The roots of language ⋆
The roots of language ⋆
Abstract and Keywords
Recent morphological theories have highlighted two very different views of what roots are. Within Distributed Morphology, roots are the basic meaningful lexical units of language. Within autonomous morphology, where the meaningful lexical units are lexemes, roots play a very different role and their necessary meaningfulness has been called into question (Aronoff 1976, 1994, 2007). In this article, I will provide detailed evidence from English irregular verbs that roots are purely morphological entities and that English verb roots cannot be meaningful. I will then trace the history of the study of roots in Semitic grammar, beginning with the earliest Arabic grammarians and moving on to the early Hebrew grammatical tradition. I will show that, although traditional Arabic grammar dealt with roots, it was the Hebrew grammarians who gave roots the special status that they acquired in Western grammatical description from the Renaissance on. I will finally review evidence from inside and outside formal linguistics on the status of roots in modern Semitic languages and conclude that, although much points to the reality of roots in these languages, there is little or no evidence that these roots are meaningful as opposed to purely morphological entities. This calls into question the basis of Distributed Morphology, or any theory that relies on meaningful roots as fundamental theoretical entities.
In this chapter, I will try to cast light from several angles on the term root as it is used in current morphological theory and on the various notions that the term embodies. I will contrast two very different broad types of morphological theories that traffic in roots, which I will term root-based and lexeme-based theories. In root-based theories, morphological structure is built up from meaningful simplex roots. In lexeme-based theories, the starting point of morphological structure is lexemes, which are meaningful but need not be simplex; in this type of theory, roots may be part of morphological structure, but they play a less central role and their relation to meaning is more complex.
From a very general point of view, the major difference between the two lies in aesthetics. Root-based theories value reduction and perfection highly, both as analytical strategies and as aesthetic desiderata. For these theories, it is highly desirable on aesthetics grounds that the basic lexical elements of language be simplex signs. Lexeme-based theories give both these considerations much less privileged places in both theory and practice, so that it does not much matter whether the ultimate units of language are simplex signs or not. The apparent fact that simplex signs do not occupy a place of honour is interesting, but not surprising.
As exemplars of the two types of treatments of roots, I have chosen to contrast Distributed Morphology, because it is the most widely discussed root-based theory, with my own work, especially as exemplified in Aronoff (1994). Distributed Morphology (DM) has been around for some twenty years in a variety of slightly (p.162) different incarnations. I will try to confine myself to the version presented in Halle and Marantz (1993), the locus classicus of the theory, but much of what I have to say applies more broadly.
My prejudices are well known, so it should not be surprising that I come down on the side of lexeme-based theories. My goal, though, is not simply to excoriate root-based morphology, but to understand its motivation and sources. In the first part of this chapter, I will explore how roots have been treated in modern linguistics, from both perspectives (§9.3). In the second part, which is more substantive in many ways, I will take a longer perspective and look at the development of the treatment of roots in the grammatical traditions surrounding the two languages where roots have figured most prominently: Arabic and Hebrew (§9.4). I began researching this question under the assumption that the origins of the grammatical theory of roots lay in the Arabic tradition, but I have concluded that it is not obvious that the root as we understand it today was isolated as a distinct entity by the early Arabic grammarians. Instead, it is only in the tradition of Hebrew grammar (itself a somewhat late historical offshoot of Arabic grammar) that we can conclude unequivocally that the root assumed the theoretical role that most linguists have assigned to it. In the last part of the chapter (§9.5), I will look at the evidence for roots in modern-day Semitic languages from more empirically oriented approaches: neurolinguistics, experimental psycholinguistics, and first language acquisition. I will show that this evidence comes down on the side of lexeme-based theories of morphological roots.
Overall, my aim is to show that roots are real linguistic entities but that there is no point in assuming that all instances of a given root have the same meaning in any language, that roots are semantically invariant. Roots are part of the autonomous morphological system of a language and what is special about Arabic and Hebrew are the obligatory templates for verbs, not the roots. Overall, this conclusion should not be surprising.
The idea that roots are semantically invariant is old and grows out of a grammatical tradition whose object of study was, in the eyes of its practitioners, the sacred language of the first known divinely-inspired book, the Old Testament, and even more specifically the Pentateuch. It should not be surprising that students of such a language might want to treat it as a perfect system. This tradition persisted into the nineteenth century in such fanciful ideas as those of Fabre d’Olivet (1815–1816), who believed that not only Biblical Hebrew roots but the Hebrew letters themselves had invariant meanings, a view that greatly influenced Benjamin Lee Whorf’s work on inner meaning. What should surprise us is the persistence of such beliefs when considering the languages of humans rather than God.
The discussion of roots has generated as much heat as it has light over the last twenty years. But hot debate among grammarians is nothing new. William Chomsky, in his book on David Kimḥi’s grammar of Hebrew, tells the following story about a pair of medieval grammarians:
The versatile and brilliant statesman, soldier, and poet, Samuel ha-Nagid (11th century) found time amidst his multifarious duties and occupations to engage in verbal clashes on grammatical issues with the profound grammarian Ibn Janah and to write, according to Ibn Ezra’s testimony, twenty-two books of ‘supreme quality’ (v’eyn lm’alah mimenu) on Hebrew grammar […] The grammatical controversy between Ibn Janah and Samuel ha-Nagid is recounted at some length in the writings of some of the medieval Hebrew grammarians. Samuel ha-Nagid apparently aroused by Ibn Janah’s criticism of some of the views of his teacher, Yehudah Hayyuj, sent a messenger from Granada to Saragossa, the place of Ibn Janah’s residence, charged with the task of challenging Ibn Janah to a verbal duel on certain grammatical issues and of exposing publicly the ‘fallacy’ of his theories. On his arrival in Saragossa, the messenger stayed at the home of a communal leader in that city, named Abu Soleiman ben Taraka, a friend of Ibn Janah. A public reception was arranged in honor of the visitor, to which Ibn Janah was invited. The latter, without suspecting the chief purpose of the gathering, accepted the invitation. During the reception, the visitor began to inveigle Ibn Janah gradually and subtly into a discussion. Some of the questions raised by him were readily disposed of and adequately answered by Ibn Janah. But others followed, and Ibn Janah, unprepared for this barrage of questions, was befuddled, and he promised to reply at some future time. He did so and sent his reply to the visitor. The latter, however, superciliously remarked that it would be wiser for Ibn Janah to withhold his reply until the Nagid’s book was published, where he would find even more serious criticisms leveled against him. This Ibn Janah refused to do. He issued his reply in book form and called it Kitab at-Taswiya. After the publication of the Nagid’s attack on him Ibn Janah retorted with a violent counter-attack in a book, which he called Kitab al-Tashwir. Ibn Janah was very proud of this book, and he frequently referred to it in glowing terms, but, unfortunately, only a fragment of it is now extant. (Chomsky 1952: xv–xvii).
