¹ They are chapters 37–41 (Būlāq ed., I, 82–108, Derenbourg ed., i, 70–88). Ch. 40 is not considered here, however, as it deals only with the similarities in behaviour between the agent noun and the verbal noun (maṣdar).

² al-ṣifatu l-mushabbdhatu bi-l-fā‘ili fīmā ‘amilat fihi; the full extent of the resemblance is easily discernible in the table below, and those familiar with the methods of the Arab grammarians will recognize the excellent specimen of qiyās (analogical reasoning) provided by these chapters.

³ Notwithstanding that later grammarians do detect examples of the pattern *ḥasanun wajhan, e.g. Zamakhsharī, Mufaṣṣal, ed. Broch, J. P., 1879, 101Google Scholar.

⁴ In the Kitāb this is purely an anthropomorphic metaphor, but it is interesting to observe that similar notions are emerging in modern linguistics, e.g. ‘dependency grammar’ (cf. Droescher, W. O. in Ellis, R. (ed.), Australasian Universities Language and Literature Association. Proceedings and papers of the thirteenth congress, held at Monash University, 12–18 August 1970, Melbourne, 1970, 326–9)Google Scholar.

⁵ The ghost has long since been laid that these terms have anything to do with the Latin idea of governance or régime in grammar: v. Weiss, J., ‘Die arabische Nationalgrammatik und die Lateiner’, ZDMG, LXIV, 1910, 349 fGoogle Scholar.

⁶ It no longer seems desirable to use the old Latin case names nominative, accusative, and genitive. They are replaced in this paper by ‘independent’, ‘dependent’, and ‘oblique’ respectively, names which at least have some correspondence with the syntactic functions they denote.

⁷ It is thus possible to conclude that sabab is the semantic bond between all elements that are syntactically Connected but are neither included in nor identical with each other. It follows that a correct utterance cannot contain elements that are unconnected with each other both semantically and structurally, which is precisely what Sībawaihi means when he describes such dis-located utterances as muḥāl ‘wrong’, ‘when the end of your utterance contradicts the beginning’ (v. ch. 6, Būl. ed., I, 8; Der. ed., I, 7).

⁸ e.g. Būl., I, 295; Der., I, 254.

⁹ This would seem to contradict Wehr, who regards it as ‘most unlikely’ that ’’ƒḍalu is felt to be undefined in ’’ƒḍalu rajulin (Wehr, H., Der arabische Elativ, Wiesbaden, 1953, 582)Google Scholar. Būl. 1, 105; Der., 1, 85.

¹⁰ Thus the constructions of the numbers 3–10 and 100 upwards are ‘improper’ annexations and the explanations for ‘ishruna apply equally to them.

¹¹ But by anticipating himself Sībawaihi surely shows that he regarded the Kitāb as a coherent whole, and did not merely stumble aimlessly from topic to topic.

¹² It is, of course, possible to say ’inna fīhā Zaydan wa- ‘Amran ’adkhaltuhu if ‘Amran is intended as the preposed object of ’adkhaltu, but that does not concern Sibawaihi here.

¹³ For example, the second element of the ‘true’ annexation construction is said by Khalil to be equivalent to tanwīn (Būl., 1, 323; Der., 1, 281), also the feminine suffix and the second element of compound words such as ḥaḍramawtu (Būl., n, 12; Der., n, 12) and some other phenomena which would benefit from a more detailed study.

¹⁴ See p.488, n. 8.

¹⁵ The phrase ‘correlated syntactically’ is an attempt to render maḥmūlan ‘alā mā ḥumila ‘alayhi, which is easier to grasp than to express in English.

¹⁶ Sībawaihi has slipped here: he ought perhaps to have mentioned that it is only the grammatical effect of ‘ishrūna upon dirhaman which is relevant, since the non-identity principle cannot possibly apply to Zaydan, who is the same as ’aḥadin, cf. the next item.

¹⁷ It would be unfortunate if readers of Rabin's, C.Ancient West-Arabian, London, 1951, 181Google Scholar, were to connect his expression ‘logical category’ with Sībawaihi's term naw‛ in this passage, as there is not the slightest reason to believe that Sībawaihi made use of logical categories in the Kitāb or, indeed, knew of their existence.

¹⁸ The work of Reuschel, W., Al-Halīl ibn-Aḥhmad, der Lehrer Sībawaihs, ah Grammatiker, Berlin, Akademie-Verlag, 1959Google Scholar, if anything, makes it more difficult to separate the master from the pupil.

¹⁹ cf. p. 491, n. 13.

²⁰ It is significant that Ḵẖalīl plays almost no part in the argument around ‘ishẖrūna dirhaman, being cited only twice in all five chapters (Būl., 1, 84, 99; Der., 1, 71, 81), and then only in a marginal capacity.

²¹ The structure of this phrase is unique, but it is clearly convenient to attach it to ‘ishẖrūna dirhaman (which is what Sibawaihi does after discussing the numerals, Būl., 1, 107; Der., 1, 87).

²² With the outstanding exception of Ibn Jinnī.

²³ Along with the concept of ƒā’'ida ‘informativeness’ and jumla (with its prescriptive con-notation of ‘sentence’), ƒaḍla is one of several terms generally regarded as typical of Arabic grammar which are not found anywhere in the Kitāb: what interested Sibawaihi was ‘good’ (hasan) and ‘right’ (mustaqīm) speech (kalām).

²⁴ N. 8 to ch. 110 in Jahn's, Sībawaihi's Buch über die Grammatik, Berlin, 1895–1900, referring to Būl., 1, 230; Der., 1, 262Google Scholar.

²⁵ Rabin, , op. cit., 183–4Google Scholar, referring to Būl., 1, 192–5; Der., 1, 161–5. The same non-identity principle enables Sībawaihi to observe that ‘in lahu ṣawtun ṣawtu ḥimārin you know very well that the voice of the donkey is not the same as the previous voice, but ṣawtu is allowed to be independent in form by latitude of speech (sa‘atu l-kalām) ‘(Būl., 1, 182; Der., 1, 152).

²⁶ Another attempt has been made by the present writer, and, it is hoped, will appear shortly under the title ‘An Arab grammarian of the eighth century A.D.’.