On Sandhi in Pali
by the late
R. C. Childers
revised with additions and corrections
by
Ānandajoti Bhikkhu
Originally Published by Journal of the Royal Asiatic Society,
Volume 11, Issue 1, January 1879, pp. 99–121.
Introduction
The rules of sandhi in Pāḷi are the effect of the spoken word and a living, spoken—not literary—language of its time. All Prakrits have similar features: unpredictability, and a confusion of mights and maybes. Childers’ work on the subject was made when only a small number of works had been published, when rules for presentation were still in flux, and when even the way we represent Indian words and syllables had not been standardised. It is still to this day the only work I know of its kind in the grammatrical literature. AJ: K R Norman did write a paper on External Sandhi in Pāli, JPTS, Vol XIX, 1993 pp. 203-213, but that was mainly focused on the Suttanipāta, not a general assessment.
In preparing this work I have changed the formatting so as to be more regular, and have rewritten some of the signs, so they can be more easily understood today, and accord with modern standards. In this regard the circumflex over the vowels â, î and û have been replaced with the macron: ā, ī and ū. In Sanskrit what we now write as ś was represented by ç, the ṣ was written as sh and ṛ was written as ri.
In presenting the work Childers has italicised the Pāḷi words in the General Remarks, but thereafter forewent doing so. In this edition in order to make the examples in Pāḷi stand out I have emboldened them. Italics, where they remain, are for foreign words in English and emphasis. Sanskrit was not italicised in the original, but it is here.
Childers usually noted the elision with an apostrophe, as in n’ atthi, n’ accuṇho, pan’ aññaṁ. In most texts these days we do not mark the elision—as is the standard of this site—so I have omitted them: natthi, naccuṇho, panaññaṁ, except in the case where proper names are involved: pañcah’ Upāli, āyām’ Ānanda, which is simply to show the names more clearly, and save having a capital letter mid-word.
It is not usually the practice elsewhere, but on this website I indicate where sandhi has affected a change in one of the sounds by including a hyphen. For instance where Childers wrote: evam me, evam āha, evañ ca, I prefer to write evam-me, evam-āha, evañ-ca. This clarifies the effect of the sandhi on the phonology.
In the case of writing ti, I prefer to separate it from the word quoted (as in pucchatī ti), because the lenghtening does not seem to be due to sandhi (i.e. pucchati+iti), but owing to the particular case of quotations bringing about lengthening of the final vowel of the quoted word, as if to separate it off clearly. If it was a combination with iti we would expect to find such combinations as pucchāma iti giving pucchāmeti, and pucchaṁ iti giving puccham-iti, but these forms are not seen, except, in the last case, occasionally in verse, metri causa.
When a glide consonant has been inserted for euphony, then I usually separate it off, e.g.: na-y-idam=na idaṁ, na-y-ime=na ime, tava-y-idaṁ=tava idaṁ, idha-m-āhu=idha āhu, idha-m-eva=idha eva.
Many of the sandhi forms listed below come from verse, and only verse. It seems than in Pāḷi, metre trumped phonology, meaning that if the metre demanded it, then all sorts of irregular forms were allowed, whereas the reverse was the case in Sanskrit which strictly applied the sandhi rules and versifiers had to work around that issue. AJ: It is interesting to look at partially Sanskritised texts like the Udānavarga, and the efforts made to Sanskritise an underlying Prakrit.
Original page numbers are given within square brackets [], and occasional numbering mistakes have been silently corrected. I have also corrected the occasional misspelling, and have completed abbreviations and written out the examples in full wherever it seemed it would be helpful to the student.
I have occasionally marked the text in red when I believe that Childers had been misled, or made a mistake. I have checked these against the CST4 corpus, but not against the originals he was quoting from, which are now hard to find.
Some of the examples Childers gave were from medieval grammars which seem to have covered all possible forms, but they are not necessarily found in any of the texts. They can be considered as being comprehensive as to possibility, rather than to actual usage. A similar thing has happened in the medieval prosodies. If I have found an example that only occurs in the grammars, I mark it with
My own annotations are preceded by the initials AJ, otherwise they are Childers own. If anyone notices a mistake in the transcription—which is not unlikely in such a complex document—please do contact me, so that corrections can be made.
Because of the changes of presentation the document should not be quoted for Childers actual writing, but reference should be made to the original.
Sanskrit Sandhi
Childers brings out well the difference between Sanskrit and Pāḷi sandhi, but he gives no overview of the former, which I feel would be useful to the student. In literary classical Sanskrit the rules are very regular, and always applied. It helps, I think, to keep in mind that in Sanskrit, sandhi is mandatory in saṁhitā (continuous text) because the language treats the sentence—not the word—as the primary unit of phonology.
