a qtulim abstract noun, like e.g. nurim youth from N--R, with an epenthetic n added to avoid the hiatus after the disappearance of the *y in *znuyim from Z-N-Y. This does explain the plural number of the word, and n-insertion is not unheard of, cf. Brockelmann (ibid.). man deception seems to be a rare case of a *maqtalu word with an n suffix tacked on. As with other highly productive categories, individual instances of this type do not carry very much weight when establishing sound laws, as they could always be secondary or analogically restored. 3.1.12 hetln, qitln, qitln, qorbn, ulhn (BL: *qitlnu, *qutlnu) This pattern forms verbal nouns and action nouns, cf. Syriac neshn victory, Arabic ityn coming. Several words in this category have direct cognates23 showing *. Possibly due to the small number of words in this category, some interesting tendencies can be observed. Firstly, there is the restriction of *qitlnu forms with to Qohelet24. It is tempting to see this pattern as a dialectical feature unique to Qohelet; but inyn task, with an in the afformative, is also attested in this book (and nowhere else), while the other qVtln words are attested in other books. Secondly, most of the qitln words have Y as their third radical (i.e. they are qityn forms). An exception is formed by kibn kiln, a word of uncertain origin. Koehler-Baumgartner (1985) and Klein (1987) derive it from K-B- to subdue, rape, reasoning that a kiln is that in which metal is subdued. This seems slightly far-fetched, especially since in the words three attestations, the simpler meaning of oven fits just as well, while it does not match the root at all. Additionally, all the other *qitlnu and *qutlnu words (except for ulhn table, see below) have patientive meanings: binyn building = what is built, inyn task = what must be answered for, yitrn profit = what is left, etcetera. Following this pattern, kibn should mean **what is subdued, perhaps **smolten metal if the association with smelting is correct. Tellingly, we do find the expected semantics in post-Biblical Hebrew kibn secret, i.e. what is suppressed (Klein 1987). Given the uncertain derivation and lack of cognates of kibn, its reconstruction must remain uncertain. As noted in 1.1.2, Bergstrsser (1918) attributes the non-operation of the Canaanite Shift in qorbn sacrifice and ulhn table to the rounded back vowel in the first syllable, which is plausible (see 5.2.1). The etymology of both words is uncertain; qorbn is from Q-R-B to be near with *-nu, but has no cognates that are not suspected loanwords from Hebrew (such as Aramaic qurbn, Arabic qurbn, Geez qwrbn). The semantic transparency of the word does, however, support the reconstruction *qurbnu. For ulhn, alphabetic Ugaritic has <tlhn>, disproving an association with Aramaic ilh etc., Arabic salk skin. Considering the unsure etymology and the reduction of the in the construct state ulhan (sg.) and ulhant (pl.), we cannot reconstruct the word with *. 3.1.13 Derived adjectives in -n and -n (BL: *-nu)
23 24

hebn: Syr. hebn; binyn: Syr. benyn; inyn: Syr. enyn. hebn reckoning is also attested in Sirach, but this may be conscious imitation of Qohelets style.

32

A very productive category in many branches of Semitic. Cf. Akkadian /ndinnu/ seller from /ndinu/ giver, Aramaic arn earthly from ar earth, Arabic qarnn horned from qarn horn, Amharic (dialect of Gafat) tekurn black. rin first is unexpected; as it is derived from r head (as the spelling <rwn> bears witness to), one would expect **rn, more or less like qorbn sacrifice. In this case, however, the first vowel seems to have dissimilated, not that of the afformative. Compare this to the same unexpected vowel in *qisn last from qs end and tikn middle from twek (tk when unstressed) middle. The etymology (and even meaning!) of hammn incense-stand? is unclear, and there are no forms showing that its -n comes from *-nu. liwytn Leviathan seems to be a more recent formation than its Ugaritic cognate <ltn>, probably /ltnu/. If this form were preserved in BH, we would expect it to yield **ltn, with -n like qorbn. Suppose this were the case, then liwytn might also have taken on that form of the suffix by contamination (see 4.2). nhutn serpent idol shows the same preservation of -n as qorbn. It is tempting to see this as another possible source of analogical retention of that suffix in liwytn, but I believe this is a bit too far-fetched. Finally, the -n in tn adversary is only a suffix if the word is derived from W-T to rove, as Bauer-Leander (1922:500) suggests; but as the sibilants do not match, it might be better to see tn as a *qatalu form of -T-N to bear a grudge, rather than to see the root as secondary and based on the noun. 3.1.14 Diminutives in -n (BL: *-nu) This category, too, is attested in other languages: cf. Arabic aqrabn earworm from aqrab scorpion, Mehri gyn youngster from gayj man. Aramaic brn little son from br son may have borrowed the Canaanite form of the suffix. Note the reduction of *a in sawwrnayik your (f.sg.) necklaces < *sawwarnayki, supporting the reconstruction of sawwr neck < *sawwaru. 3.1.15 Derived adjectives in -ni (BL: *-niyyu) As these words simply have two derivative afformatives tacked on one after the other, their reconstruction is fairly secure. 3.1.16 The feminine plural ending -t, Phoenician -<uth> (L), Amarna Canaanite -<Cu-tV> (BL: *-tu) This ending can securely be reconstructed as *-tu based on forms like Akkadian /-tu/, Aramaic -t etc. and Arabic -t. 3.1.17 ql <ql>, br, msr, mt, Amarna Canaanite <ru-u->, <s--nu> (BL: *qalu, *masiru, *mawtau) See footnote25 for the attested cognates of all words but br. The three words r head, sn small cattle and nd skin-bottle offer an interesting problem. They can quite securely be reconstructed to *qalu based on cognate
r: Akk. /ru/ and /ru/, Ug. <ri>, Aram. r etc. (<r, ry, r>), Arab. ras; sn: Akk. /snu/, Ug. <sin>, Aram. n, Arab. dan; nd: Akk. /ndu/ (Ug. <nad> little bag cannot be /nadu/, rather /naadu/; but it shows the root has * as its second radical).
25

33

evidence. In Tiberian Hebrew, their etymological * is no longer pronounced (whereas Samaritan Hebrew does have it: ro head, son small cattle). Based on the spellings in the Amarna Letters, this pronunciation seems to be quite old, and the common explanation, *qalu > *qlu > *qlu > ql, has the words undergoing the CS, which had already taken place (in Amarna Canaanite, at least) by the 14th century. Yet the words are almost exclusively spelled with a <>, almost never with a <w> mater lectionis. How can this be? One could assume that the <> was retained based on other forms. If there was a plural *raam with preserved * (as reflected by Ugaritic <ram>) synchronically opposing the singular *r, then the spelling of <> could be extended from the plural into the singular (thus Bauer-Leander 1922:224). But sn is a collective noun; there could not have been an accompanying plural form, and indeed none is attested, yet the word is always spelled <sn>, so this cannot be the whole solution. Another option, the one embraced by Brockelmann (1908) and Bauer-Leander (1922), among others, is that the <> in these forms is a case of historical spelling: <r> and <sn> started to be written that way when they were still pronounced with an audible *, and retained their spelling after the development of *qalu > *qlu > *qlu. This means that the alphabetic writing tradition that ultimately resulted in Hebrew and the other Canaanite languages must have been transmitted since before the 14th century BCE. Although this is long before the generally assumed date of composition of the earliest parts of the Hebrew Bible, it is not that much earlier; after all, Ugarit was using its own version of the Semitic alphabet by the 12th century. According to Driver (1948), the alphabet was in common use in the Levant by the mid-first millennium BCE. The historical spelling theory is backed up by data from a language that has not contributed very much to the discussion so far: Moabite. In line 20 of the Mea inscription (KAI 181), we find the word <rh>, probably to be translated as his officers (= heads)26. Here we see a very old attestation of head spelled without <>, probably reflecting a plural /r-/ or similar based on the singular (cf. Ugaritic /rim/; BH rw his heads in Isaiah 15:2 is a reflex of the same form), rather than the broken plural **/raa/ (cf. Ugaritic /ram/, BH rim) which would have retained the *. That this spelling is probably a slip-up, a case of the scribe forgetting to employ the historical spelling rather than an independent development of Moabite, is shown by the normal spelling of small cattle as <sn> in line 31. It could be argued that these two forms show a different development of * in *rau and *anu, but this seems very unlikely. Furthermore, the alternation between the spellings <r-kd> (/r-qudi/) and <ri-kd> (/ra-qudi/) in Egyptian texts (Burchardt 1910:I 53) support an early date for elision of * in this word. The consistent spelling of r and sn with <> (also in inscriptions: <sn> in Phoenician KAI 26, <r> in Phoenician KAI 24, 30, Hebrew KAI 189 (the Shiloam inscription) and Neo-Punic KAI 145; <r()> a few times in late and peripheral inscriptions), then, is probably due to the high frequency of these words. nd is less frequent; perhaps in this case the <> was in analogy to an unattested plural *naadm.
26

A cognate of BH r poor does not fit the context; connecting <rh> with Aramaic <ry> entitled or re creditor (Beyer 2004:II 484) also seems far-fetched, given the context and the fact that this root is not otherwise attested in Canaanite.

