1The verbal root √hlk, with the original meaning ‘to go’, is attested in almost every branch of the Semitic language family, including Hebrew. In most cases it behaves irregularly, exhibiting unexpected paradigms. The phonological changes responsible for the anomalous forms are often equally irregular and therefore ad hoc.

2Irregular forms of √hlk are well known and many scholarly opinions on their appearance in the individual languages have been shared. Despite parallels between the various Semitic languages, an investigation from a general Semitic point of view is lacking. In this paper I will point out these parallels, albeit with a slight focus on Biblical Hebrew and its closest Northwest Semitic relatives. Moreover, I will make use of modern linguistic theory in order to explain why some developments are more likely to have happened than others.

1.0. The Data

3We find verbal forms of √hlk in Akkadian, Arabic, and almost all Northwest Semitic languages.2 The table on the following page sums up the most common forms in each language.

4In most languages the first radical h is missing or behaves irregularly in many forms, marked in the table with shading. The rationale behind this peculiar feature of √hlk is the main subject of this paper and will be treated in detail below. Here I confine myself to select observations.

5Excluded from the table are the Biblical Hebrew ‘strong’ formations, namely prefix conjugations that preserve the first radical, such as imperfect ְ י ַ ה ֲֹ ךל and construct infinitive ְ ה ֲֹ ךל . They are rare: the 19 such formations are contrasted with exactly 1000 ‘weak’ formations. Furthermore, their restricted attestation in presumed foreign or ‘non-Judahite’ passages, such as the book of Job (6x) and the Balaam narrative (3x), suggests that they are style-switching features (cf. Rendsburg 2003, feature II.C.7).3 These forms have been linked to the Moabite consecutive imperfect wʾhlk ‘and I went’ (Meshaʿ lns. 14–15), but it should be noted that the strong consecutive imperfect is very rare, being attested only twice in Biblical Hebrew, of which ו ַ ת ּ ה ֲ ל ַ ך ‘and (fire/lightning) ran’ (Exod. 9.23) may actually be a piʿʿel form (Joüon and Muraoka 2011, §75g: lectio mixta) (the second is ַּ ו ַֽ י ַּ לה ְֹ֑ ך ‘and he passes’ [Job 14.20]).4

Table 1

6Prior to the discovery of Ugaritic, the absence of the first radical was commonly explained as a result of dissimilatory loss of this consonant in the hifʿil, i.e., *hahlīk > *hālīk > ךילוה , after which it spread to the qal by analogy with the I-w verbs (Praetorius 1882). Not only is it most unlikely for analogy to act from a less commonly occurring paradigm to a more common one (see §3.0), this explanation was also no longer tenable when Ugaritic turned out to have lost the first radical in the G-stem (impf. ylk, inf. lkt), but not in the C-stem (impf. a͗šhlk_1S). Although Praetorius’ theory of dissimilation remains plausible for the hifʿil (Meyer 1992, §78.9b), it is found nowhere else in Biblical Hebrew, and is therefore an ad hoc solution.5

2.0. Frequency

7There are multiple ways in which frequency, more specifically token frequency,6 plays a significant role in language change.

8High token frequency can either stimulate or prevent formal change. This paradox is explained by the Frequency–Implementation Hypothesis:

Changes which require analysis—whether syntactic, morphological, or phonological—during their implementation affect the least frequent words first; others affect the most frequent words first. (Phillips 2001, 134)

9‘Analysis’ in this context is to be understood as the fragmentation of an utterance into individual words, morphemes, or sounds as performed by the speech recipient in order to interpret the message. Common phrases and words, however, are perceived as single units rather than sequences of elements (Phillips 2001, 127). An example of this is the grammaticalised use of American English I am going to, which can be phonetically reduced to [ˈɑiməɾ̃ə] without becoming misunderstood (Bybee 2001, 11). From the speaker’s perspective, recurring utterances of a word or phrase undergo ‘automation’ like other highly repetitive behaviours, and become less clearly articulated (Bybee 2015, 41). At the other side of the conversation, phonetic reduction (i.e., deficient articulation) generally causes interpretation issues for the listener, but this does not apply to contextually predictable words—and the more frequent a particular form, the more predictable it is.