Similar tales are told about the Arabic grammarians from a much earlier time. Kufa and Basra are two cities in Southern Iraq founded by early Muslims, acknowledged to have been the intellectual centres of early Islam. The city names came to represent two schools of Arabic grammar, with the Kufan school being in the minority. The two schools differed on, among other matters, whether the maṣdar verbal noun was the ʔaṣl (base) of the verb or vice versa. The battle raged for centuries, though the Basrans are generally acknowledged to have won in the end.
Plus ça change, plus c’est la même chose. The battle between DMers and their opponents has raged with equal ferocity over the last two decades. Here is a small sample of the invective in Marantz (1997: 202):
This paper brings the reader the following news: Lexicalism is dead, deceased, demised, no more, passed on […] The underlying suspicion was wrong and the leading idea didn’t work out. This failure is not generally known because no one listens to morphologists. Everyone who has worked on the issues of domains—what are the domains for ‘lexical phonological rules’, what are the domains of ‘special meanings’, what are the domains of apparently special structure/meaning correspondences—knows that these domains don’t coincide in the ‘word’ and in fact don’t correlate (exactly) with each other. But the people that work on word-sized domains are morphologists, and when morphologists talk, linguists nap.
Marantz’s opponents are more polite, but no less damning. Here are a few samples from Williams (2007):
These awkwardnesses all stem from the idea that idiomatic meanings can all be fixed on the ‘roots’ that occur in the idiom, and not on the idiom itself; and that stems from the decision that the lexicon(s) in DM do not list any derived forms, and that in turn stems from the decision that morphemes are the sole units of insertion. (Williams 2007: 360)
As DM stands, all applications of the Pāṇinian principle which are not morpheme vs. morpheme contests must be recast as one or another kind of rule of special allomorphy, turning gold into clay. (Williams 2007: 364)
There is no reason to accept any part of Marantz’s analysis of nominalization. (Williams 2007: 367).
9.3 Roots in modern morphological theory
9.3.1 Roots and the intellectual sources of DM
Meaningful roots are central to DM. Marantz (1997: 212–13) proclaims that ‘Things with special meanings are roots’. The source of DM’s insistence on the centrality of roots lies in the linguistic ideas of Roman Jakobson. Jakobson was a staunch reductionist who was strongly influenced by the successes of physics in the early twentieth century and believed that linguistics consisted mainly of removing the surface variance of languages to reveal the underlying invariance. Jakobson admired Baudouin de Courtenay and Kruszewski for their work on invariance, which he often argued to be the main source of modern synchronic linguistics.1 Jakobson applied this aesthetic of invariance most fully and famously to his analysis of Russian nominal cases, in which the various meanings and uses of each case were reduced to a single abstract meaning (Jakobson 1971a).
Classic generative phonology was most centrally an attempt to apply a radical version of Jakobsonian invariance to sound systems. Its goal was to reduce the (p.165) linguistic signifiers of a language to an underlying level at which there was a one-to-one correspondence between the forms and meanings of simple signs. The DM notion of the root is a direct outgrowth of this Jakobsonian vision of languages: lexical items should be reduced to roots and each root should have a single form and a single meaning at some underlying level.
I have found no mention of roots in Chomsky and Halle’s (1968) The Sound Pattern of English (SPE), the great masterwork of classic generative phonology; not surprising, since SPE makes no theoretical distinction between morphology (where roots reside) and syntax. The internal structure of a word like theatr-ical-ity is not provided by morphology but by syntax, in the form of a bracketed phrase structure tree to which the rules of phonology apply cyclically, with the cycles determined by the tree structure. There is implicit morphology distinct from syntax in SPE, though. The affixes in this same word theatricality, for example, are provided with internal morphological structure: ic+al and i+ty. In a footnote, the analysis of i+ty is said to be ‘well motivated on morphological grounds’ (Chomsky and Halle 1968: 33), based on the existence of the noun-forming affix -ty and such sets as sanctity–sanctify–sanctitude and clarity–clarify, although it is noted in the next footnote that this analysis of ity is not required by the phonology.
SPE makes use of the term and notion stem. For example, the dox of orthodox and the graph of photograph are labelled as stems and given their own (initial) cycle. Elsewhere, in particular in the analysis of Latinate verbs of the form ‘prefix–stem’ such as permit, transfer, and compel, the authors do not give the stem its own cycle, marking this particular construction by the famous ‘=’ boundary (which is identified formally as neither a formative boundary nor a word boundary). In conclusion, nowhere that I can find does SPE use the term root; and though stem occurs in SPE it seems to be used by and large for bound stems, and apparently those without much semantic content.