34

The plural of r has been given as an example of a long * that did not undergo the CS. Its absolute form is rim, while construct state r shows that the in the first syllable is indeed historically long (short *a should yield **r). But as Bauer-Leander (1922:620) show, these forms can be recent developments, the result of the late elision of *a when more than one syllable away from the stress: Proto-Hebrew absolute state *raam > *rim > rim, construct state *raa > *ra > r. By the time the new instances of * were formed due to pretonic and compensatory lengthening, respectively, the CS had not been operative for a very long time, and they therefore ended up as BH . The spellings <b(w)r> and <mwsr> for br cistern and msr bands show that etymological * had been lost in pronunciation by the time the Hebrew Bible was written. Whereas r, sn and nd preserved their <> in writing due to frequency or analogy, it could be left out when unneeded. In the case of br, a spelling <br> would be identical to that of br well, and as the two words are semantically similar, a reading mistake would be easily made. As for the reconstruction of the word, Koehler-Baumgartner (1985) give Arabic bura cooking-hole as a cognate. Other *qulu forms yield ql, e.g. *muda very > md, but the may have been reintroduced secondarily in these forms based on the spelling (<md> etc.). With br, this could not happen, as it was not spelled with <>. Therefore, we cannot be sure that it goes back to *baru. The same should go for msr, but as *maqtilu is a reconstructible noun pattern and *muqtilu is not, the word must derive from *masiru. Note that ms exit and mr fear do not show *a > in Auslaut. The long in ms, construct state plural, may be the result of secondary gemination and subsequent degemination of the *; if it is original, it mayb be a case of * > after a back rounded vowel, like qorbn sacrifice. 3.1.18 Miscellaneous ekl grape has a clear cognate in Aramaic etkl and Arabic itkl. tb resident alien should be reconstructed as *tawtbu, cf. Syriac tawtb; the non-operation of the CS in this word is probably due to the rounded back vowel in the first syllable. The other miscellaneous words have no clear cognates and are not formed after very productive patterns, so it is probably best to leave them out of consideration. 3.2 Verbs 3.2.1 Pl and pll As mentioned above, the rare pl matches the L(engthened)-stem of Arabic (stem III, fala) and Geez (03 in Tropper (2002)s terminology, qtala) quite well formally; all of these perfect forms could derive from *qtala. As Gzella (2010) notes, however, they do not share the L-stems semantics. Gzella suggests they may be ad hoc neologisms, built in analogy to the pll of hollow and geminate roots, and it is to the pll that we shall turn our gaze. Arabic and the Ethiopic languages do not have anything like the pll: in both the L-stem and the D-stem (Hebrew pil, Arabic stem II, Geez 02), the types of roots that form plls in BH behave as strong verbs, hollow roots taking a y or w as their middle radical. Biblical Aramaic, however, does have a pll, with forms like mrmm praising (Bauer-Leander 1927:146-7). As Bauer and Leander note, the in these Aramaic forms cannot come from *, but only from *aw. They posit *qawlila for the perfect, which

35

would indeed result in the right forms in BH and Aramaic, and which is supported by the Egyptian spelling <wbb>, probably /awbiba/, a pll of -W-B (Burchardt 1910:I 53). Furthermore, they state that this stem formation probably originated in geminate roots like S-B-B and then spread to hollow roots like Q-W-M (*sawbiba *qawmima); an opposite development, however, could explain the origin of the *w in the form, which would then simply be the second radical of II-W roots. Seen in this light, *qawmima etc. strongly resemble the form of quadriradical perfects, *qatlida, while the formal similarity of both of these forms to the normal pil, *qattila, explains the plls semantic equivalence to the pil. Be all this as it may, we can conclude that the various forms of the pl and pll are probably not outcomes of the Canaanite Shift. 3.2.2 Verbs with a weak middle radical (hollow roots) and verbs with a geminate second radical (geminate roots) In the qal perfect of most hollow roots, we find an in BH (e.g. qm he stood) and reflexes of * in other languages, e.g. Aramaic qmu they stood and Arabic kna he was. Brockelmann (1908) does indeed reconstruct *qla as the Proto-(West-)Semitic form of the qal perfect for these roots, deriving it from earlier *qawala and *qayala with contracted triphthongs *awa and *aya. Geez, however, has forms like qoma he stood and ema he set, which cannot derive from *qma and *ma, but must come from *qaw(a)ma and *ay(a)ma, respectively. Therefore, the contraction cannot be a common development of all the West-Semitic languages, but rather a separate development. There are no objections to assuming that this contraction only took place in Hebrew once the CS had ceased to operate. The in tb, r and b may be analogical to the same vowel in the associated nominals that the verbs were patterned after; but given the existence of /bu/ to be ashamed in Akkadian, b may actually be a primary stative verb, deriving its vowel from a contraction of *bawua. The in yb he will come and other imperfect and imperative forms of b and b is best explained by reconstructing them with an *a as the theme vowel: *yabwau. As described in 3.1.9, Proto-Semitic *Cwa > *C, resulting in *yabu (a form paralleled by Arabic hollow alif verbs, e.g. yanmu he sleeps); with the CS and a few other developments, this regularly yields yb. Note the different vowel in the prefix of yb he will come vs. yb he will be ashamed; if the etymology described immediately above is correct, one of the two must be secondary. The most plausible possibility seems to be that yb retained the original form of the prefix, while yb was based on other stative verbs; Joon-Muraoka (2003:118) holds that yb is the outcome of BarthGinsbergs Law (the *a of the imperfect prefixes becomes *i in verbs with an *a-vowel in the stem), but this should have resulted in *yabtu > *yibtu > **yb (cf. *diru > zra), not yb. If, instead, yb reflects the original form, that means Barth-Ginsbergs Law operated after the Canaanite Shift had taken place. Bauer-Leander (1922:465) derives the qal participle of hollow roots from *qawilu, i.e. the stative participle (like *kabidu > kbd heavy) rather than the fientive participle (like *ktibu > ktb writing). This is not the form attested in Aramaic (Biblical: qm standing) or Arabic (kin being), nor should it normally yield ql, but rather ql (Bauer-Leander (1922) explain this with their mixed language theory: one stratum of preHebrew had developments like *gawiru > gr refugee, while the other had *gawiru > gr dwelling (participle of G-W-R)). Reconstructing these forms as stative *qatalu

36

adjectives (Fox 2003:162; cf. qtn small etc.), however, does yield the attested ql pattern, and although this is not the form reflected by Aramaic and Arabic, it is supported by evidence from Ugaritic: here, the participles of hollow roots usually have forms like <qm> standing, <bn> understanding, showcasing a contraction of *awa, *aya > *; *wi and *yi would have remained uncontracted (Tropper 2000:196). There are some Ugaritic forms like <qym> standing, probably /qyimu/, but the presence of the wrong semivowel, from an historical perspective, betrays this words secondary status. As evidenced by the perfect, contraction of triphthongs probably occurred after the Canaanite Shift, in Hebrew at least, resulting in the unchanged * still seen in qm standing ones of (construct state plural). The same development may have been shared by the more nominal forms n raw (~ Arabic na to be raw) < *nawau and b cloud < *gayabu from -Y-B to be dark27. The two forms with , bsim treading down and qmim standing, are probably based on analogy with the strong verb (qtlu : qtlim = qmu : x; x = qmim). Like the imperfect forms of b, the nipal perfect arises regularly from the expected strong form: *nakwana (cf. niqtal < *naqtala) > *nakna > nkn it is firmly established. The nipal imperfect, on the other hand, is problematic. The corresponding form of the strong verb, yiqqtl, goes back to *yanqatilu, as shown by Arabic yanfailu and the Akkadian N-stem preterite (cognate to the source of the Central Semitic imperfect) /ipparis/. For hollow roots, Arabic has forms like yanhru it will decline, collapse, which could regularly come from *yanqawilu28. As *awi is contracted to in BH (cf. *mawita > mt he died), *yankawinu should > **yikkn, yet the attested form is yikkn. Bauer-Leander (1922) explains this based on the idea that hollow roots were not really triradical, but had a long vowel as a part of their root. The * vowel of verbs of the Q--L type would then have spread through the paradigm of the nipal imperfect for hollow roots and then have changed to * due to the CS. This scenario seems extremely unlikely and does not fit the triradical model that the rest of this paper relies on. For lack of a better explanation, we will not use forms like yikkn as evidence of the CS. Finally, there is the question of the -- linking vowel between the stem and consonantal endings in the nipal and the hipil (shared by geminate roots, which also have it in the qal). This could possibly have originated in verbs III-W, e.g. *hayawta you lived > *hayta, and spread from there; but this seems unlikely, as no trace of this phenomenon is left in that category of verbs, and one would think that if a certain form is frequent enough to spread to other categories, it is also frequent enough not to be replaced itself. The only morpheme that resembles this linking vowel is indeed the // found before consonantal endings in the Akkadian stative. This linking vowel does then seem to have undergone the Canaanite Shift. 3.2.3 Verbs with a weak third radical Forms like bn he built probably go back to *qataya and *qatawa, cf. Geez bakaya to cry and fatawa to desire, the spelling of Arabic ban he built etc. with a form of the letter y that serves as a mater lectionis, the so-called alif maqsra (so for this verb,
27

Koehler-Baumgartner (1985): -W-B, but cf. Aramaic b cloud, Old South Arabian <yb>; no attested BH forms show W as a second radical. 28 No preterite N-stem verbs from hollow roots are attested in Akkadian.