10The other aspect of the Frequency–Implementation Hypothesis, the stability of frequent words, can be attributed to the fact that “the more token frequency a form has, the better and earlier it is stored in the brain” (Dressler 1985, 333). As such, frequent forms are accessed more easily, irrespective of their morphological irregularity (Bybee 2015, 95). A speaker of English will reproduce the past tense of to go correctly as went without hesitation, whereas he needs to derive the past tense form of a verb like to vamoose or to abscond. Because went occurs frequently, it is stored in the human mind as a lexical item separate from the present tense of to go. Hence, very frequent forms are resistant to paradigmatic levelling, which is the most significant change that requires morphological analysis (cf. Phillips 2001, 134). For instance, as word-initial *w had become *y in Proto-Northwest Semitic, Biblical Hebrew stative I-w verbs like *√wqd > דקי ‘to burn’ were reanalysed as I-y verbs on the basis of the forms where it had changed, e.g., the perfect *waqada > *yaqada, causing the analogical replacement of word-internal *w with *y, such as the imperfect *yiwqadu → *yiyqadu > קיי ַ ד ‘it will burn’. The frequently occurring *yiwkalu ‘he will be able’, however, did not undergo this analogical levelling, resulting in the form כוי ַ ל (Blau 2010, §§1.16.2, 1.16.4). Similarly, ןתנ ‘to give’ is the only III-n verb in which the third radical is assimilated to a following consonant, e.g., נ ָ ת ַ֫ ית ‘I gave’ vs. ׁ ש ָ כ ַ נ ְ ית ‘I dwelt’; the verb’s high frequency blocked a morphological analysis nāṯan + - tī (cf. Blau 2010, §4.3.8.3.4).

11In conclusion, frequently occurring forms exhibit consistent but irregular phonetic reductions that are resistant to paradigmatic levelling.

12The verb for ‘to go’ (i.e., denoting basic motion) belongs to the most basic lexical stratum of every language and is uttered very often. Within the selection of Biblical Hebrew discourse clauses7 (based on the annotations of Van Peursen, Sikkel, and Roorda 2015) ֲ ה ָ ל ְַּ ך ranks among the six most frequently used verbs. Among the forms of this verb the 2ms ל ֵ ך (and morphological variants, 140 attestations) and 2mpl ( ל ְ וכ 83) imperative forms are the most common forms, accounting for more than a quarter of all qal forms of ה ָ ל ְַּ ך in direct speech (774 in total). This aligns with the paradigmatic independence that is found with imperatives, in particular basic directives like ‘go! ’, ‘come! ’, ‘give! ’, ‘take! ’, all over the world (cf. Veselinova 2006, 139; Maiden, Swearingen, and O’Neill 2009, 105–6).8 Hence, it is by no means surprising that this form is phonologically reduced throughout Northwest Semitic. At some stage the *h in *hlik,9 standing in a phonetically unstable position, was dropped.10 Due to a lack of data, it is impossible to date this development more accurately than somewhere between Proto-Semitic and Proto-Northwest Semitic;11 at least the Akkadian imperative alik, probably going back to *halik, did not participate. Akkadian does show another irregular phonological reduction in the paradigm of alākum: original *h exceptionally assimilated to an adjacent consonant, e.g., illik < *yihlik-, ittalak < *yihtalik- (or *yithalik-; Kouwenberg 2010, 546, fn. 114).12 Most likely, this assimilation first occurred in the most frequent form where *h stood before a consonant, for instance the preterite *yihlik-, later spreading to other forms, presumably on analogy with I-n verbs.13

3.0. Analogy

13In the past, historical linguistics has been accused of using analogy as a ‘wastebasket’ for linguistic changes that could not be explained by regular change or borrowings (Campbell 2013, 91). There are, however, several tendencies that aid us in understanding how and—perhaps most importantly—in which direction analogical changes operate (Bybee 2015, 115).

14In the first place, forms that influence each other must be semantically related (belonging to the same morphological category) as well as phonologically similar. For instance, the past tense of verb A cannot be analogically reshaped directly on the basis of the present tense of verb B. In the second place, analogical change regularly operates from a basic to a derived form within a paradigm (Kuryłowicz 1945, 23). The basic form is identified as the ‘least marked’ in terms of phonology (short > long), morphology (stem only > with affixes), semantics (e.g., singular > plural; masculine > feminine; active > passive; present > past, future), and, above all, token frequency (frequent > rare). Bybee stipulates the following hypothesis:

High-frequency forms are resistant to change on the basis of the structure of other forms or patterns, and more likely to serve as the basis of such change in low-frequency forms. (2015, 102)