1. Vowel Sandhi (Ac Sandhi)
When two vowels meet, they must merge or transform based on their quality.
Savarṇa Dīrgha (Homogeneous): Two similar vowels (short or long) merge into their long form. Example: a-a=ā.
Guṇa: a/ā followed by i, u, ṛ results in e, o, ar. Example: a-i=e.
Vṛddhi: a/ā followed by e, o results in ai, au. Example: a-e=ai.
Yan: i, u, ṛ, ḷ (when followed by a dissimilar vowel) transform into their corresponding semivowels y, v, r, l. Example: i-a=ya.
Ayādi: The diphthongs e, ai, o, au transform into ay, āy, av, āv when followed by any vowel. Example: e.g., śre + anas =śrayaṇas.
2. Visarga Sandhi
The visarga (ḥ) changes based on the following sound:
Before Voiceless Stops: Before ca/cha, it becomes ś. Before ṭa/ṭha, it becomes ṣ. Before ta/tha, it becomes s.
The "O" Rule: If a word ends in –aḥ and is followed by a voiced sound (including vowels and voiced consonants), and before a, the –aḥ becomes –o. Example: gajāḥ gacchanti becomes gajo gacchanti.
Ropa (Deletion): If a word ends in –āḥ followed by any voiced sound, the visarga simply disappears.
3. Consonant Sandhi (Hal Sandhi)
Sanskrit makes much use of regressive assimilation (the second sound influences the first).
Schwa-like Assimilation: A dental t will transform to match the place of articulation of the following sound (t-l=ll, t-j=jj, t-c=cc).
Anunāsika: A stop becomes a nasal if followed by a nasal. Example: d-m=nm.
Anusvāra: A final m becomes an ṁ (anusvara) if followed by any consonant, but remains m if followed by a vowel.
Crucially, Sanskrit sandhi operates across all word boundaries in a sentence (saṁhitā), treating the sentence as a single phonological unit. Pāḷi, as Childers shows, treats the word as a more independent unit.
These are simple rules, mainly governed by ease of pronunciation, and they are easily learned, and what is more they are consistently applied—this is a major difference between Sanskrit and Pāḷi and other Prakrits.
Ānandajoti Bhikkhu
April 2026
A Summary List of the Rules
The number given in round brackets is the number of the paragraph where Childers gave the rule. Note Childers restarted the numbering when he came to compound sandhi, so you have to know which section the rule comes under.
V = vowel; C = consonant.
Word Sandhi: Vowel Sandhi
a + a = ā (1)
a + aCC = a (2)
a + a = ā (before double consonant, rare) (3)
a + a = a (elision, rare) (4)
a + i/ī = e (5)
a + i = a (elision, rare) (9)
a + i = a (elision) (10)
a + ī = a (elision) (11)
a + u = o (6)
a + ū = ū (rare) (8)
a + u = a (elision) (10)
a + ū = a (elision) (11)
a + e = a (elision) (11)
a + o = a (elision) (11)
a + iti = ācc (7)
ā + a = ā (12)
ā + aCC = ā (13)
ā + a = ā (before simple consonant, rare) (14)
ā + i/ī = e (5)
ā + i = ī (15)
ā + ī = ā (elision) (12)
ā + u = o (6)
ā + ū = ā (elision) (12)
ā + u = ā (elision, rare) (13)
ā + e = ā (elision) (12)
ā + o = ā (elision) (12)
i + V = i (elision) (16)
i + a = i (elision) (18)
i + a = ā (crasis) (20)
i + i = ī (19)
ī + V = ī (elision, rare) (17)
ti/ttī + V = ty + V (21)
iti + V = cc + V (23)
iti + V = v + V (rare) (22)
di + V = jj + V (24)
api + V = app + V (25)
u + V = u (elision) (26)
u + V = u (elision of following short vowel) (27)
u + i = ū (28)
u + V = v + V (29)
e + V = e (elision) (30)
e + VCC = e (elision) (30)
e + V = e (elision) (31)
e + a = ā (32)
e + a = yā (33)
e + aCC = ya (34)
e + aCC = yā (rare) (35)
o + V = o (elision) (36)
o + V/CC = o (elision) (37)
o + a = ā (38)
o + a = vā (39)
o + aCC = va (40)
o + aCC = vā (rare) (41)
o + V = vV (42)
Word Sandhi: Euphonic Insertions
V + idaṁ/ime = V-y-idaṁ/ime (43–44)
V + iva = V-viya (45)
V + u/ū = V-v-u/ū (46)
V + V = V-m-V (47)
V + iva = V-r-iva (48–49)
V + V = V-n-V (rare) (50)
V + V = VC-V (revival of lost consonant) (51)
V + V = VC-V (revival of wrong consonant) (52)
Word Sandhi: Mixed Sandhi
V + CC = V + CC (doubling) (53)
V + C = VC + C (revival of lost final) (54)
ṁ + V = m + V (54)
V + C = Vm + C (metri causa, rare) (54)
Ch + Ch = C + Ch (aspirate loss) (55)
ṁ + V = n + V (rare) (56)
ṁ + V = C (historical) + V (56)
ṁ + V = C (analogical) + V (57)
ṁ + V = V (elision in verse) (58)
Vṁ + V = V (elision in verse) (59)
aṁ + a = ā (60)
ṁ + eva = ṁ-y-eva (61)
ṁ + eva = ññ-eva (61)
ṁ + iva = viya (61)
ṁ + V = V (elision in late texts) (61)
Abbreviations
I was not always sure what text or edition the abbreviations referred to as there was no key given in the article. Some have therefore been drawn from Childers’ Dictionary where there was a key. This is probably unimportant as most of the texts he was referring to are hard or impossible to find these days, but these are the ones mentioned in the work.