37

<bny>), and the fact that Arabic III-Y verbs with an i-vowel in the perfect still retain their consonantal y, cf. nasiya to forget. Some scholars, like Diem (1975), support an alternative reconstruction, which derives most weak verbs from biconsonantal roots; the W/Y of III-WY roots would then be a secondary adaptation to the triconsonantal system. Diem (1975:31), however, cannot convincingly explain the origin of these extra radicals; he derives them from unattested forms in which they are supposed to have arisen naturally, e.g. *yarm-na they throw > *yarmiyna; but the actual attested forms are like Arabic yarmna. To prove that weak roots did not originally have a third radical, Diem thus ironically has to posit extra forms with a third radical. Another problem which is more relevant to the present paper is this reconstructions inability to explain the non-operation of the CS in forms like bn, if they come from *qat. Diem suggests Aramaic influence based on forms like rm he threw (1975:33), but one wonders why the Hebrew form was not then changed to **bn. Either way, it seems prudent to derive BH III-WY verbs from *qatawa and *qataya, whether this was the Proto-West-Semitic form of the perfect or a secondary development. In Ugaritic, forms both with and without contraction are attested, e.g. <ly> and <l> he went up (Tropper 2000:653 ff.), which supports a relatively late date for contraction. 3.2.4 Verbs I- and verbs III- That forms of these verbs where the * has elided, like ymar he will say and yims he will find, did originally contain * is highly probable, based on other forms of the same verbs where it is preserved, like mar he said and yimsu they will find. The CS did not occur in III- verbs; cf. ms and mr in 3.1.17. 3.2.5 - endings on verbs The proposed source of the - ending of the cohortative and long imperative, the subjunctive ending, is short -a in Arabic, e.g. an yaktaba that he write. See 3.3.1 directly below on the difficulties of reconstructing the length of word-final vowels; but regardless of its origin, the - must have become long at some point in pre-Hebrew, or it would have dropped off. This may have happened after the operation of the Canaanite Shift, however, as Amarna Canaanite still has word-final short vowels, but has already undergone the CS. 3.3 Other 3.3.1 Pronouns and personal suffixes The Akkadian first person singular pronoun, /anku/, shows that nki should be reconstructed with *. Besides ze this (m.) and zt this (f.), there is also a fairly rare byform, z (spelled <zh> and sometimes <zw>), which can be used for the feminine or neuter (i.e. no gender specified). As two other demonstrative pronouns, ze and the archaic zu which, this, are commonly related to various cases of the Arabic word d one of29, it seems plausible to derive z < *d, the accusative.

29

E.g. d l-qarnayn the one of the pair of horns, the horned one, an epithet of Alexander the Great or another ancient ruler; PCS nom. *d > BH zu, gen. *d > ze.

38

As for zt this (f.): although it is spelled <zt>, the words cognates show the lack of *; Ugaritic has <dt>, Geez has ztti, ztta (acc.), although these last forms might have dropped *, cf. Tropper (2002:33). Without external evidence of *, it is better to reconstruct the word as *dt- and to consider the spelling with <> to be secondary. The length of word-final vowels in some other pronouns and personal suffixes is hard to reconstruct. As an example, let us consider the second person suffixes, BH -k (m.) and -k (f.). As word-final short vowels dropped off in later stages of North-West Semitic, cf. Arabic kataba he wrote ~ BH ktab, these forms suggest that the masculine suffix should be reconstructed with a long vowel following the *k, while the feminine suffix had a short one (*i, based on cognate evidence). Intriguingly, this conflicts with evidence from other languages: Arabic -ka, -ki both have short vowels, Geez -ka < *-ka and -ki < *-k show a short vowel for the masculine and a long one for the feminine, while Tigria has reflexes of long vowels in both suffixes (Brockelmann 1908:I 309). A diagram clearly shows that all different options are attested: feminine *-ki *-k masculine *-ka Arabic Geez *-k Hebrew Tigria

A traditional explanation for this conundrum is that these vowels were anceps or semilong (cf. Brockelmann 1908:I 74). They then later became either long or short in individual languages. This variable nature of certain word-final vowels also receives support from language-internal evidence in BH: consider the vowelless suffixes in pausal forms like lk to you (m.sg.), or the fact that unlike other words ending in -, -k is not written with a <h> mater lectionis (contrast att, also supposedly anceps, but usually written <th>). Hasselbach (2004) attempts to derive all the relevant suffixes from forms with short vowels, but his haphazard invocation of analogy does not solve the problem satisfactorily. Whether these words (-k, the verbal suffix -t you (m.sg.), att you (m.sg.), third person singular feminine suffix -h) truly were in between long and short vowels or whether these are cases of long vowels being shortened or vice versa is hard to say, but we cannot be sure at all that any of these vowels was long when the Canaanite Shift occurred. 3.3.2 Other k here, now, thus, kk thus and k how show two different reflexes of *, if they correspond directly to Biblical Aramaic k here. The word-final -, however, may be the reflex of an anceps vowel (see 3.3.1 above), which is supported by the words similarity to the preposition k- like, cf. Arabic ka- with short a < *a. KoehlerBaumgartner (1985) convincingly explains k as the reflex of *ka-hu, the preposition *ka with the third person singular masculine suffix. kk may then not simply be a reduplicated form of *ka, but the same preposition with the second person singular masculine suffix, which also happened to be *-ka. Perhaps the two words originally reflected different nuances in meaning, *ka-hu meaning like him > like that, and *ka-

39

ka meaning like you > like this. Either way, we do not need to reconstruct * for these words. m what is paralleled by Aramaic and Arabic m, but as the - is word-final, this may be a case of an anceps vowel as well. The precative particle -n (spelled <-n>) and its independent form nn have no unambiguous cognates: Syriac n (<n> and <ny>) is said to be a Hebrew loanword by a Syriac author (cf. Sokoloff 2009), while Geez na (also naa and n) come has the wrong guttural, which is confirmed by other Ethiopic languages. The so-called h locale ending -, which serves as a marker of place or direction, is probably cognate to Ugaritic -<h> with the same function and should therefore be reconstructed as *-ah(V). There are no indications that it had *. Finally, l (<l>) no, not comes from *la, as shown by the dialectal Arabic form la (standard Arabic l with *-a > -). 4. Analogy and contamination Having considered which words and vowels patterns can confidently be traced to ProtoNorthwest Semitic forms with *, we should now consider the effects that analogy may have had on the dataset due to analogical retention of or spread of . Becker (1990) makes a distinction between proportional analogy and contamination, both of which are commonly referred to as analogy. Proportional analogy (henceforth: analogy) is the formation of new words based on a productive morphological rule. This can be exemplified by a formula expressing that the newly formed word has the same relationship to the word it is derived from as another word or set of words, as I have already done a few times (e.g. walk : walked = dream : x; x = dreamed); as Becker (1990) points out, however, this suggests that an analogy is based on a single example30, while quite often, it would be more accurate to describe the analogy as the application of a rule (e.g. /X/ VERB (present tense) /Xd/ VERB (past tense)). We will discuss the possible effects of analogy on the distribution of the CS in 4.1. Contamination, on the other hand, does not work with precise rules, but involves the assimilation of words to other words that are very closely associated to them semantically and often occur with them, mainly antonyms and adjacent members of series such as numbers. Some examples of this process from English are femelle >> female, based on male, and (PIE *kwetur >) **whour >> four, based on the following five (< *penkwe)31. We will deal with this in 4.2. 4.1 Analogy On page 66, Becker (1990) writes: Ein morphologisch mgliches Wort, das mit einem usuellen Wort in einer morphologischen Relation steht, ist eine mgliche Analogiebildung. This entails that the process by which the analogical form is produced must already exist in the language; analogy does not simply make words look like other words haphazardly. For this reason, I reject the notion that imperfect forms of hollow roots like yikkn it will be established have an in analogy to the perfect nkn it has been established (see 3.2.2); BH has no rule like /nX/ NIPAL PERFECT /yiCX/
30

Which does happen; cf. Elean Greek Zns Zeus (gen.) : Zes Zeus (nom.) = mns month (gen.) : x; x = mes month (nom.) (example from Becker 1990:18). 31 Latin underwent the opposite development, resulting in **pinque >> quinque based on quattuor.