The DM distinction between a list of roots and a list of words can be traced directly to Halle (1973). Halle uses the term root on the second page of this article, noting that ‘the list [of morphemes] must include not only verbal, nominal, and adjectival roots but also affixes of various sorts’ (Halle 1973: 4). One of the main points of this article is to provide an account for the fact that not all possible words actually occur, and those that do occur are often idiosyncratic in form or meaning. Halle encodes these idiosyncrasies by means of a dictionary that he characterizes as a filter:
I propose that idiosyncrasies of the type just illustrated be listed in a special filter through which the words have to pass after they have been generated by the word formation rules. The special information given in the filter under each entry is then added to the representation of the word. In the case of semantic idiosyncrasies such as those exemplified by the special meaning of nouns like recital and transmittal the filter would supply the (p.166) appropriate indications about their semantics. In the case of phonological idiosyncrasies like those exhibited by nouns like obesity, the filter would supply the information that the noun in question is not subject to the Trisyllabic Shortening Rule, or, more formally, would supply the noun with the feature [–Trisyllabic Shortening Rule]. Finally, ‘gaps’ in the dictionary […] would be accounted for by providing the ‘missing’ words with the rule feature [–Lexical Insertion]. In other words, the fact that English lacks the nouns *derival and *arrivation would be reflected in the grammar by marking these words, which would be generated by the word formation rules, as not being subject to lexical insertion and therefore incapable of appearing in any actual sentence of the language, in spite of the fact that they are neither semantically nor syntactically or phonologically anomalous. (Halle 1973: 5).
In other words, I am proposing that the list of morphemes together with the rules of word formation define the set of potential words of the language. It is the filter and the information that is contained therein which turn this larger set into the smaller subset of actual words. This set of actually occurring words will be called the dictionary of the language. (Halle 1973: 6).
Lexical insertion, within this framework, selects fully inflected words from this dictionary.2 Researchers outside DM have not pursued the idea that the dictionary is a filter, because it is difficult to understand what Halle meant by the term filter. Are the dictionary entries the words that make it through the ‘exception filter’ or the ones that remain behind, those that are filtered out? The filter cannot be uniform in texture in any sense, as normal filters are, because it singles out the words that actually occur, which presumably comprise some sort of list. Furthermore, the filter provides some of these words, at least, with additional information that the rules of word formation do not provide. Overall, the relation among the dictionary, the filter, and the morphemes is not at all clear in this framework, which is admittedly not meant to be more than a prolegomenon.
DM makes use of another traditional term, morpheme, albeit in a very untraditional way. For DM, morphemes are not signs, sound/meaning pairs, but rather ‘terminal elements of the trees [that] consist of complexes of grammatical features’, which ‘are supplied with phonological features only after Vocabulary insertion at M[orphological]S[tructure]’ (Halle and Marantz 1993: 114). In other words, what they call morphemes are what most others call morphosyntactic features or properties or values, and what they call Vocabulary insertion (or ‘the addition of phonological material’ (p. 131)) is what others call morphological realization.3 But Vocabulary insertion does more: ‘In addition to phonological features, Vocabulary insertion supplies morphological features that signal idiosyncratic properties of specific Vocabulary items’ (p. 136).
(p.167) Nowhere do Halle and Marantz discuss the term root in their 1993 paper. As far as I can tell, roots are introduced into DM by Marantz (1997), who ‘explodes the Lexicon and includes a number of distributed, non-computational lists as Lexicon-replacements’ (p. 203). The first list of three, which he calls the ‘narrow lexicon’, ‘contains the atomic roots of the language and the atomic bundles of grammatical features’ (p. 203). The second list is the Vocabulary of Halle and Marantz, i.e. morphological realization or spell-out. The third list is Halle’s dictionary redux under another name: ‘Encyclopedia—the list of special meanings. The Encyclopedia lists the special meanings of particular roots, relative to the syntactic context of the roots, within local domains’ (p. 204) and includes not only words but also all idiomatic expressions. In short, it is a Bloomfieldian lexicon of idiosyncratic complex items.
I won’t speculate as to why DM plays so fast and loose with traditional terminology, here conflating dictionary and encyclopaedia, elsewhere morpheme and morphosyntactic feature value. The problem in this instance is that there is a well-established dichotomy in lexicology between the two terms at issue here, which is lost by calling one the other: a dictionary is a list of words: a dictionary definition contains only what is necessary to distinguish one word from another; while an encyclopaedia is about concepts and things: its entries contains all that there is to know about the entity named in each. Contrast the Oxford English Dictionary of 1928 with the Encyclopaedia Britannica of 1911. The first sought to include everything there was to know about the words of English. The other sought to include everything there was to know about everything besides the words of English. The DM Encyclopedia is not an encyclopaedia, but instead a dictionary, in fact a direct descendant of Halle’s dictionary, a Bloomfield lexicon of ‘idioms’, irregular entities, including words and phrases. But within a theory that wishes to give words no special place, there can be no dictionary of words, and so the former dictionary is now called an encyclopaedia, even though it is not one, or at least this is the only explanation I can provide for the peculiar use of terminology here.
Which brings us back to roots. Roots in DM are mysterious objects. They have no phonological form, at least not at first, since they only receive form at the point of Vocabulary insertion and they are also subject to readjustment rules, but they do bear meaning: ‘Things with special meanings are roots’ (Marantz 1997: 212–13). As Williams (2007) notes, for DMers, it is not the case that what DM calls idioms have idiosyncratic meaning, but rather that the roots have idiosyncratic meanings in the context of individual expressions (idioms). DM roots are apparently not just abstract meaning-bearing elements, but rather the only true bearers of lexical meaning. In a way, they are like lexemes, except for being simplex (Aronoff 1994). And, as Williams argues, it is this claim about simplex roots bearing meaning that brings DM to grief.
(p.168) But why make such a claim, especially when the existence of the ‘Encyclopedia’ makes the claim empirically empty, given that roots can have ‘special contextual meanings’ (Marantz 1997: 213)? The answer is that DM is a last-ditch attempt to preserve the vision of Jakobsonian invariance of elementary signs in the form of roots. From a signifier point of view, DM abandons this invariance to Vocabulary insertion, but it apparently cannot go so far as to abandon the signified side and so resorts to the artifice of supplying roots with abstract meanings that are then somehow moulded in context by means of the Encyclopedia. The upshot is that DM roots, in truth, have neither constant form nor constant meaning.