Abhidh. = Abhidhānappadīpikā, probably Ven. Subhūti’s edition.
Alw. = James d’Ālwis, probably his Introduction to Kaccāyana.
Aṭṭh. = Aṭṭhakathā
Cl. Gr. = A Compendious Pāli Grammar. By the Rev. Benjamin Clough. Colombo, 1824.
Das. = The Dasaratha-Jātaka. Edited by V. Fausböll. Copenhagen, 1871.
Dh. = Dhammapada (pg., not number)
Dict. art. = Article in Childers Dictionary
Ev. = Maybe = Gog. Ev.?
F. Jāt. = Faüsboll, Jātakatthavaṇṇanā
Gog. Ev. = The Evidences and Doctrines of the Christian Religion. By the Rev. D. Gogerly. Part I. On Buddhism. Colombo, 1862.
Gov. Ev. = Government Evaluation ?
Jāt. = Faüsboll, Jātakatthavaṇṇanā
Kacc. = Kaccāyana, probably d’Alwis’ edition
Mah. = The Mahawanso. Edited by the Hon. George Turnour, Esq. Colombo, 1837.
Par. = probably Mahāparinibbānasutta (DN 16)
Pay. S. = Probably Pāyāsi Sutta, DN 23
Ras. = Anecdota Pālica. Rasavāhiṇī, Uraga Sutta. Edited by Dr. F. Spiegel. Leipzig, 1845.
Raṭṭh. S. = Raṭṭhapāla Sutta, MN 82
S. = Sutta
Sansk. = Sanskrit
Sen. = Kaccāyanappakaraṇaṁ. Edited by Émile Senart. Journal Asiatique, Mars—Avril, 1871, and Mai—Juin, 1871.
Sutta N. = Sutta Nipāta
Ten J. = Fausboll, Ten Jātakas
On Sandhi in Pāḷi
by the late
R. C. Childers
[The following fragments were found among the papers of the late R. C. Childers. When his last illness overtook hin, he was known to have been engaged for some time on a comprehensive handbook of the Pāḷi language. Before, however, he had far proceeded with the elaboration of the greater work, he proposed to bring out an elementary grammar first, to meet the pressing want of intending students of that language. After his death, it was hoped that some other Pāḷi scholar would be found willing to complete the work in either form from the materials collected by Mr. Childers: and with this view these were submitted to Prof. Pischel at Kiel. But as this gentleman has since expressed his intention of publishing a grammar on a plan of his own, it is thought that the following two fragments may, even as such, be of interest to Pāḷi students. Each of them treats of the Sandhi rules, and is complete in itself. But while the former, intended for the shorter work, was left ready for the press, the latter, intended for the larger, had not yet received the benefit of the gifted author’s revision.—R. R. I have no key to whom these initials refer, but I very much suspect it was Reinhold Rost, librarian of the India Office Library.]
General Remarks
Sandhi [100] as a branch of Grammar requires very different treatment in Pāḷi and in Sanskrit.
The whole of the important division of Internal Sandhi is wanting in Pāḷi—at least it comes under the domain not of Grammar but of Philology. It would not only be a misapplication of labour, but positively misleading, to work out rules of internal Sandhi from, for instance, such forms as sabbhi and lacchate. Our only proper course is to trace them to their Sanskrit originals sadbhis and lapsyate, and bring them under rules, not of Sandhi, but of phonetic change.