40

NIPAL IMPERFECT, nor any other rule that would produce forms like yikkn. Similarly, the seemingly trivial fact that analogy can only be based on existing words excludes Bauer-Leander (1922)s explanation that the is based on an analogy with verbs with an in their root (i.e. verbs like b to come with the imperfect yb); none of these few verbs actually occur in the nipal, or have meanings that make that likely. What determines how productive a morphological rule is? For a large part, nonlinguistic factors (Becker 1990:115-6, also for following); but also language-internal factors, namely frequency and semantic transparency. Three kinds of frequency are involved: there is type frequency, i.e. to how many words the rule can be seen to apply; token frequency, i.e. how often examples of the rule occur individually (in the example of note 30, the rule has very low type frequency but very high token frequency, as the rule originally only applied to but one, very frequent word); and satiation, i.e. the ratio of possible to actually listed words (Becker 1990:115, citing Aronoff 1976:36), or to how many of the eligible words the rule has applied (in English, the rule /X/ NOUN (sg.) /Xs/ NOUN (pl.) has much higher satiation than /X/ NOUN (sg.) /X/ NOUN (pl.)). Semantic transparency is the degree in which the morphological rule results in a word with a predictable meaning; the rule that forms qattl patterns, for instance, has high semantic transparency, as it takes roots and turns them into profession nouns or adjectives denoting a habitual characteristic. Note that several rules with different levels of productivity can compete; cf. the coexistence of *midyn and mdn, both strife32. Now, suppose that the Canaanite Shift was conditioned by some phonetic factor. This conditioning could appear less clearly after time due to the workings of analogy. If a certain pattern in which the CS had occurred was productive enough, new words would be coined based on that pattern, resulting in forms that should not have undergone the CS and yet show its result. For this reason, we can only use evidence from productive patterns if the relevant datum applies to every attested token of the pattern; for instance, all mql nouns are stressed on the ultimate syllable, but they have different consonants preceding and following the . Therefore, the fact that *maqlu > mql does count as an indication that stressed * > at least some of the time, but the existence of mdn does not mean that * > after *d, as this individual form could have been formed secondarily after the CS was over. Let us examine various morphological rules that may have affected the reflexes of the Canaanite Shift. 4.1.1 Analogy in the verbal system In 3.2, I have argued to limit the reconstruction of * to a small number of verbal forms: the qal active participle qtl and inflected forms, the qal absolute infinitive qtl (and other absolute infinitives), the imperfect and imperative of b to come and b to be ashamed, the nipal perfect of hollow roots and forms with a linking vowel formed from hollow and geminate roots; the imperfect of I- verbs like ymar probably had * < *a at the time of the CS. The qal active participle qtl is highly productive; if there were ever forms that preserved *, they have all changed it to . The same goes for the qal absolute infinitive, and the other absolutie infinitives, which were probably formed analogically to this one
Respectively formed according to the rules /Q-T-L/ VERB (root) /miQTL/ ABSTRACT NOUN and /Q-Y-L/ VERB (root) /mQL/ ABSTRACT NOUN, both applied to the root D-Y-N.
32

41

(*qatala VERB (perfect) : *qatlu ABS.INF. = *naqtala, *qattila VERB (perfect) : x ABS.INF.; x = *naqtlu, *qattlu), either before or after the operation of the CS. The existence of b being ashamed (abs.inf.) : b he was ashamed : yb he will be ashamed probably allowed the analogical formation of r it was light and yr it will be light from r light, reinterpreted as an absolute infinitive. Besides the perfect of tb to be good, the ql pattern did not spread any further in the verbal paradigm (it is probably original in b to come, imperfect yb). The forms with -- before consonantal suffixes in the hollow and geminate roots have become regular, i.e. they have spread to all relevant verbs. As only a small number of the verbs with as their first radical show *ya- > yetc., those that do not have probably undergone analogical restoration of the *, based on the proportion *qatal VERB (perfect) : *yiqtul VERB (imperfect) = *atal VERB (perfect) : x VERB (imperfect); x = *yitul. 4.1.2 Plural formation Brockelmann (1908:I 77) attributes apparent exceptions to the CS to the leveling of paradigms that contained both * and *, based on whether the vowel had been stressed or not. In cases where * was generalized, this would be a case of rules like /X/ NOUN (sg.) /Xim/ NOUN (pl.) (sus horse : susim horses = dd dear : x dears; x = ddim); in cases where * was generalized, this would have happened according to rules /X/ NOUN (sg.) (susim horses : sus horse = tbim like /Xim/ NOUN (pl.) refugees : x refugee; x = tb). In BH as we know it, these kind of rules are not actually all that common, as most nouns undergo some change in their vocalism in the plural; but this is mainly due to vowel reduction, a fairly late development. Paradigmatic leveling may well have been an important force at the time right after the occurrence of the Canaanite Shift. Nevertheless, the two examples of an - alternation within the same word (ml left (hand), mli left, r head ~ rim heads) can both be explained differently (see 3.1.3 and 3.1.17)33. 4.1.3 mql /QL/ VERB (perfect, hollow root) /mQL/ ACTION or PLACE (cf. *hl to dance mhl dance for the first meaning, ln to spend the night mln place to spend the night for the second) seems to have been a fairly productive rule, both due to its frequency (although that may be the result of its productivity, rather than the source) and its high semantic transparency. Additionally, it only applied to a subset of verbs, which (perhaps counterintuitively) increases productivity too (Becker 1990:116). 4.1.4 maqtl and miqtl These patterns are less productive than miqtl and mql, as reflected by their lower frequency. They mainly form action and instrument nouns. 4.1.5 *qatalnu

Blau (2012:137) lists another example which I have not yet discussed, en human ~ anim men; but I think it much more likely that anim < *anama, not *unma, as is shown by the construct state, an < *anaay.

33

42

Two productive patterns are the BH reflexes of *qatalnu: qtn for roots with a weak second (properly qln) or third radical34 and qittln for all roots (not just strong roots, cf. hizzyn vision from H-Z-Y to see). These patterns mainly form abstract nouns and are quite frequent. 4.1.6 *qitlnu, *qutlnu Although few of these forms are attested in Hebrew, their high semantic transparency (all of the words are abstract nouns) may point to a period of productivity. 4.1.7 *-nu With its various meanings, the -n suffix was highly productive, as shown by the large number of words that bear it. The -n version of the suffix was rather less productive. 4.1.8 *-tu (feminine plural ending) The -t ending is found on the plural of practically every feminine word and even on some masculine nouns, like bt fathers. As an inflectional ending, it is highly susceptible to analogy. 4.2 Contamination A few forms in our dataset may show the effects of contamination. In some cases, I have already noted this in 3.1: qann jealous may have been contaminated by qann idem, abbtn sabbatical is probably contaminated by abbt sabbath; liwytn Leviathan may have been contaminated by the unattested cognate of Ugaritic /ltnu/ idem; and it is hard to say which development came first, *linu tongue > *lanu or *girnu throat > *garnu, but whichever of the two contiguous body parts shifted last probably did so due to contamination by the other one. An interesting conundrum is posed by, once again, mli left, which has the same vowel pattern as ymni right. Neither is the expected form, based on their corresponding nouns, ml and ymin; both can be explained by cognate forms in Arabic, amal < *amalu and yaman < *yamanu. Barth (1906:794) thinks that ymni was contaminated by mli, rather than the other way around, because there are two cases of a ktib <ymyny>, i.e. the expected **ymini. Additionally, the explanation given in this paper requires mli to come from *imaliyyu to account for the <> in both this word and ml. No other cases of contamination seem to have affected the relevant words. 5. Phonetics of the Canaanite Shift As we are treating the Canaanite Shift as a phonetically motivated sound change, the precise phonetics of the shift deserve to be discussed. We shall consider the two main factors that could effect a qualitative change from [a:] to [o:] or similar: stress, and the
Originally from roots with a weak third radical, like *hazaynu > *hazaynu > hzn vision; roots with a weak second radical should have a long * in the first syllable due to the contraction of *awa and *aya, but -W- to rejoice n joy, construct state n for **n shows that qln has merged with qtn analogically, based on the identical form of the absolute state singular. Brockelmann (1908:I 77) attributes this to a sound law instead, stating that in a sequence of two long vowels, the first one becomes short, i.e. *awan > *n > *an > n, n.
34