9.3.2 Roots within lexeme-based morphology
On 29 May 2011, Simon Winchester published a piece in The New York Times on the three ‘most complex’ verbs in the English language: set, put, and run. The last has now assumed top position as the verb in the Oxford English Dictionary (OED) with the most senses: 645 according to the official OED count. I will not attempt to prove that none of these 645 can be reduced to any other one, but consider just a few, listed in (1):
a. He ran the table
b. The trains run between Jerusalem and Tel Aviv (ambiguous)
c. Run a bath
d. Run before the wind (of a sailboat)
f. Run over
g. Run a seam
h. Run up a bill vs. run up a hill
i. Run someone through with a sword
Within a lexeme-based theory, each of the however many distinct senses of run we finally decide there are is a separate lexeme with its own meaning, even though all are listed together in the dictionary. And here is where the concept of root becomes useful. All these distinct senses share the same three idiosyncratic forms: run, ran, run (the last two forming the past tense and past participle). Within a lexeme-based theory, we say that all three share a single abstract root √RUN. √RUN does not have a constant sense, because it occurs in all these distinct lexemes; nor does it have a constant form. But all these lexemes share this single root, because they all vary in the same way under the same morphological conditions. Within lexeme-based theory, then, a root is a purely abstract morphological entity, a morphome. I suspect that the same is true in DM.
(p.169) This analysis extends to all highly frequent irregular root/lexemes in all languages, which show the same consistent irregularity across multiple senses. In (2), I have provided a list of the most common English irregular verbs in rank frequency order:
(2) BE, HAVE, DO, SAY, GET, MAKE, GO, KNOW, TAKE, SEE, COME, THINK, GIVE, FIND, TELL, FEEL, LEAVE
Just like √RUN, all of these verbs are highly polysemous and not one of them has a regularized analogue. We therefore say in a lexeme-based theory that each of these irregular verbs is a root that occurs in a large number of homophonous lexemes. We can also now understand how polysemy and homophony apply within such a theory: the roots are polysemous (inasmuch as they each occur in many lexemes), but the lexemes are homophonous (inasmuch as they share a single root).
Roots like English √RUN identify the paradigm of the lexemes that share this root. The root is not a semantic or phonological entity. Compare the homophonous English roots √RING1 (with past tense form rang and participle form rung), √RING2 (with past tense and participle form ringed), and √WRING (with both past tense and participle forms wrung). The root is purely morphological and similar to the flexeme of Fradin and Kerleroux (2003), Spencer’s (2005) lexemic index, and Olivier Bonami’s Paradigm Identifier, which he has discussed in various presentations. All specify the idiosyncratic information about the shape of an inflectional paradigm that may be shared among multiple lexemes. A root, in this sense, is not just a sound form but a set of forms.
I now turn to the notion of a root in Semitic morphology, specifically verbal morphology. I will first present, perhaps for the first time for an audience of non-Semitists, the history of the notion of the root in the Semitic grammatical tradition, with special attention to the (mistaken) idea that roots and morphological patterns have meaning. I will then discuss how verb roots (understood as purely morphological objects in the sense established here for English roots) interact with the binyan system of patterns in Semitic languages, drawing mostly on Israeli Hebrew. I close with evidence drawn from outside the narrow domain of pure linguistic analysis for the claim that Semitic roots are purely morphological entities.
9.4 The origins of roots in Semitic grammatical study
I will explore the historical origins of the idea of roots through a brief survey of the earliest works on Semitic morphology and of the received tradition that has persisted to this day in both the Arabic and Hebrew grammatical traditions. My goal is to understand what the traditional views of Semitic morphology were, how they have developed up to the current time, and to then use these views, in the final section of this chapter, as an aid to understanding what Semitic verb morphology is really like. The main conclusion of this section will be that Arabic grammar operated on roots (p.170) but did not recognize them as a special category separate from other ‘bases’. The isolation of the root as a distinct concept began in early Hebrew grammar shortly after the turn of the millennium and hundreds of years after the establishment of the Arabic tradition.
9.4.1 Arabic dictionaries
The first significant linguistic work on Arabic was a dictionary, Kitāb al-ʕayn ‘the book of ʕayn’, by al-Khalīl ibn ʔAhmad al Basrī (718–791 CE), who lived in the century following the founding of Islam. Al-Khalīl used an Indian-style order for the letters of the Arabic alphabet, starting at the letter representing the farthest point back in the vocal tract, which is pharyngeal ʕayn, and ending at the furthest point forward, which is aleph. Al-Khalīl aimed at a comprehensive Arabic lexicon, though he never got past the first letter, hence the title of the book. His method was to put into one entry all words with the same root. He then grouped together all occurring roots that are permutations of the same set of consonants. So, all k-t-b words are grouped together, but then this root is grouped with all occurring permutations of these three consonants. The fact that the dictionary was organized around roots, as were all subsequent dictionaries of Arabic and other Semitic languages, is a prima facie argument for the value of roots in describing Semitic languages. The fact that roots sharing the same consonants were grouped together, regardless of order, is an indication that al-Khalīl was concerned largely with form.
Homophonous roots were grouped together in one entry, as in the following partial entry for ʕ-sh-q from al-Azhari’s (d. 981) Tahdhīb, a later lexicon in the same tradition:
Abūl-Abbās Aḥmad ibn Yahyā was asked whether love or passion is more praiseworthy. He said: ‘love, because passion includes a degree of exaggeration’. Ibn-al ‘Aʕrābī said ‘ʕushuq are the men who trim the sets of sweet smelling plants; when said of a camel, ʕushuq means one that keeps to its mate and does not desire any other’. He said: ‘ʕashaq is the lablab-tree; the singular is ʕashaqa’. He said: ‘ʕashaq is also the arak-tree. An ʕashiq “lover” is called thus because he withers from the intensity of his passion in the same way as the ʕashaqa “lablab-tree” when it is cut.’ (Tahdhīb I: 170, from Versteegh 1997: 30)
Dictionaries of Arabic ever since have been arranged by the forms of roots, not their meanings. All senses of a given root are listed together. The Wehr/Cowan Dictionary of Modern Written Arabic (cf. Cowan 1971) lists the numbered senses listed in (3), among others, under the verb entry najada for the root njd (where verb measures or patterns are in bold). Essentially synonymous definitions are separated by commas; semicolons separate the major senses. Nouns derived from the same root are listed in (4.)