15Accordingly, after the collapse of the case systems, the Bronze Age Northwest Semitic languages generalised the presumably more frequent oblique plural ending *-īm/n rather than the morphologically less marked nominative *-ūm/n. Perhaps the most famous case of analogy in Biblical Hebrew verbal morphology is the introduction of ī in the stem of the strong verb hifʿil imperfect י ַ ק ְ ליט on the basis of analogy to II-wy hifʿil, e.g., י ָ םיק :

16It may seem unlikely that a rather limited verb class can influence the paradigm of a regular verb, but less so when we see that among hifʿil imperfect forms in Biblical Hebrew direct speech clauses the II-wy ה ֵ ש ִׁ בי ‘to return (trans.)’ is the most common verb, with ה ֵ ב ִׁ אי ‘to bring, insert’ being third.14

17Thus, even though frequent forms behave independently of the rest of the paradigm, they can conversely serve as an analogical model for less frequent forms within the paradigm.

18Given the morphological relationship between the imperative and the prefix conjugations (including the construct infinitive), we may expect the latter to be reshaped in analogy to the phonologically reduced imperative of √hlk, as shown in (2). After all, the imperative * (h) lik is more basic than the imperfect *ya- (h) lik- in terms of phonology, morphology, and frequency.15

19Additionally, the anomalous paradigm of √hlk was associated with the scheme of the active I-w verbs already before Praetorius (1882). The active I-w verbs, most of which are motion verbs like √hlk (e.g., √wrd ‘to descend’, √wṣ́ʾ ‘to go out’, √wθb ‘to sit down’, √wbl ‘to bring’), are characterised by a biradical stem with stem vowel i throughout the Semitic languages.

Table 2

20The similarities of this verb class with √hlk may have formed an additional impetus for the previous analogical development and also explain the feminine ending -t for the construct infinitive (another hallmark of I-w verbs in Northwest Semitic):

21This development could have happened at any time in the history of the (West) Semitic languages after the loss of the first radical in the imperative. Only in Phoenician did it also spread to the qal perfect ylkt_1S */yalaktī/, parallel to the levelling in the paradigm of the verb √ntn ‘to give’ in both Phoenician and Ugaritic (impv. */tin/, impf. */yatin-/, pf. */yatan (a)/16). On the basis of Hebrew (Blau 2010, §3.3.5.5.1) and Phoenician (yipʿil pf. ylk and impf. ylk), I would postulate an additional extension of I-w forms to the causative stem of √hlk in Proto-Canaanite. The dissimilatory rule *hah- > *hā- (> hō-) put forward by Praetorius (1882) would explain the dissimilarity with Ugaritic a͗šhlk_1S */ʾašahlik-/, though it is unnecessary and otherwise ad hoc (see above). If it happened, it must have done so before Phoenician shifted the causative morpheme from *hi- to *yi-.

22But were the imperfect forms of √hlk not frequent enough to resist paradigmatic levelling?17 It is possible that the loss of h in the imperfect *yahlik- should indeed be attributed to a phonological reduction similar to that in imperative *hlik, but there is evidence that the imperfect of √hlk (and active I-w verbs) was not as paradigmatically independent as the imperative. If Biblical Hebrew י ֵ ל ֵ ך goes back to *yaylik- (Joüon and Muraoka 2011, §75c; Suchard 2019, 250),18 it reflects insertion of a first root consonant *y. This can only be explained by two-stage analogical levelling with I-wy verbs:

23This analogy is additional proof of the paradigmatic autonomy of the imperative, where the *y was analogically added in neither I-w verbs nor √hlk. It is unclear whether this development happened only in (Biblical) Hebrew, as the Canaanite data provides no information on the prefix vowel of the relevant verbs. In Ugaritic, we can infer from 1s impf. a͗lk /ʾalik-/, a͗rd /ʾarid-/ (vs. *ʾay- > /ʾē-/ <i͗>) and the attestation of syllabic representations of /yatin-/ (vs. BH י ִׁ ת ְֲֵ ן ) and sim-ilar forms (Sivan 1985, 292) that this development did not occur. Given that in Biblical Hebrew this -ē- < *-ay- is never spelled with a mater lectionis י (contrasting with I-y, e.g., קיי ַ ד ‘it will burn’), this analogy might be as late as the reading traditions.19

24The forms of √hlk attested in Arabic have not been discussed so far, as they are completely regular. The origin of this regularity is a semantic shift ‘to go’ > ‘to perish’ and a concomitant loss of frequency. No longer afforded the protection of high token frequency, any morphological anomalies were ousted by paradigmatic levelling.