43

place of articulation of nearby vowels and adjacent consonants (M. van Oostendorp, personal communication). 5.1 Stress There is a tendency for rounded vowels to be restricted to stressed syllables, which generally allow more different vowels than unstressed syllables to begin with. In Russian, for instance, [o] can only occur in stressed syllables; in pretonic and word-initial syllables, /o/ is reduced to [], elsewhere to [], thereby merging with /a/ (Pugh 2007:30). Note that this process, known as akane, involves the development of historical unstressed *o to an a-vowel, not stressed *a to an o-vowel. Additionally, see 5.2.1 for an example of a stressconditioned shift of an a-vowel to an o-vowel from Huave. 5.2 Phonetic environment A shift from [a] to [o] could also be the result of coarticulation effects. When transitioning between different phonemes, the articulatory apparatus has to coordinate various different movements, and often, this results in articulatory features of some phonemes carrying over to adjacent sounds. As [o] is a back, rounded vowel, the presence of these kinds of vowels, consonants that are articulated far back in the mouth or labial consonants could make an /a/ sound [o]-like. 5.2.1 Vowels Vowel assimilation, often called vowel harmony or umlaut, can be caused both by preceding or by following vowels. In Old Icelandic (Gutenbrunner 1951:57), both long and short *a > [] (preserving its length) if the following syllable contained *u > closed o. This so-called uumlaut is clearly illustrated by the first person plural suffix -om < *-um: kllom we call < *kallum, cf. kalla to call, tom we ate < *tum, cf. t he ate ( is long [a:]). In Saami, we see that a-vowels are rounded by a preceding rounded vowel:
Lule Saami short stressed "mid" round vowels trigger rounding of an immediately following short or long low vowel. Rounding harmony does not occur elsewhere. E.g. dll+a surfaces as 'dll (using an orthography-based transcription) "fire-sg.nom." and dl+a surfaces as 'dl "fire-pl.nom." (predictable vowel lengthening in this context, but it also happens with underlying long low vowels). Although most of the stressed vowels that trigger this harmony are in the first syllable (which carries main stress), there are examples of non-initial stressed syllables triggering rounding harmony. E.g. telefvnn+a surfaces as 'tele,fvnn "telephone-sg.nom.". The round "" does not occur in post-stressed syllables otherwise. Note that what is spelled and described as "u" does not trigger rounding, nor does the "uo" diphthong. E.g. ull+a = 'ulla "gutter-sg.nom", guoll+a = 'guolla "testicle-sg.nom. (B. Morn-Duollja, p.c.)

The fact that u and uo do not trigger rounding may be some form of dissimilation, cf. 7.2.1 and Ohala (1993).

44

An interesting case where only stressed vowels participate in rounding due to vowel harmony is found in a dialect of Huave:
There's a claim about the Huave dialect of San Mateo del Mar that rounding harmony skips unstressed syllables. Stress is final, and supposedly you get e.g. ta-mOng-Os 'I passed' but t-a-mOng-As-On 'we (excl.) passed', where caps = participating vowels w[ith] the first O as the trigger in both cases. (Yuni Kim, p.c.)

5.2.2 Consonants Labial or velar consonants which are directly adjacent to an [a] can also pull it back and round it (M. van Oostendorp, p.c.). For an example from Semitic, consider the quality of /a/ and /a:/ in Arabic, which is backed before and after emphatic consonants (including /q/) and before /r/, but fronted elsewhere: qalb heart and nr fire have [] [:], kalb dog and ns people have [] [:] (Ryding 2005:26). Similarly, palatal consonants may prevent an [a] from being backed (M. van Oostendorp, p.c.). 6. * > in other languages Having considered phonetic conditions which can cause an a-vowel to move back and be rounded, we will try to establish what would be normal conditions for the Canaanite Shift by taking a look at a more or less random sample of three other examples of a shift of * to : the Germanic languages (Indo-European), Welsh (Celtic, also Indo-European) and rural Maltese (Semitic). Unfortunately, all of these languages are European, but this can hardly be helped due to the relatively poor availability of historical resources concerning non-European (and non-Middle-Eastern) languages. I have purposefully excluded cases from the area where the CS operated, such as the > shift in Western Syriac, to avoid any chance of some lingering after-effect of the CS skewing the results. 6.1 Germanic At some point between Proto-Indo-European and Proto-Germanic, the precursor of the various Germanic languages underwent the development (*eh2 35 >) * > * in all environments, stressed or unstressed (Voyles 1992:48); cf. PIE *bhh2geh2 beech > PGm *bk, PIE *ghbheh2s gift (gen.) > PGm *gebz. In later Germanic languages (i.e. all but Gothic), unstressed * > * > [u] > [] (> in most languages). In many languages, stressed * developed further to [u], cf. Dutch broe(de)r, German Bruder brother < PIE *bhreh2tr. 6.2 Welsh While PIE *eh2 shows up as * in Proto-Celtic, cf. Irish brathair brother < PIE *bhreh2tr, it consequently shifts to o and aw in Welsh, a member of the Brythonic subgroup of the Celtic languages. This development must be fairly late, as it also affected Latin loanwords. Morris Jones (1913:94) shows a diagram similar to the following:

As is customary in the Leiden Indo-Europeanist tradition, I will reconstruct Late Proto-Indo-European * as coming from *eh2, *h2 being one of the so-called laryngeals, consonants of an uncertain phonetic realization.

35

45

Bryth. (Lat.) *brteres brothers trnittem trinity (acc.), Lat. *brton judgement

Bryth. Early Welsh

Old Welsh o

Middle Welsh o

Modern Welsh o o

Modern Welsh broder trindod

(environment) penultimate syllable ultimate syllable monosyllable

[] au aw

aw

brawd

As can be seen from the table, all instances of Brythonic * shifted to Early Welsh []. In Old Welsh, it became closed o in penultimate (i.e. unstressed) syllables, while breaking to au in ultimate syllables and monosyllabic words (i.e. when stressed). Later, this au > aw developed to o as well, except in monosyllables and following *w (ibid.). 6.3 Rural Maltese36 Although Maltese has undergone a lot of influence from various Romance languages, it is still quite clearly an offshoot of Arabic and therefore a Semitic language. Where many contemporary Arabic dialects feature a phenomenon known as imla, where // is pronounced as [e:] or another front vowel, Standard Maltese preserves the pronunciation [a:]. Rural dialects of Maltese go even further in the other direction. As reported in Schabert (1976), practically all rural dialects of Maltese have shifted * backwards, with realizations ranging from [:] to [u:], a pronunciation feature which is known as imm (p. 4); cf. /dr/ house < Arabic dr, /b/ he brought < Arabic jb. Most dialects that feature imm also show the raising and backing of Standard Maltese /o/ to [] or [u] and /a/ to [], as well as the diphthongization of // to [w] or [w] and // to [j]. Imm only occurs in stressed syllables, but this is because in Maltese, long vowels are limited to stressed syllables; they were shortened elsewhere, cf. alternations like /hfy/ barefoot (sg.) ~ /hafyyn/ barefoot (pl.). The shift also affected *a that was secondarily lengthened: /spi/ it ended < Italian spiccia, but /ma-spi-/ it did not end. Sometimes, however, /a/ and // are lengthened to //, as in /rayna/ we saw, /ma-rayn-/ we did not see. Schabert (1976) can not find any phonetic conditioning for this interchange between // and // and says it is morphologically conditioned. 6.4 Summary In two of the three examined languages (or language groups), we have seen no phonetic conditioning for the * > shift: all instances of * originally shifted to * in ProtoGermanic, and although there are some exceptions to * > // in Rural Maltese, these are not phonetically conditioned. In Welsh, all instances of Brythonic * shifted to [], but this phoneme developed two allophones later on, becoming [o] when unstressed and [au]
I maintain Schabert (1976)s phonemic transcriptions, but have adapted his phonetic symbols to IPA and have maintained the Arabic transcription used elsewhere in this paper; e.g., Maltese // is realized as [d] = Arabic j.
36