(3) I: to, help, aid, assist, support; to sweat, perspire;
II: to furnish, upholster; to comb, card, tease (cotton);
III: to travel in the highlands (of Arabia);
X: to ask for help, seek aid; to take liberties, make bold
(4) highland, upland, tableland, plateau; help, aid, succour; emergency, crisis, trouble, difficulty, distress, calamity; courage, bravery, intrepidity, undauntedness; sword belt; upholsterer; upholsterer’s trade; upholstery work; upholstery
Cowell (2005: 36–8) remarks about Arabic dictionaries in general:
Words with the same root commonly have related meanings […] There are countless exceptions however. In Arabic dictionaries, for instance, which are alphabetized by roots—not by bases as Western dictionaries are—‘homonymous roots’ are sometimes entered separately […] This policy has never been consistently carried out, however; the more usual type of entry is the purely ‘formal’ roots, whose sub-entries may include words of various word-families, arranged without regard to meaning.
It is often difficult, if not impossible, to decide without arbitrariness whether two words with the same (formal) root have ‘related meanings’ or not. The use of etymology to resolve some of these difficulties only makes the concept of ‘root’ still more ambiguous.
9.4.2 The influence of the Greek grammatical tradition on Arabic grammar
Versteegh (1977) has demonstrated that the Greek grammatical tradition is a major source for Arabic grammar, to some extent indirectly through Syriac, the Aramaic language in which many of the writings of the early Church fathers were composed, at a time when early Christians and early Muslims lived side by side. Some of his strongest evidence comes from terminology, where we can sometimes see a direct chain from Greek through Syriac to Arabic and even further to Hebrew. The terms for verbs with two arguments, for example, are remarkably similar in meaning in all these traditions, as well as in Latin grammar. All are calques on the Greek word meaning ‘going across’, as shown in (5):
‘going across, passing over’
Versteegh provides more extensive evidence for the intellectual continuity among the Greek, Syriac, and Arabic grammatical traditions, which I cannot rehearse here. But his claim makes eminent sense, given other well-known continuities between the Greek and Islamic intellectual traditions.
Sībawayhi (Abū Bišr ‘Amr ibn ‘Uṯmān ibn Qanbar Sībawayhi) is generally considered to be the author of the first book on Arabic grammar. He studied Arabic grammar in Basra, at about the time that al-Khalīl was writing his dictionary in the same city, and died about 796 CE, soon after al-Khalīl. He wrote one book, known as Kitāb Sībawayhi, often called simply the Kitāb, a grammar of Bedouin Arabic, which is generally regarded as the founding work of Classical Arabic grammar.4 He often refers to al-Khalīl. All subsequent traditional grammars of Arabic fall within the template of the Kitāb.
The Kitāb and subsequent traditional Arabic grammars are divided into two parts. The first part is devoted to syntax: parts of speech (noun, verb, particle) and their distribution, and includes inflection. The second part is devoted to the forms of words: morphology and morphophonology. The coverage of morphology is exhaustive. According to Carter (2004: 100), within this tradition ‘The word is formed from a combination of its root consonants (called ʔaṣl, lit. “trunk or base”, hence “radical”) and the pattern (wazn, lit. “measure” or bināʔ, lit. “structure, building”), in which those consonants are embedded’. Each bināʔ gets its name from the third person masculine past tense form of the verb in that bināʔ, using the root f-ʕ-l, e.g. faʕala.
In traditional Arabic morphology, however, derivation does not operate directly from the consonantal root, as it does in traditional Hebrew grammar, but de mot à mot ‘from word to word’ (Bohas 1984). This type of word-based derivation makes sense if we assume the influence of the Greek and Syriac grammatical traditions, which were strictly word-and-paradigm and admitted of no morphemes. Thus, one word form is said to be the ʔaṣl of another, as follows:5
- The maṣdar is the initial form of the verb
- The maṣdar is the ʔaṣl of the past form of the verb
- The past form of the verb is the ʔaṣl of the present
- The present form is the ʔaṣl of the imperative
In short, traditional Arabic morphology is not root-based in the modern sense. Each form in a derivation was regarded as the base (ʔaṣl) of the next and there was no distinct term for the consonantal root as opposed to other bases. Yes, dictionaries from the start were organized around consonantal roots, but homophonous roots with different senses were never systematically distinguished.
The first grammarians of Hebrew lived in the Golden Age of Jewish culture in Arab Iberia, Al-Andalus, from about the tenth to the twelfth centuries, centred around Cordoba. The earliest Hebrew grammarians were clearly influenced greatly by the Arabic grammatical tradition. Hebrew grammar as we know it was later codified by the members of the Kimḥi family, who lived in Provence, much further north. The Michlol of David Kimḥi (1160–1235) (Chomsky 1952) is the best known of their works. Kimḥi also wrote a dictionary, called Sefer Hashorashim ‘the book of roots’. Traditional Hebrew grammar after Kimḥi largely lost contact with Arabic grammar, due in part to the fall of Muslim Andalusia to the Christians in 1236. Later Hebrew grammarians consequently had little knowledge of Arabic or the Arabic grammatical tradition.
Traditional Hebrew morphology is resolutely root-based. The names of the radicals are identical to those used in Arabic grammar, the three letters p, ʕ, l from the verb ‘do, work’, demonstrating clearly the Arabic origins of the tradition. But a special term ʃɔrεʃ ‘root’ is used in Hebrew grammar to mean only the consonantal root of a verb or noun, which may consist of two, three, or four radicals.6 There is no equivalent term in Arabic grammar and no Hebrew term equivalent to the Arabic ʔaṣl ‘base, source’, which has a much wider use, as we saw in the last section.
It is likely that the Western grammatical term root is a calque of the Hebrew grammatical term. Of the four earliest citations of the English word root in this sense in OED, three refer to Hebrew. Furthermore, only Hebrew grammar, not Arabic grammar, had a distinct term that referred just to the root, and Western Christian grammarians had little or no contact with the Arabic tradition, although some did know Hebrew well because of the sacred nature of the Old Testament. The Old Testament books of the King James Bible, for example, were translated directly from the Hebrew original.