We have, therefore, only to deal with external Sandhi, but here we find the most important differences between Pāḷi and Sanskrit. Pāḷi Sandhi is almost wholly independent of Sanskrit Sandhi, and cannot be brought under Sanskrit rules. At first sight combinations like natthi, dukkhassantaṁ, may appear to be merely phonetic corruptions of the Sanskrit nāsti, duḥkhasyāntam; but this view is soon seen to be untenable, since in the great majority of cases a Pāḷi sandhi-change differs from the corresponding Sanskrit sandhi-change. Thus we have pañcime for Sanskrit pañceme, sādhūti for sādhviti, sammad era for saṁyageva, kiñcāpi for kiñcid api, yatharira for yathaiva, and innumerable others opposed to Sanskrit usage.
In Sanskrit sandhi is imperative, in Pāḷi it is to a great extent optional: between separate words it takes place but seldom, and even in compounds hiatus occurs. Again, while sandhi is regular and uniform in Sanskrit, in Pāḷi it is very irregular. For example, while in Sanskrit na upeti must always become nopeti, in Pāḷi it might become nopeti, or nupeti, or nūpeti, or remain na upeti without sandhi change taking place.
Every word in the Pāḷi language without exception, if written separately, ends either with a vowel or anusvāra. This can easily be seen by inspecting a Pāḷi dictionary. But in a sentence a word occasionally ends in a consonant, owing to a lost final consonant being restored for the sake of euphony. Thus the Sanskrit abhūt ‘he was,is ahu in Pāḷi; but if ahu is followed by eva, the lost consonant may be revived to avoid hiatus, and we get ahud eva, which in Sanskrit would be abhūd eva. Other examples are given further on in § 51.
Sandhi may be divided into Vowel Sandhi, Consonant Sandhi, and Mixed Sandhi. The Pāḷi rules of vowel sandhi are numerous enough, but consonant sandhi, which in Sanskrit forms so important a branch of the subject, in Pāḷi is limited to cases of contact between anusvāra and a following consonant. The reason for this is that in Pāḷi, as we have just said, all words which do not end in anusvāra end in a vowel. Thus in Sanskrit we are told that pūrṇaḥ followed [101] by ca becomes pūrṇaś ca; but the Pāḷi equivalent of pūrṇaḥ is puṇṇo, which, of course, remains unaltered before a consonant. Mixed sandhi has two divisions, viz. when a word ending in a vowel is followed by a word beginning with a consonant, and when a word ending in a consonant is followed by a word beginning with a vowel. Cases of mixed sandhi are very few, and in the second division are limited to the changes of anusvāra before a vowel. This is, again, because anusvāra is the only consonant with which a word can terminate. Cases like ahud eva come under vowel sandhi.
We have seen that in Pāḷi we have external sandhi only to deal with. I propose to separate external sandhi into the two great divisions of the Sandhi of Words and the Sandhi of Compounds. The distinction is a real and important one, as the rules of the two divisions differ in many respects.
Sandhi of Words
In Sanskrit word-sandhi is imperative, in Pāḷi it only takes place in certain cases. To take a representative text, in the first page of Mahāparinibbāna Sutta there are nearly thirty instances in which a word-sandhi change does not take place when it would have been imperative in Sanskrit. The first five are samayaṁ bhagavā, māgadho ajātasattu, vajjī abhiyātukāmo, ahaṁ ime, vajjī evaṁmahiddhike. In the whole page word-sandhi only occurs nine times, viz. in evam-me, evam-āha, tenupasaṅkama, pucchatī ti, evañ-ca, bhaṇantī ti, pattikova, tenupasaṅkami, etad-avoca.
In prose, word-sandhi is almost confined to indeclinables and pronouns. Thus at Par. 22 we have sandhi only in connexion with yaṁ, nūna, ahaṁ, iti, tena, etaṁ, kāci, kiñci, eva, pana, tattha, mama, ha, and kathaṁ. In a great many cases of sandhi we find two pronouns brought together, as myāyaṁ=me ayaṁ, or two indeclinables, as yan-nūna, or an indeclinable and a pronoun, as nūnāhaṁ=nūna ahaṁ, tasseva=tassa eva. But frequently, also, the indeclinable or pronoun is connected with a verb or noun, as in tatth’ Ānanda, purisam-pi, etad-avoca.
Even with indeclinables and pronouns sandhi only takes [102] place in certain cases. Thus at Par. 22 we have ahaṁ imaṁ, taṁ ābādhaṁ, so ābādho, kho āyasmā, and many others. Some words cause sandhi change much more certainly than others. Thus there is scarcely any exception to the rule that anusvāra before the particles ca, ti and pi is modified to ñ, n and m (e. g. kathañ-ca=kathaṁ ca). Eva almost always causes or suffers sandhi change, e.g. kiñcid-eva=kinci eva, pattikova=pattiko eva. The same may be said of iti, but at Par. 9 we have an exception, bhavissanti iti. AJ: see the note in my introduction regarding iti/ti. Na followed by a vowel is generally modified, e.g. natthi, nevra, nāhosi, though exceptions occur, as na evaṁ (Par. 22), na idāni (F. Jāt. 12). On the other hand, forms like nakkhamati, nappahosi, are far less common than na khamati, na pahosi.