46

when stressed. Later, we see that a preceding w maintained this [aw] < [au], a case of dissimilation blocking or reversing sound change, while [aw] shifted on to [o] in other cases, except in monosyllables. Welsh, therefore, does not exhibit a true stressconditioned change of vowel quality; rather, the stress causes and later preserves diphthongization 37 , as does the presence of a consonant with a similar place of articulation. 7. Conclusions and discussion 7.1 *a$ > * Words which contained the Proto-Northwest-Semitic sequence *a$, i.e. a closed syllable containing an *a as its nucleus and a * as its coda, show the same behaviour as words with *. Hence, it follows that *a$ > *, probably after Proto-Northwest-Semitic, as Northwest Semitic neighbour Ugaritic has not undergone it, but before the operation of the Canaanite Shift. Note that it must also predate the apocopation of final short vowels, as word-final *aV > , cf. *yaqrau > *yiqra > yiqr he will call (the is the result of a second, post-CS round of *a$ > *). In the perfect of the verb, the * was restored before consonantal suffixes, based on the morphological rule */QaTaLa/ 3SG PERFECT */QaTaLti/ 1SG PERFECT. The // of Phoenician /qarti/ does not contradict this, as it is the result of the Phoenician Shift. 7.2 Conditions of the Canaanite Shift Of the words listed in 2.1 38 that have fairly certain reconstructions with *, whether original or from *a$, the great majority show as that sounds reflex. As shown by *la > l no(t) and *d > z this, the CS also occurred word-finally. Of the words that do not have , only yim shovels has in a position that would have been unstressed in earlier stages of pre-Hebrew; while the tendency for only stressed vowels to be backed and rounded (see 5.1) does support stress-based conditioning of the Canaanite Shift, there is no real evidence to back it up. Upon closer examination, the small group of words that show (*a$ >) * > can be divided into two groups: words with back vowels and words in which the * follows *y. Note that no members of unproductive word patterns that either confirm or contradict these conditions are attested in other Canaanite languages. 7.2.1 Words with back vowels As already noted by Bergstrsser (1918), the Canaanite Shift does not occur after back vowels. While I agree that qorbn sacrifice is an example of this category, Bergstrssers other example, ulhn table, has an unclear etymology and might come from either *tulhnu or *tulhanu. Additional examples that Bergstrsser does not mention, however, can be found in nhutn serpent idol and two or three words with < *aw: tb resident alien, ms exit if this comes from *mawsu and possibly *ltn Leviathan, the expected cognate of Ugaritic <ltn> and, as was suggested in 3.1.13, the origin of the -n ending in BH liwytn. The fact that tb retained its * shows that the

37

Cf. pre-Hebrew *tawk middle > BH twek (absolute state, i.e. stressed), tk (construct state, i.e. unstressed). 38 Of the Canaanite languages, only BH has attested forms with * > , so the others cannot contribute all that much to a discussion of the conditions of the CS.

47

Canaanite Shift was predated by the contraction of diphthongs, at least in some environments. The non-operation of the CS in these words is probably a form of dissimilation: the [o:] (or similar) in *[qurbo:nu] etc., the normal reflex of these words, was interpreted as an // that had been assimilated to the preceding back rounded vowel, leading to the restoration of [a:] in careful speech (cf. Ohala 1993). 7.2.2 Words with *y In all the four remaining nouns with < * (yim shovels, binyn building, inyn task and qinyn possession), the long vowel is preceded by y. While this might be coincidental, due to the small number of cases, non-occurrence of the Canaanite Shift after *y is plausible, as the palatal place of articulation of the consonant may have kept the * from shifting back; cf. 5.2.2. While this condition seems to hold for the nouns, there are plenty of unproductive verb forms with y, namely the imperfect forms of I- verbs in which the * was not analogically restored (2.1.3.2). If * was preserved after *y, these verbs should have forms like **ymar, not ymar he will say. Interestingly, there is the attested pausal form wayysel and he took away from -S-L, mentioned in 2.1.4.2. This form cannot have arisen by analogy39, and is therefore probably original. The more common y- forms, however, can: paradigm pressure may have introduced the based on the other prefixes (*aqtul, *taqtul, *naqtul 1SG, 2SG, 1PL IMPERFECT : *yaqtul 3SG IMPERFECT = *mar, *tmar, *nmar 1SG, 2SG, 1PL IMPERFECT : x 3SG IMPERFECT; x = *ymar). 7.2.3 Summary: formulation of the Canaanite Shift Bearing the above in mind, the following three rules (plus one) describe the regular outcome of the Canaanite Shift: (0. Proto-Northwest-Semitic *a > pre-Proto-Canaanite * before consonants and wordfinally.) 1. Pre-Proto-Canaanite * > Proto-Canaanite * if the previous syllable contains a back rounded vowel. 2. Pre-Proto-Canaanite * > Proto-Canaanite * after *y. 3. Pre-Proto-Canaanite * > Proto-Canaanite * elsewhere. 7.3 Dating the Canaanite Shift 7.3.1 Relative chronology We have seen that the CS interacts with several other Canaanite and Hebrew sound changes. Based on this interaction, they can be arranged into a relative chronology, ordered from the earliest to the latest development. Developments that cannot be ordered relative to one another are given the same number and a different lowercase letter. Note that 4a could have preceded 3b, and 5a could have preceded 4b.

39 Nor is it likely that this form is the result of a later elision of * in closed syllables, i.e. *yasilu > *ysil >> *yasil > (way)ysel, like *qarati > qrti I called: when the * in the prefix of the imperfect was restored, the prefix vowel was *i, like in the strong verb; cf. yesr he will bind < *yisur, not **yasr < *yasur.

48

?a. Proto-Semitic *w becomes Proto-Northwest-Semitic *y word-initially. To my knowledge, no instances of Proto-Semitic *w or *wa$ are attested in Canaanite (besides members of productive categories), so we cannot say whether this change preceded or followed the CS. 1a. *a > * before consonants and word-finally. Predates 2: *la no(t) > *l > *l (2). Predates 4b: *yiqrau he will call (stays *yiqrau under 1a, 2) > *yiqra (4b) > *yiqr (5b). 1b. Contraction of some diphthongs (in unstressed syllables?): *aw > *, *ay > *. Predates 2: *tawbu resident alien > *tbu (stays *tbu under 2). Predates 4a: *qawamay standing ones of > *qawam40 > *qm (3a). 2. Canaanite Shift. Postdates 1a, 1b, 2. Predates 3a: *banaya he built (stays *banaya under 2) > *ban (3a). Predates 3b: *atta (stays *atta under 2) > *att (3b). 3a. Contraction of triphthongs: *aWa > *, *aWu > *, *aWi > *. Postdates 2. Predates 4a: *qawam standing ones of > *qm before elision of non-pretonic unstressed *a. Predates 4b: *banaya > *ban before loss of final *a. 3b. Lengthening of anceps vowels that are retained in BH. Postdates 2. Predates 4b: *atta > *att (stays *att under 4b). 4a: Loss of non-pretonic unstressed *a. Postdates 3a. Predates 5a: *raam heads > *ram > *rm (5a). Predates 5b: *raa heads of > *ra > *r (5b). 4b. Loss of final short vowels. Postdates 1a, 3a, 3b. Predates 5b: *yiqrau > *yiqra > *yiqr (5b). 5a. Elision of * between a consonant and a vowel. Postdates 4a. 5b. *a > * before consonants and word-finally. Postdates 4a, 4b. ?b. Barth-Ginsbergs Law: *a > *i in imperfect prefixes of verbs with *a as their root vowel. Postdates 2: *yabu > *yabu (2) (stays *yabu under Barth-Ginsbergs Law). To illustrate this chronology more clearly, the following table shows the development of several different words.

40

I.e. not > *qawmay > *qm, which would be the case if 4a preceded 1b.

49

ProtoNWS 1. 2. 3. 4. 5.

*ddu *tawbu *la *yiqrau *qawamay *raaay *raama *atta *ddu *ddu *ddu *dd *dd *tbu *tbu *tbu *tb *tb *l *l *l *l *l *yiqrau *yiqrau *yiqrau *yiqra *yiqr *qawam *qawam *qm *qm *qm *raa *raa *raa *ra *r *raama *raama *raama *ram *rm *atta *atta *att *att *att

7.3.2 Absolute chronology Having established the relative chronology of the Canaanite Shift and several related sound changes, we can try to date the CS more precisely: how many years ago did it take place? For this, we largely rely on written evidence, as it links certain linguistic stages to buildings and artefacts that can be dated using other means. The oldest written evidence of the Canaanite Shift is not to be found in the Amarna Letters, but in slightly older texts: Egyptian texts from the New Kingdom contain many Semitic loanwords, and the first instances of words that have undergone the CS are attested from the 15th century onwards (Hoch 1994:423). Albright (1954) reads the 18th-century name <tw-ti-wt> (in an Egyptian list of slaves) as /ddhuat(u)/ he is my beloved/uncle, but it is not clear whether Egyptian <w> could be used as a mater lectionis by this time already. According to the relative chronology established above, the elision of * in coda position preceded the Canaanite Shift. Burchardt (1910:I 53) states that Egyptian transcriptions of Canaanite words, which are less concerned with historical spelling, show the first signs of *-elision in place names attested in eighteenth-dynasty texts, i.e. around the same time as the first attested examples of the CS. It would seem, then, that the elision of * in closed syllables and the CS occurred in fairly rapid succession, in the fifteenth century BCE. Incidentally, Kitchen et al. (2009) arrives at a similar date based on a computational linguistic approach, situating the last shared ancestor between Hebrew and Aramaic (i.e. the split between Canaanite and Aramaic) around 1500 BCE. 7.4 Canaanite, Northwest Semitic and the family tree model If the dating of the Canaanite Shift to the 15th century BCE that was reached in 7.3.2 is correct, it has some interesting consequences for the classification of the Northwest Semitic languages. Traditionally, they have been grouped into a family tree, consisting of the Canaanite and Aramaic branches, with Ugaritic being thrown in with either one or constituting a third branch. Canaanite and Aramaic then branch out further, eventually ending in the various languages and dialects that make up these families. The idea of a family tree is that different languages and speech varieties can be traced back to one single ancestor. This proto-language can then be connected to related families, which all descend from another, earlier single ancestor, and so on, until the limits of the historical method have been reached. In this model, no language can inherit any feature that was not present in its ancestral language (although it can, of course, develop such features). For example: if the Canaanite Shift had already occurred in ProtoNorthwest-Semitic, Ugaritic and Aramaic could not have inherited the Proto-Semitic *