4.0. Suppletion

25In his thorough study of suppletion as a synchronic state (rather than a morphological process), Mel’čuk (1994) summarises his definition of suppletion as follows:

… for the signs X and Y to be suppletive their semantic correlation should be maximally regular, while their formal correlation is maximally irregular… (1994, 358)

26Thus, e.g., the semantic difference between go and went is a regular grammatical difference (viz. ‘−/+ past tense’), but there is no rule in English grammar that can explain their morphological dissimilarity. After all, past tense is regularly marked by the suffix -ed (as in play: played) or vowel alternation (as in swim: swam). Biblical Hebrew qal ש ָ ת ָ ה ‘to drink’ has no hifʿil, but instead the causative ‘to make to drink’ is expressed by using the hifʿil ִׁ שה ְּ ק ֲָ ה . The formal correlation between ת and ק is, however, not “grammatically corepresentable” (cf. Mel’čuk 1994, 347).

27While both of these examples are instances of suppletion through lexical change, a suppletive paradigm can also be the result of irregular and regular phonological changes.20 This type of suppletion is generally rarer, as it is often eliminated via paradigmatic levelling (Kim 2019, 457). Moreover, most regular sound laws lead only to regular alternations. For instance, as post-vocalic spirantisation in Biblical Hebrew is a regular phenomenon, כ ָ ת ַ ב [kʰɔːˈθaːv] does not stand in a suppletive relationship to כי ְ ת ֹ ב [jiχˈtʰoːv], even though two of the three root consonants are phonetically different.21 The absence of the first radical of active I-w verbs in several forms of the paradigm is another alternation, as it occurs regularly in this class of verbs.

28Contrastively, there is no allophonic alternation (*w >) y: h in Northwest Semitic that can account for the difference between √hlk in the perfect, absolute infinitive, and participle, on the one hand, and the imperative, imperfect, jussive, and construct infinitive, which are derived from a root *√wlk, on the other. Biradical imperatives regularly correspond to a perfect with initial n (*śaʾ: *naśaʾa) or y (*rid: *yarada), while perfect forms with initial h also regularly have it in the imperative (*haraga: *hrug). The meanings of the forms in which this alternation is distributed correlate like any other verb, with the differences being completely grammatical. As both conditions of Mel’čuk’s definition are met, it follows that √hlk and a hypothetical *√wlk were in a suppletive relationship in Proto-Northwest Semitic and to varying degrees in most of its descendants Ugaritic (not in the C-stem), Phoenician (also in the perfect), and Hebrew.

29Further proof of this suppletive relationship is seen in the fact that the √hlk forms did not participate in the Biblical Hebrew analogical levelling of I-w verbs presented in (4)– (5), whereas the imperfect and jussive did.

5.0. Conclusions

30We have seen three linguistic phenomena that account for the anomalous forms of √hlk in Biblical Hebrew and other Semitic languages. These explanations are not completely new; Blau (2010, §3.3.5.5.2; cf. 1979, 145, fn. 14) already pointed towards the role of high token frequency in the shortening of the imperative *hlik > *lik; analogy has been commonly accepted for the similarity between √hlk and active I-w verbs for a long time (e.g., Praetorius 1882); and Gesenius (1816, §86) already mentioned the suppletive character of √hlk. What is new, however, is the use of modern linguistic theories in order to explain the parallels and differences in this paradigm across Semitic languages.

31The imperative *hlik was likely the first form to behave anomalously, being phonetically reduced to *lik because of its high token frequency, probably in an early stage of West Semitic. Subsequently, the prefix-conjugations (and the construct infinitive) were reshaped in analogy to the imperative and the morphologically and semantically similar active I-w verbs. This resulted in a paradigm in which affected forms, going back to a hypothetical *√wlk, and unaffected forms, regularly formed from the root √hlk, are in a suppletive relationship with each other. Along the way we saw irregular, high frequency-induced assimilation (and subsequent analogy) in Akkadian, and paradigmatic levelling in Arabic, where the verbal forms had lost their paradigmatic independence due to a semantic shift.

32By way of a summary, the following table provides a concise overview of the development of the G-stem forms of √hlk from Proto-Semitic to Biblical Hebrew. Phase I is not unlikely to correspond to Proto-West Semitic; the changes in Phase II occurred thereafter, the latest in Proto-Northwest Semitic; the loss of the first radical in the hifʿil (Phase III) can be postulated for Proto-Canaanite; Phase IV is the (probably Hebrew-only) insertion of *y in the prefix of I-w verbs and √hlk by analogy with the strong verb, which could have happened at any time before the split of the Tiberian and Babylonian reading traditions.

Table 3