I have said that word-sandhi is in prose almost confined to indeclinables and pronouns. The limited number of exceptions will generally be found to be cases in which either (1) a vocative beginning with a vowel is preceded by a word ending in a vowel, as gacchāvuso, pañcah’ Upāli, āyām’ Ānanda; AJ: = gaccha āvuso, pañcahi Upāli, āyāma Ānanda. or (2) when a verb is preceded or followed by a noun in grammatical relation with it, as utthāyāsanā, āsanā vuṭṭhāya, upajjhāyassārocesuṁ (Dh. 103) AJ: Dh. is normally an abbreviation for Dhammapada, but by page number, not verse number. What edition is unknown. Here it seems to be a mistaken reference though. , sametāyasmā, āgamhā; or (3) when two nouns are in grammatical relation, as dukkhassantaṁ, dvīhākārehi (Dh. 91). There is a small residuum of cases not coming under these three categories, e.g. at Dh. 337, sandhāyāha.
But even under the three categories sandhi only sometimes takes place, e.g. we have gamissāma Ānanda (Dh. 170), vajjī abhiyātukāmo.
Sandhi is more extensively used in the early texts of the Tipiṭaka than in the late texts of the Commentaries.
In verse, word-sandhi is much less restricted and much more frequent than in prose, being in great measure governed by the question of metrical exigency. Thus in the first two pages of Dhammapada there are nine sandhis, of which only two, nādhigacchanti and nappasahati, would occur in prose. The remainder, for instance sammantīdha, vantakāsārassa, are used metri causā. Some of the bolder sandhis, as the elision of ṁ and aṁ, are confined to verse. [103]
Sandhi of Compounds
In Sanskrit, the rules of sandhi for words and for compounds are the same; in Pāḷi, they present several points of difference. Pāḷi compound words are of two classes—first, compounds, which are phonetic corruptions of corresponding Sanskrit compounds; and, secondly, compounds in which two Pāḷi words are independently combined, without reference to Sanskrit. To the first my remarks at p. 100 AJ: Written as p. 99 in the original. The pagination must have been changed before publication. on internal sandhi are applicable; they are by their nature excluded from the department of sandhi. Thus jaraggava cannot be brought under any Pāḷi rule of sandhi; all we can do is to trace it to an older Sanskrit form jaradgava, of which it is a phonetic corruption. On the other hand, kulitthi, at Par. 3, cannot be identified with a Sanskrit compound kulastrī, but is an independent combination of the Pāḷi form itthi with kula.
Rules of Word Sandhi
I. Vowel Sandhi
1. If a word ending in a is followed by a word beginning with a, the two short vowels sometimes coalesce into ā. Ex. nāhosi=na ahosi, nātisīto=na-atisīto, nāsakkhi=na asakkhi, nāyaṁ=na ayaṁ, panāyaṁ=pana ayaṁ, etthāyaṁ=ettha ayaṁ (Jāt. 8), vatāyaṁ=vata ayaṁ (Dh. 408), nāyaṁ=na ayaṁ, nāhaṁ=na ahaṁ (Ten J. 36), tatrāhaṁ=tatra ahaṁ, nāparaṁ=na aparaṁ, imassāpi=imassa api, cāhu=ca ahu, senāsantuttha=sena-asantuttha (Ten J. 30).
2. But if the initial a of the second word is followed by a conjunct consonant, the final a of the first word is usually elided. Ex. natthi=na atthi, naccuṇho=na accuṇho, panaññaṁ=pana aññaṁ, tatrassa=tatra assa, tatthaddasaṁ=tattha addasaṁ, yassatthāya=yassa atthāya, dukkhassantaṁ=dukkhassa antaṁ, kirassa=kira assa, tavantikaṁ=tava antikaṁ (Jāt. 28).
3. There are a few exceptions in which the two as coalesce before a double consonant, as nāddasa=na addasa, nāñño=na añño, nāssa=na assa (Dh. 23). [104]
4. Rarely a is elided before a, e.g. vatayaṁ=vata ayaṁ (Kuhn 60), cahaṁ=ca ahaṁ (Jāt. 3), nahosi=na ahosi (Dh. 155), upadhāyahaṁ=upadhāya ahaṁ.