50

that the CS turned into *; but they did, so the CS must have happened at a more recent node in the family tree. The family tree model works best when describing languages that are separated from one another, usually due to migration. This is due to something known as the founder effect: when a sample from a larger population migrates to a new area, they only take a small part of the diversity that is present in the entire population with them, and only that small pool can be passed on to their descendants (Mayr 1942:237). In linguistics, the founder effect results in low diversity in the speech of settlers of a new area, typically the situation in which a new branch of the family tree splits off. Once a speech community stays in the same place for a longer period of time, however, it diversifies into different dialects and drawing neat family trees becomes more difficult. Such a dialectal landscape is more adequately represented by a wave model, a map that characterizes the various dialects by features they share, drawing so-called isoglosses between dialects that differ in a certain respect. To what extent can we speak of a Canaanite language family? To bear any meaning, the term should refer to a group of languages that share more than one feature. Otherwise, saying a certain language is Canaanite is just shorthand for saying that it underwent the Canaanite Shift. Now, there are other features that are generally held to be common Canaanite developments, such as the use of the definite article <h> (BH ha-:) and the consonantal sound change *t, *d, *t > *, *z, *s. Of these three changes, only the use of the article <h> is unambiguously a feature of the less-attested Canaanite languages, Moabite, Ammonite, and Edomite. No plene spellings of // < * are attested, and the graphemes <>, <z> and <s> may also represent the interdental fricatives, as in Old Aramaic (Hug 1993:51). May we assume that these less-attested languages have undergone the CS and the shift of interdentals to (post)alveolars, simply because we know they are Canaanite languages on account of their use of <h> and their geographic location? It depends on whether we are dealing with a family tree situation or a wave situation. If the Proto-Canaanites arrived in Canaan as immigrants from somewhere else, perhaps from the Proto-Semitic or Proto-Northwest-Semitic Urheimat, then we can safely assume that changes that affected several different branches of Canaanite (in this case, Hebrew and Phoenician) are shared features of the entire family. If the Canaanite languages acquired their typically Canaanite features in situ, however, it is quite possible that the wave of the Canaanite Shift spread to different dialects than the wave of the interdental-(post)alveolar shift, overlapping in some dialects but not in others. To answer this question, we must make a very brief excursion into archaeology. Although it can be hard to link archaeological remains to particular ethnic or linguistic groups, there seems to be a tentative consensus about the advent of the Semites in the Levant. In the Southern Levant, i.e. the biblical land of Canaan, a new culture seems to have emerged at the beginning of the EB I (Early Bronze Age I) period; Richard (2003:287) calls this (proto-)Canaanite culture. As the EB I period started around 3500 BCE (Dever 2003), it is somewhat anachronistic to call this culture Canaanite, as the characteristic developments of the Canaanite languages would probably not yet take place for another two millennia (see 7.3.2); but this does appear to be the first period in which a Semitic population settled in the area.

51

Further to the north, in the area surrounding Ugarit, the first evidence for Semitic settlement comes from the late 4th millennium (Von Reden 1992:47-9). As the core area of the various Northwest Semitic languages was thus reached by Semitic speakers around the same time, it would seem that this was the result of one migratory movement. The diversification into Ugaritic, Canaanite and Aramaic would then have been a gradual process, not an abrupt branching off from the family tree. This wave model history of the Northwest Semitic languages receives some support from the linguistic finds at Deir Alla (Beyer 2012:123 ff.), and other inscriptions that have many features from both Canaanite and Aramaic, a phenomenon which is hard to square with a family tree modelbased approach. In conclusion, the concept of the Canaanite language family is in need of some serious reconsideration. While the Canaanite shift is attested for Amarna Canaanite, Hebrew, and Phoenician, and the shift of interdentals to (post)alveolar sibilants is attested for the latter two, determining whether other languages and dialects participated in these and other Canaanite developments must be done on a case-for-case basis. 7.5 Topics for future research In our investigation of the Canaanite Shift, some problems have come up that have not yet been solved. I will briefly discuss them here, as well as possible extensions of the methodological approach I have used to related topics. 7.5.1 yikkn etc. The in the nipal imperfect forms of hollow roots, like yikkn it will be established, has not received a satisfactory explanation in this paper. As it is directly related to the question of the reconstruction of hollow and other weak roots, finding a convincing etymology of this verbal form may well shed new light on the Proto-Semitic morphological system. 7.5.2 Non-family tree-based relationships between Semitic languages As argued in 7.4, the family tree model is not the most adequate approach to describing the relationship between the different Northwest Semitic languages. Applying the wave model to other parts of the Semitic family might help resolve some controversies surrounding its subdivision, such as the status of Arabic and Old South Arabian, which are usually either grouped with Northwest Semitic as Central Semitic, or with Ethiopic and Modern South Arabian as South Semitic; Huehnergard-Rubin (2011) examines exactly this problem from a hybrid family tree-wave model, while Garr (1985) takes a similar approach to the Northwest Semitic dialects. A clearer picture of the various relationships between the Semitic languages would also have implications for our understanding of the prehistory of the various Semitic and Afro-Asiatic peoples. 7.5.3 A stricter use of analogy In this paper, I have used the rather rigid definition of analogy explained in 4.1. This stricter use of analogy has resulted in the problematization of verbal forms like yikkn (3.2.2, 7.5.1) and the necessity of finding a set of sound laws that regularly yield the attested forms of II-WY and III-WY verbs (3.2.2, 3.2.3). As analogy is often invoked as a cure-all whenever a reconstruction or posited sound law does not match the attested

52

forms, I propose that only those analogies that are based on a clear morphological rule be accepted as having explanatory power. Often, there are two or more competing sets of reconstructions or sound laws that aim to explain the same forms which both heavily use analogy to explain away the deviations from their predicted forms; a more restricted use of analogy might show which of these competing theories could actually be right and which cannot.

53

Acknowledgements I would like to express my sincere gratitude to my supervisor, prof. Gzella, for his enthousiastic and extremely helpful guidance on the researching and writing of this paper; to prof. van Oostendorp and his colleagues, for their essential contribution to section 5; to drs. Ronald Kon for answering my questions on Arabic lexicography; to my (as of yet unknown) second reader; and finally, to my girlfriend, parents, housemates and all my other friends for their interest, critical questions and moral support. Thank you. References ALBRIGHT, W.F. 1954: Northwest-Semitic names in a list of Egyptian slaves from the eighteenth century B.C., in Journal of the American Oriental Society, volume 74, no. 4, pp. 222-233 1976: Word formation in generative grammar, MIT Press, Cambridge, Mass. 1906: Formangleichung bei begrifflichen Korrespondenzen, in Bezold, C. (ed.), Orientalische Studien, Theodor Nldeke zum siebzigsten Geburtstag, Alfred Tpelmann, Gieszen, pp. 787-796. 1922: Historische Grammatik der hebrischen Sprache des Alten Testaments, Max Niemeyer, Halle 1927: Grammatik des Biblisch-Aramischen, Max Niemeyer, Halle BECKER, TH. BEN-HAYYIM, Z. 1990: Analogie und morphologische Theorie, Wilhelm Fink, Mnchen 1977: ibrit waarmit nsh mrn, The Academy of the Hebrew Language, Jerusalem 2000: A Grammar of Samaritan Hebrew, Eisenbrauns, Winona Lake BERGSTRSSER, G. 1918: Hebrische Grammatik, 1. Tl., Einleitung, Schrift- und Lautlehre, Vogel, Leipzig 2004: Die aramischen Texte vom Toten Meer samt den Inschriften aus Palstina, dem Testament Levis aus der Kairoer Genisa, der Fastenrolle und den alten Talmudischen Zitaten, Vandenhoek & Ruprecht, Gttingen

ARONOFF, M.

BARTH, J.

BAUER, H. and LEANDER, P.

BEYER, K.