In traditional Hebrew grammar, words are not derived from one another in the Word-and-Paradigm manner used in Arabic grammar and its precursors. Instead, words are derived by putting a root into a binyan ‘structure’ for verbs and a miʃkal ‘weight’ for nouns.7 Hebrew grammar traditionally recognizes seven verb binyanim in which roots can occur and a larger number of noun miʃkalim. Each binyan is said to have a meaning or syntactic function. The meanings of the miʃkalim are not given as large a role. The term used for binyan in traditional Western grammars of Hebrew (p.174) (e.g. Gesenius et al. 1910) is conjugation. One binyan is termed qal ‘simple’. It is equivalent to the bināʔ aṣliyya ‘root structure’ of Arabic grammar and is similarly not named by its form. All others have names corresponding to the third person singular past tense, as in Arabic, called the ground form by Gesenius et al. (cf. German Grundform).
In traditional Hebrew grammar, a binyan is not a pattern or template, terms that have been used recently. The literal translation is ‘structure’ but conjugation is used in English, because binyanim are treated like conjugations in traditional Hebrew grammar, with full paradigms provided for each. For some binyanim it is possible to argue that there is one pattern/template or ground form or principal part. For other binyanim, most prominently qal and nifʔal, there are at least two principal parts in the paradigm. For qal, there are also subconjugations, mostly vowel-driven, similar in spirit to the subconjugations of the Latin third conjugation, with and without the theme vowel i.
9.4.5 Form and meaning in Hebrew morphology
In traditional Hebrew morphology, each root is said to have a constant meaning and each binyan is said to embody a constant syntactic type, defined largely in terms of argument structure. The meaning of an uninflected verb (lexeme) is composed of the meaning of its root and the syntactic type of its binyan. In fact, little of this is cashed out in practice. Consider first the binyan system. First, the syntactic types of only three of the seven binyanim are reliable. Puʕal and hofʕal are more or less restricted to being passives of piʕel and hifʕil respectively, and have always been fairly marginal. Of the five major binyanim, hitpaʕʕel is the only truly reliable one, being almost always reflexive. Of the remaining four, Hifʕil is the most reliable, being usually causative, while piʕʕel is frequently ‘intensive’, not a structural type (though see the discussion of Doron below) and is also used with all quadriliteral verbs, regardless of meaning; a usage that is clearly driven by phonological factors. Qal embodies no syntactic type (it is basic). Nifʕal is usually intransitive but sometimes the passive of qal, which surprisingly has no regular passive, though it is the most common binyan. Overall, there are many verbs whose binyan must be listed lexically. For example, hitpaʕʕel verbs include hitpallel ‘pray’, histakel ‘look’, and hiʃtameʃ ‘use’. Berman (1978: chapter 3) provides a summary of the nuances of the binyan system in Modern Israeli Hebrew. Her conclusion is that the traditional treatment is incompatible with the facts of Hebrew. Instead, ‘each binyan can be said to have certain salient or “unmarked” properties’ (p. 93), with many verb senses listed lexically.
Much less reliable are the roots. No root occurs in every binyan, let alone every nominal mishkal. And, as I have discussed at length elsewhere (Aronoff 2007), many roots have quite distinct senses when they occur in different binyanim. Overall, (p.175) Hebrew roots are very similar to English strong verb roots. None may be quite so polysemous as run, but what unites all the uses of a given root is its participation in the morphology of the language, not its semantics.
Hebrew (and all Semitic languages) has what are called weak roots, which are completely analogous to what are called strong verbs in Germanic. The terminology is confusing, but in the Semitic tradition the consonants in these roots were called weak because they are subject to loss or change in certain cells in the paradigm, while those of ‘full’ roots persist unchanged throughout. As I demonstrated at length in Aronoff (2007), these verb roots, like the Germanic strong verbs of English, can be placed into subclasses, according to their alternations. For example, there is a class of roots that we may call missing n-initial. The members of this class consist of a subset of the roots whose initial consonant is /n/, along with several roots whose initial consonant is /j/, and one with initial /l/. These form an inflectional class because they all lose their initial root consonant under certain circumstances, while regular roots with these same initial consonants do not. Semantics is irrelevant, just as it is with Germanic strong verbs in English.
All Semitic languages investigated to date, except to some extent Maltese (Hoberman and Aronoff 2003), exhibit almost exceptionless root-and-binyan verbal morphology, just as with any other system of verb conjugations (e.g. Latin). Roots and binyanim are morphological entities first and foremost. The difference in the obligatoriness of patterns between verbs and nouns may lie in the fact that verbs are inflected in all Semitic languages, while nouns are not always. Overall, though, neither the verb roots nor the binyanim have constant meaning in any Semitic language so far investigated, despite or perhaps because of the robustness of the inflectional patterns.