5. a and ā may coalesce with i or ī into e. Ex. kokilāyeva=kokilāya iva, bandhusseva=bandhussa iva (Sen. 14), pajjotasseva=pajjotassa iva, netaro=na itaro, neresi=na īresi, yathedaṁ=yathā idaṁ.
6. Similarly a and ā may coalesce with u into o. Ex. nopeti=na upeti, nopagacchittha=na upagacchittha. (Mah. 28), alatthopaddhabhikkhūhi=alattha upaddhabhikkhūhi (Mah. 174), mamopamaṁ=mama upamaṁ (Ras. 29), noyāti=na uyyāti, pakkhanditodadhiṁ=pakkhanditā udadhiṁ (Mah. 117).
7. When a is followed by iti, we always get the crasis ā, e.g. pivathāc=pivatha icc (Jāt. 3).
8. a and ā with ū rarely become ū instead of o, as cūbhayaṁ=ca ubhayaṁ, tadūpasammanti=tadā upasammanti (Jāt. 18).
9. Very rarely a elides i. Ex. yename=yena ime (Par. 64), paname=pana ime (Cl. Gr. 9).
10. a is very frequently elided by i and u, whether followed by a conjunct consonant or not. AJ: This is one of the most common sandhis in Pāḷi prose, especially with ime, imaṃ, imesu, etc. Ex. passathimaṁ (Dh. 31), bhamarassiriyato, pahāyimaṁ=pahāya imaṁ, pañcime=pañca ime, ito cito=ito ca ito, tatridaṁ=tatra idaṁ, nūnimaṁ=nūna imaṁ, vatidaṁ=vata idaṁ (Jāt. 56), yassindriyāni=yassa indriyāni, yatthitthi=yattha itthi (Jāt. 155), tenupasaṅkami=tena upasaṅkami, nāmupaṭṭhāko=nāma upaṭṭhāko (Jāt. 29), cupeto=ca upeto (Jāt. 84), etenupāyena=etena upāyena, assuppajjati=assa uppajjati (Das. 36), nuppajji=na uppajji (Jāt. 15), nuddhaṁsati=na uddhaṁsati (Jāt. 18).
11. a is elided before ā, ū, e and o. Ex. yenāyasmā=yena āyasmā, yassālayā=yassa ālayā, uṭṭhāyāsanā=uṭṭhāya āsanā, idhāvuso=idha āvuso, ekenūno=ekena ūno, gaṇhathetaṁ=gaṇhatha etaṁ, cetarahi=ca etarahi, panesa=pana esā, nettha=na ettha, ivotataṁ=iva otataṁ, mamokāsaṁ=mama okāsaṁ (Jāt. 13), samuddenottharāpesuṁ=samuddena ottharāpesuṁ.
12. ā sometimes elides a short vowel, and less often a long [105] vowel. Ex. imāhaṁ=imā ahaṁ, tadāyaṁ=tadā ayaṁ, vāssa=vā assa (Jāt. 136), yathāyaṁ=yathā ayaṁ, yassāyaṁ (Dh. 118), kasmāsi=kasmā asi, disvāpanissayaṁ=disvā upanissayaṁ (Mah. 30), vihesāva=vihesā eva, sutvāva=sutvā eva (Das. 4), attanāva=attanā eva.
13. ā is often elided before a long vowel, or before a short vowel followed by a conjunct consonant. Ex. tadāsi=tadā āsi, māpādi=mā āpādi, tadeva=tadā eva (Mah. 244), tatheva=tathā eva, mahiyekarattivāso=mahiyā ekarattivāso. (Dhaniya S.), netvekamantikaṁ=netvā ekamantikaṁ, gorakkhetta (Mah. 80), tayajja=tayā ajja, tadassu=tadā assu (Jāt. 196), daṭṭhukāmamhā=daṭṭhukāmā amhā (Dh. 84), natthamhā=natthā amhā (Dh. 177), jitamhā=jitā amhā.
14. Very rarely ā is elided before a followed by a simple consonant. Ex. muñcitvahaṁ=muñcitvā ahaṁ (Jāt. 13).
15. ā followed by i sometimes gives the crasis ī. Ex. saddhīdha=saddhā idha, seyyathīdaṁ=seyyathā idaṁ.