54

2012: The Languages of Transjordan, in Gzella, H. (ed.), Languages from the World of the Bible, de Gruyter, Berlin, pp. 111-127 BIRKELAND, H. BLAU, J. 1940: Akzent und Vokalismus im Althebrischen, Dybwad, Oslo 1993: A grammar of Biblical Hebrew, Harassowitz, Wiesbaden 2010: Phonology and Morphology Eisenbrauns, Winona Lake BURCHARDT, M. of Biblical Hebrew,

1910: Die altkanaanischen Fremdworte und Eigennamen im gyptischen, Hinrichs, Leipzig 1908: Grundriss der vergleichenden Grammatik der semitischen Sprachen, Reuter & Reichard, Berlin 1943: Studien ber hebrische Morphologie und Vokalismus, Brockhaus, Leipzig

BROCKELMANN, C.

BRNNO, E.

CAD: see Gelb (1959) CROSS, F.M. Jr. and FREEDMAN, D.N. 1975: Studies in Ancient Yahwistic Poetry, Scholars Press, Missoula

DEL OLMO LETE, G., 2003: A dictionary of the Ugaritic language in the alphabetical and SANMARTN, J. tradition, Brill, Leiden DEVER, W.G. 2003: Chronology of the Southern Levant, in Richards, S. (ed.), Near Eastern Archaeology, A Reader, Eisenbrauns, Winona Lake, pp. 82-87 1975: Die Verba und Nomina tertiae infirmae im Semitischen, in Zeitschrift der Deutschen morgenlndischen Gesellschaft 127, pp. 15-60 1999: From Proto-Semitic to Hebrew, Centro Studi CamitoSemitici, Milan 1962- Kanaanische und aramische Inschriften, Harassowitz, 1969: Wiesbaden 1948: Semitic writing: from pictograph to alphabet, Oxford University Press, London 1989: qnqrdansy had ltr, nbiim uktubim, Kiryat

DIEM, W.

DOLGOPOLSKY, A.

DONNER, H., and RLLIG, W. DRIVER, G.R.

EBEN-ON, .

55

Sper, Jerusalem EJ: see Skolnik-Berenbaum (2007) FRIEDRICH, J. and RLLIG, W. 1999: Phnisch-Punische Grammatik, Pontificium Institutum Biblicum, Rome

FOX, J. GARR, W.R.

2003: Semitic noun patterns, Eisenbrauns, Winona Lake 1985: Dialect geography of Syria-Palestine, 1000-586 B.C.E., University of Philadelphia Press, Philadelphia

GELB, I.J. et.al. (edd.) 1959- The Assyrian Dictionary, University of Chicago Oriental 2010: Institute, Chicago GOSHEN-GOTTSTEIN, 1979: The Aleppo Codex and the Rise of the Massoretic Bible M.H. Text, in The Biblical Archaeologist 42, pp. 145-163 GUTENBRUNNER, S. 1951: Historische Laut- und Formenlehre des Altislndischen, Carl Winter, Heidelberg 2010: So-Called poel-Forms in Isaiah and Elsewhere, in Van der Meer e.a. (edd.), Isaiah in Context, Brill, Leiden, pp. 63-81 2012: Mimation, in Khan, G. (ed.), Encyclopedia of Hebrew Language and Linguistics, Brill, Leiden (forthcoming) HASSELBACH, R. 2004: Final vowels of pronominal suffixes and independent personal pronouns in Semitic, in Journal of Semitic Studies 49, pp. 1-20 1994: Semitic Words in Egyptian Texts of the New Kingdom and Third Intermediate Period, Princeton University Press, Princeton 1987: Ugaritic vocabulary in syllabic transcription, Scholars, Atlanta 1992: Historical Phonology and the Hebrew Piel, in Bodine, W.R. (ed.), Linguistics and Biblical Hebrew, Eisenbrauns, Winona Lake, pp. 209-230 HUEHNERGARD, J., and RUBIN, A.D. 2011: Phyla and Waves: Models of Classification of the Semitic Languages, in Weninger, S. (ed.), The Semitic Languages, An International Handbook, de Gruyter, Berlin, pp. 259-

GZELLA, H.

HOCH, J.E.

HUEHNERGARD, J.

56

278 HUG, V. 1993: Altaramische Grammatik der Texte des 7. und 6. Jh.s v. Chr., Heidelberger Orientverlag, Heidelberg 1972: The Neogrammarians, a re-evaluation of their place in the development of linguistic science, Mouton, The Hague 1953: Das Wort lm im Alten Testament, Alfred Tpelmann, Berlin 2009: A grammar of biblical Hebrew, Gregorian & Biblical Press, Rome

JANKOWSKY, K.R.

JENNI, E.

JOON, P. and MURAOKA, T.

KAI: see Donner-Rllig (1962) KITCHEN, A. et al. 2009: Bayesian phylogenetic analysis of Semitic languages identifies an Early Bronze Age origin of Semitic in the Near East, in Proceedings of the Royal Society B: Biological Sciences 276, pp. 2703-2710 1987: A comprehensive etymological dictionary of the Hebrew language for readers of English, Macmillian Publishing Company, New York 1915: Die El-Amarna-Tafeln, Hinrichs, Leipzig 1985: Lexicon in veteris testamenti libros, Brill, Leiden

KLEIN, E.

KNUDTZON, J.A. KOEHLER, L. and BAUMGARTNER, W.

KRAHMALKOV, C.R. 2000: Phoenician-Punic Dictionary, Peeters, Leuven KHNE, H. 1974: Mit Glossenkeilen markierte fremde Wrter in akkadischen Ugarittexten, in Ugarit-Forschungen 6, pp. 157-168 1982: A history of the Hebrew language, Brill, Leiden 1863- An Arabic-English lexicon, Williams and Norgate, London 1893: 1987: Comparative Dictionary of Geez, Harassowitz, Wiesbaden 1975: A Latin dictionary, Clarendon Press, Oxford

KUTSCHER, E.Y. LANE, E.W.

LESLAU, W. LEWIS, C.TH. and SHORT, C. MAYR, E.

1942: Systematics and the Origin of Species, Columbia University

57

Press, New York MEYER, R. 1966- Hebrische Grammatik, De Gruyter, Berlin 1972 1913: A Welsh Grammar, historical and comparative, Clarendon, Oxford 1880: Prinzipien der Sprachgeschichte, Max Niemeyer, Halle 2007: A New Historical Grammar of the East Slavic Languages, Volume 1: Introduction and Phonology, Lincom, Munich 2003: The Early Bronze Age in the Southern Levant, in Richard, S. (ed.), Near Eastern Archaeology, A Reader, Eisenbrauns, Winona Lake, pp. 286-302 2005: A Reference Grammar of Modern Standard Arabic, Cambridge University Press, Cambridge 1993: The phonetics of sound change, in Downing, P. e.a. (edd.), Historical Linguistics: Problems and Perspectives, Longman, London, pp. 237-278 1976: Die Nominalbildung des gyptischen, Zabern, Mainz 1976: Laut- und Formenlehre des Maltesischen anhand zweier Mundarten, Palm & Enke, Erlangen 1984: Grammatical Analysis and Glossary of the Northwest Semitic Vocables in Akkadian Texts of the 15th-13th C.B.C. from Canaan and Syria, Neukirchener Verlag, Neukirchen

MORRIS JONES, J.

PAUL, H. PUGH, S.M.

RICHARD, S.

RYDING, K.C.

OHALA, J.J.

OSING, J. SCHABERT, P.

SIVAN, D.

SKOLNIK, F. and 2007: Encyclopaedia Judaica, 2nd edition, Thomson Gale, Detroit BERENBAUM, M. (edd.) SOKOLOFF, M. 2009: A Syriac Lexicon, Eisenbrowns, Winona Lake, and Gorgias Press, Piscataway 2000: Ugaritische Grammatik, Ugarit-Verlag, Mnster 2002: Altthiopisch, Ugarit-Verlag, Mnster ULLENDORFF, E. 1977: Is Biblical Hebrew a language?, studies in Semitic languages and civilizations, Harassowitz, Wiesbaden

TROPPER, J.

58

VON REDEN, S.

1992: Ugarit und seine Welt, die Entdeckung einer der ltesten Handelsmetropolen am Mittelmeer, Gustav Lbbe, Bergisch Gladbach 1995: Grundriss der Akkadischen Grammatik, 3rd edition, Pontificium Institutum Biblicum, Rome 1992: Early Germanic Grammar, Academic Press, San Diego 1966: Die lexikalischen und grammatikalischen Aramaismen im alttestamentlichen Hebrisch, Alfred Tpelmann, Berlin 1971: A dictionary of modern written Arabic, 3rd edition, Spoken Language Services, Ithaca

VON SODEN, W.

VOYLES, J.B. WAGNER, M.H.

WEHR, H.

59

You might also like