9.4.6 A modern analysis: Doron on Israeli Hebrew binyanim
Doron (2003) makes an excellent case for the semantic value of the ‘derived’ binyanim in Israeli Hebrew (those other than qal). But Doron admits that many individual verbs are idiosyncratic. Overall, her analysis supports a theory in which certain binyanim may be used productively as part of the derivational morphology of the language, while simultaneously serving as inflectional classes:
All the meaning contrasts discussed above are achieved by the pairing of equi-rooted verbs [two or more verbs with the same root]. On the other hand, when a single verb is derived from a root, i.e. when the verb is not paired with another equi-rooted verb, then the contribution of the template is more erratic. Even then, a lot of systematicity can be shown to exist if one also takes into account equi-rooted nouns and adjectives. But when no contrast whatsoever is expressed by the choice of morphology, then, as often as not, the template is arbitrary. For example, verbs such as listen, climb, or urge are not semantically causative, despite their derivation by the causative template; neither do perfume, end, or disperse necessarily denote actions, despite their derivation by the intensive template. (p. 23)
It is well known that derivational morphology allows a certain amount of deviation from compositional meaning […] Many intensive verbs are associated with rich encyclopedic knowledge […] Crucially, templates can be specified to have marked features in the environment of certain roots […] Therefore, idiosyncratic verbs have listed templates. The templates of these verbs are completely uninformative—that is, the form/meaning correspondence is rendered as opaque as in the more familiar languages with poorer morphology. (p. 38)
Doron’s account is very similar to that of Cowell (2005) for Syrian Arabic, where a change in binyan is treated as a form of derivational morphology or word formation: each derivational process has a syntactic and semantic effect that is expressed through a change in binyan, but the binyanim have no constant meaning outside their use in derivational morphology. This type of account fits easily into any framework that distinguishes degrees of productivity in derivational morphology, e.g. Aronoff (1976). The idea that assignment to conjugation class can be part of derivational morphology is discussed in detail in Aronoff (1994), for both Israeli Hebrew and other languages, especially Latin. For example, the Latin intensive verb is formed by taking the past participle stem of a verb, e.g. canere- ‘sing’, whose past participle stem is cant-, and using it to form a verb of the first (-a:) conjugation: cantare. The desiderative verb suffix ur-, by contrast, forms verbs of the fourth (-i:) conjugation (as in e:suri:re) from e:dere ‘eat’. Just like Latin conjugations, Hebrew (and Arabic) binyanim have no predictable meaning except when used to form new words. This is what lies behind Doron’s observation about verbs that are ‘not paired with another equi-rooted verb’ being idiosyncratic. Doron frames her account in terms of DM. But in fact, nothing in Doron’s account particularly supports DM over other theories and there is much in it that quite directly supports the observation that binyanim are inflectional classes that have been co-opted for purposes of derivational morphology, which goes as far back as at least Kuryłowicz (1962).
9.5 Findings from outside pure linguistic analysis
In this section, I will review very briefly findings on Semitic roots from allied fields: child language acquisition, aphasiology, and experimental psycholinguistics. Despite claims to the contrary, the overwhelming weight of evidence favours the sort of position that I am advocating here: Semitic roots are primarily morphological in nature. Whether they have constant meaning or not is beside the point.
9.5.1 Evidence from child language acquisition
The leaders in the study of native acquisition of Hebrew and Arabic are Ruth Berman and Dorit Ravid. Berman and Ravid agree that ‘Hebrew-speaking children (p.177) can perform consonantal root extraction from as young as age three both in interpreting and producing novel verbs based on familiar nouns and adjectives’ and that the novel verbs that they produce conform to the standard morphological patterns (Berman 2003: 274). Ravid (2003: 303) concludes: ‘Developmental studies indicate that the basic ability to manipulate nonlinear structures emerges early on in Hebrew speakers.’ By kindergarten, ‘young Palestinian Arabic speakers were easily able to identify pairs of words related through the root, and also to provide another word from the same root’ (pp. 306–7).
9.5.2 Evidence from aphasia
Prunet et al. (2000) present a case study of a French–Arabic bilingual aphasic who showed root-sensitive metathesis in Arabic but no similar behaviour for French. They conclude that ‘Arabic roots can be accessed as independent morphological units’ (p. 610). Prunet et al. suggest that this finding supports a ‘morpheme-based theory that forms words by combining roots and templates’ (p. 609). Goral (2001) surveys a number of case studies of aphasia in Hebrew speakers (including one covered by Prunet et al. 2000). She concludes that ‘these two abstract morphemes [roots and patterns] have a psychological reality for Hebrew speakers’ (2001: 309).
9.5.3 Psycholinguistic evidence
In the last twenty years, there have been many psycholinguistic studies on Hebrew root and pattern morphology, using a variety of experimental paradigms: Feldman et al. (1995), Frost et al. (1997), Deutsch et al. (2000), Goral and Obler (2003), Shimron (2003), Berent and Shimron (2003), Velan et al. (2005), Bick et al. (2010). Prunet (2006) presents a valuable overview of work on Semitic languages more broadly. The overall conclusion to all this work is best expressed in the words of Joseph Shimron: ‘the sensitivity to the roots is not necessarily affected by their meanings. Rather, the root as a morpheme has an independent effect unconstrained by semantic mediation’ (Shimron 2003: 20). This conclusion applies equally well to the findings from child language acquisition and aphasia: speakers of Semitic languages are very sensitive to roots, but there is no evidence that these roots have constant meaning, only that they play an important role in the morphology of these languages.
9.6.1 Semitic roots
Roots are real linguistic entities, but there is no value in assuming that all instances of a given root have the same meaning. Roots are instead most centrally part of the (p.178) autonomous morphological system of a language and they play as robust a role in the morphology of a language like English or Latin as they do in Hebrew or Arabic.
What is special about Arabic and Hebrew is less the roots than the complex ways in which these roots interact with the obligatory syllabic templates that undergird the conjugation system of verbs. There is also the large role that the conjugations themselves play in derivation, apart from affixation. In most languages, conjugations are distinguished from one another by having different affix patterns. In Semitic languages, the inflectional affixes are identical across conjugations, which are distinguished largely by complex syllabic templates, reminiscent of the ablaut patterns of Germanic languages, but much stricter in form, as Kuryłowicz (1973: 43) notes: ‘The Sem[itic] ablaut is an essential ingredient of the morphological structure of the verb.’
It is time now to close the circle. I started off by comparing the role of the root in two quite distinct types of theories, one in which roots have constant meanings, of which DM is the modern archetype, and one in which roots play a purely morphological role, need not have constant meanings, and often don’t. I then searched for the historical origins of the theory in which roots are meaningful by definition. Although I had previously long believed that these origins lay in Arabic grammar, I have become convinced that the grammatical tradition in which roots first played the fundamental role that we find in modern theories was that of the medieval Hebrew grammarians.
Behind this search, though, lies a more fundamental issue. The idea that roots and other morphemes have constant meaning has been axiomatic in modern linguistics ever since the term morpheme was first coined by Baudouin de Courtenay (1895). Modern generative phonology, beginning with McCarthy (1981), adopted traditional Hebrew morphology almost wholesale. And though McCarthy’s phonological analysis did not depend at all on Semitic roots having a constant meaning, that traditional notion has been imported into the theoretical literature along with the rest of the framework.