16. i is very frequently elided before a vowel, whether the latter be short or long. Ex. gacchāmahaṁ=gacchāmi ahaṁ, icchāmahaṁ=icchāmi ahaṁ (Jāt. 53), dassentanappakaṁ=dassenti anappakaṁ (Jāt. 47), bhavataddharatti=bhavati addharatti (Gog. Ev. 23), pajja=pi ajja (Jāt 18), pañcahaṅgehi=pañcahi aṅgehi (Sen. 18), natthaññaṁ=na atthi aññaṁ (ib.), passasimaṁ=passasi imaṁ (Jat. 144), idānime=idāni ime (Jāt. 221), cattārimāni=cattāri imāni, yadimassa=yadi imassa (Jāt. 1, 17), These last four may, however, equally well be looked upon as examples of rule 18, and written passasimam, idānime, cattārimāni, yadimassa. dasahupāgataṁ=dasahi upāgataṁ, pañcah’ Upāli=pañcahi Upāli, dvīhākārehi=dvīhi ākārehi (Dh. 91), apāvuso=api āvuso, tāha=ti āha, sayanehāvasathehi=sayanehi āvasathehi (Brāhmaṇadhammikasutta), pesa=pi esa, teva=ti eva (Jāt. 53), hetaṁ=hi etaṁ, hevaṁ=hi evaṁ (Jāt. 169), hete=hi ete (Jāt. 22), phandanteva=phandanti eva (Das. 36), atthetassa=atthi etassa (Jāt. 55), pītunnatunnatā=pīti unnatunnatā. (Mah. 27).
17. Very rarely ī is elided. Ex. tuṇhassa=tuṇhī assa.
18. Not unfrequently i elides a following short vowel. [106] Ex. phalantisaniyo=phalanti asaniyo (Mah. 72), pihaṁ=pi ahaṁ (Jāt. 47), pissa=pi assa, itissa=iti assa (Sen. 15; Das. 2), hissa=hi assa (Das. 21), idānissa=idāni assa (Jāt. 195), kaccittha=kacci attha (Pāt. 2), kārayittharaṇaṁ.
19. i-i sometimes becomes ī. Ex. sammantīdha=sammanti idha (Dh. 2, see variæ lectiones), yānīdha=yāni idha. This is almost invariably the case when a word ending in i is followed by iti. AJ: See my comments in the Introduction.
20. i-a occasionally gives the crasis ā. Ex. kiñcāpi=kiñci api, pāhaṁ=pi ahaṁ (Jāt. 184), idānāhaṁ=idāni ahaṁ (Dh. 88).
21. Final i or ī preceded by t or tt and followed by a vowel may become y. If preceded by tt, one t is dropped. Ex. jīvantyelaka=jīvanti elaka (Abhidh. 513), tyayam=ti ayam (Alw. I. xl), gutyatha=gutti atha (Abhidh. 374), sabbavityanubhūyate=sabbavitti anubhūyate (Pay. S.), paṭisanthāravutyassa=paṭisanthāravuttī assa (Pay. S.)
22. In the case of iti we occasionally have v for y. Ex. itvevaṁ=iti evaṁ (Cl. Gr. 15), tveva=ti eva (Dh. 93, Das. 24, Ten J. 54).
23. In the case of iti the combination ty generally becomes cc, as iccevaṁ=iti evaṁ, iccanena=iti anena, iccetaṁ=iti etaṁ.
24. In one or two cases di before a vowel becomes jj. Ex. yajjevaṁ=yadi evaṁ (Cl. Gr. 14); here we must presuppose a transition form yadyevaṁ.
25. Api followed by a vowel may become app, which points to a transition form apy. Ex. appeva=api eva, appekacce=api ekacce, appekadā=api ekadā.
26. u is occasionally elided before a vowel. Ex. tusitesupapajjatha=tusitesu upapajjatha (Mah. 201), tadahuposathe=tadahu uposathe, sametāyasmā=sametu āyasmā, sādhāvuso=sādhu āvuso (Das. 22), tāseva=tāsu eva (Par. 15), tiṭṭhatetaṁ=tiṭṭhatu etaṁ.
27. Rarely u elides a following short vowel. Ex. nuttha=nu attha, nusi=nu asi (Dh. 96), kinnumā=kinnu imā.
28. u followed by i sometimes gives the crasis ū. Ex. kiṁsūdha=kiṁsu idha, sādhūti=sādhu iti. In the case of iti this crasis is invariable. [107]
29. There are one or two instances of final u changing to v before a vowel. Ex. vatthvettha=vatthu ettha (Sen. 16), sesesvayaṁ=sesesu ayaṁ (see Dict. art. Payyālaṁ).
30. e may be elided before a long vowel, or before a short vowel followed by a conjunct consonant. Ex. māsi=me āsi (Das. 3), meso=me eso (Gog. Ev. 46), vāsayettha=vāsaye ettha (Jāt. 93), sīlavantettha=sīlavante ettha, sacassa=sace assa (Dh. 156), matthi=me atthi (Pay. S.).
31. e sometimes elides a following vowel. Ex. teme=te ime (Jāt. 177), dveme=dve ime (Par. 48), sacejja=sace ajja (Ras. 84).