Meanwhile, though, another group of linguists began to work about a half century ago on descriptions of the morphology of Semitic languages unfettered by traditional notions, notably structuralists like Cowell (2005) working on Arabic dialects and the group of Israeli linguists who deliberately approached Modern Israeli Hebrew with an open mind rather than from the point of view of traditional Hebrew grammar, including Shlomo Ariel, Aaron Bar-Adon, Haim Blanc, Ruth Berman, Shmuel Bolozky, Uzi Ornan, and Ora Schwartzwald, to name only a few. What they and their successors discovered was that the root and binyan system of these languages was more honoured in the breach than in the observance, and that it constituted a system of partial regularities rather than the perfect universe that had been depicted in traditional grammar, what Berman (1978: 100) calls a ‘compromise view […] (p.179) most appropriate to the semi-productive nature of the binyan system in contemporary Hebrew’. Others, notably Waltke and O’Connor (1990), have extended similar observations to Biblical Hebrew, where the binyanim are more regular, but still not nearly as systematic as the received tradition would lead one to expect.
In other words, if one looks at Semitic root and binyan systems as they actually function within real languages, rather than through the lens of a tradition that dates back a millennium, they turn out to manifest the sort of partial and default regularities characteristic of complex morphological systems that have become familiar to morphologists over the last forty years, rather than the invariant system so dearly desired by theoreticians since Baudouin de Courtenay. But we should not despair, because in these forty years we have developed an arsenal of tools meant precisely to attack these partial regularities in very precise ways as a network of ordered defaults (Aronoff 1994, Brown and Hippisley 2011). Yes, Semitic morphology is unusual, but it is not outside the universe of morphological systems that have occupied morphologists’ attention of late. Furthermore, although roots are central to Semitic morphology, the roots of Semitic are in many ways analogous to those of English or Latin, inasmuch as their most interesting properties are purely morphological and have nothing to do with meaning, except in so far as they are realized through a lexeme. They are not identical, but the ways in which Semitic roots, English roots, Latin and Romance roots differ from each other will provide grist for the mills of theoretical concepts like flexeme, lexemic index, and paradigm identifier that have been proposed over the last decade.
9.6.2 Roots, morphemes, morphomes, and lexical categories
The most remarkable property of roots in any language is their formal stalwartness in the face of semantic variation. As I noted above, English irregular verbs like run always have the same irregular forms, no matter how wide the variation in meaning across senses. The same is true of suppletive Latin roots like fer and the weak roots of Hebrew. Much less remarked but more remarkable is the fact that this formal unity in the face of semantic variation is not a special property of irregular roots or roots in general; it is the hallmark of morphological units of every kind, including those that show no variation in form at all. Consider English derivational suffixes, like -er, -ic, and -y. If we peruse a reverse alphabetical dictionary and consider the meanings of all words ending in any one of these suffixes, it quickly becomes apparent that there is tremendous variation in the set. But their form remains constant, so much so that this formal constancy is never noticed. English derivational suffixes differ greatly from one another in productivity, as derivational affixes do in any language, and these differences are correlated closely with what I have called elsewhere semantic coherence (Aronoff 1976), the extent to which the meaning of any given word containing a particular suffix is predictable (Baayen 1992, Plag 1999, Bauer 2001). (p.180) Where they do not differ is in what we might call formal coherence. Between the most productive affixes such as -ness and the least productive like -th there is no difference in formal coherence, despite the gulf in semantic coherence. The constant heart of linguistic morphology is the form of words and this extends to roots as it does to all morphological elements. Even Alec Marantz seems now to agree, at least if I interpret one of his recent statements correctly:
This view of the relation between roots and lexical categories is generally consistent with the insights of ‘lexeme-based’ morphological theories such as Aronoff’s (2001) [sic]. The word root plays an important role in an Aronoff-style theory as well as the present one, capturing facts associated with families of words sharing the same root. However, the meaning contribution of a root is never independently realized within this version of Distributed Morphology, since the objects of interpretation are the phases, not the roots. A root with a category-determining head corresponds to the lexeme of Aronoff’s system, and such a constituent can have a particular meaning, a variety of uses, and a history, as Aronoff makes clear. (Marantz, no date: 5–6)
(⋆) This chapter originated with a presentation at a workshop on roots organized at the Institute for Advanced Study at Hebrew University, Jerusalem, in June 2011. I thank the organizers of that conference and the participants who provided feedback. Olivier Bonami provided very helpful comments on an earlier version. I owe a great debt to Bob Hoberman, who taught me what little I know about the Arabic grammatical tradition and about Arabic morphology in general.
(1) They coined the term morpheme, among many others, and were the first to use phoneme in something like its standard sense. Stankiewicz (1972) is an anthology of Baudouin de Courtenay’s work, with some discussion of Kruszewski’s influence on his teacher. Kruszewski’s work is anthologized in English in Kruszewksi (1993). See Jakobson (1971b) for an assessment of the contributions of both.
(2) Halle notes, incidentally, that ‘certain words presuppose the existence of other words’ (1973: 13), a harbinger of lexeme-based theories of morphology.
(3) It is not clear to me whether DM Vocabulary insertion differs at all from morphological realization rules or constraints. Both match phonological material with morphosyntactic information. DM allows for the morphology to manipulate the syntactic tree in various ways and even add structure before reaching Morphological Structure, but there is nothing to prevent other theories from doing the same if they like.
(4) Earlier grammarians whose work has been lost or was never written down may have influenced Sībawayhi.
(5) Here Ɂaṣl might best be interpreted as ‘base’ or ‘source’. Recall that the direction of derivations between the maṣdar and the verb was a major aspect of the war between the Basrans and the Kufans.
(6) The word ʃɔrεʃ in the sense ‘root of a plant’ is attested in Deuteronomy, which was probably composed about the seventh century BCE, and the word has since been in continuous use in various related senses.
(7) This term is clearly related to the Arabic term bināʔ and the two concepts are very similar. Compare also Arabic wazn ‘weight, measure’, used interchangeably with bināʔ.