32. Occasionally we have the crasis e-a=ā. Ex. sacāhaṁ=sace ahaṁ (Dh. 140, 165; Ten J. 12; Jāt. 5), vāhaṁ=ve ahaṁ (Alw. N.).
33. When e is followed by a, it may become y, the a being at the same time lengthened. Ex. tyāhaṁ=te ahaṁ (Sen. 15; Ten J. 111; Jāt. 135), pabbatyāhaṁ=pabbate ahaṁ (Pay. S.), myāyaṁ=me ayaṁ.
34. But if the initial a is followed by a double consonant, it is not lengthened. Ex. tyajja=te ajja, tyatthu=te atthu.
35. There are one or two exceptions to the last rule, e.g. tyāssa=te assa; this is the right reading, it is also in Pay. Siddhi, and occurs in Sigālovāda S.).
36. o frequently elides a following vowel. Ex. sohaṁ=so ahaṁ, aggohaṁ=aggo ahaṁ (Jāt. 53), bhiyyobhiññataro=bhiyyo abhiññataro (Par. S.), yodha=yo idha (Dh. 47), cattārome=cattāro ime (Sen. 14), pañcamotthi=pañcamo atthi (Cunda S.), AJ: Read: pañcamatthi. ubhottha=ubho attha (Jāt. 165), khomhi=kho amhi, natthomhi (Dh. 199), nāsitosmi (Dh. 199), pattikova=pattiko eva, kattabboposathe=kattabbo uposathe (Mah. 220).
37. o is frequently elided before a long vowel, or a short vowel followed by a double consonant. Ex. kutettha=kuto ettha, yāvantettha=yāvanto ettha, asantettha=asanto ettha, namatthu=namo atthu, katamassa=katamo assa (Pay. S.), tayassu=tayo assu, bhiyyassa=bhiyyo assa (Das. 35), perhaps ajjatagge=ajjato agge, AJ: this seems rather to be ajja+t+agge. tatuddhaṁ=tato uddhaṁ (Jāt. 25). [108]
38. The crasis o-a=ā sometimes occurs. Ex. appassutāyaṁ=appassuto ayaṁ, dukkhāyaṁ=dukkho ayaṁ (Jāt. 168).
39. o followed by a sometimes becomes v, the a being at the same time lengthened. Ex. yvāhaṁ=yo ahaṁ, svāhaṁ=so ahaṁ, svāyaṁ=so ayaṁ, khvāhaṁ=kho ahaṁ (Sutta N.), yatvādhikaraṇaṁ=yato adhikaraṇaṁ.
40. But if a is followed by a double consonant, it is not lengthened. Ex. khvassa=kho assa, svassa=so assa, yāvatakvassa kāyo tāvatakvassa byāmo=yāvatako assa kāyo tāvatako assa byāmo (Pay. S.), khvajja=kho ajja, kvattho=ko attho.
41. There are two or three exceptions to the last rule in which the a is lengthened. Ex. svāssu=so assu (Jāt. 196), khvāssa=kho assa (Pay. S.).
42. Sometimes o becomes v before a long vowel. Ex. yveva=yo eva, sveva=so eva.
43. If a word ending in a or ā is followed by idaṁ, or any of the oblique cases of the pronoun idaṁ which begin with i, the consonant y is often inserted for euphony. That the y should be attached rather to the second than to the first word is clear from the fact that we have viya=yiva. Ex. na-y-idam=na idaṁ, na-y-ime=na ime, tava-y-idaṁ=tava idaṁ (Vaṅgīsa S.), cha-y-ime=cha ime, nava-y-ime=nava ime (Pay. S.), mā-y-ime=mā ime (Jāt. 203), mā-y-idaṁ=mā idaṁ (Pay. S.).
44. In one case final ā is shortened. yatha-y-idaṁ=yathā idaṁ.
45. The same process takes place with iva, but the form yiva does not occur, having been replaced by the later form viya, which is a metathesis of yiva.
46. When a vowel is followed by u or ū, the consonant v is sometimes inserted for euphony. Ex.
So also when a word ending in a vowel is followed by eva, the consonant y is often inserted for euphony.
47. A euphonic m is sometimes inserted between two vowels. Ex. idha-m-āhu=idha āhu, idha-m-eva=idha eva (Das. 44), idha-m-āgato, yena-m-idhekacce=yena idha ekacce (Pay. S.), jeyya-m-attānam=jeyya attānam, parigaṇiya-m-asesaṁ=parigaṇiya asesaṁ (Mah. 20), parañ-ca-m-avajānāti [109] =parañ ca avajānāti (Alw. N. 120), sammati-m-eva=sammati eva (Pay. S.), kathaṁ jīvaṁ jīvati-m-āhu seṭṭhaṁ Living how do they say he lives the best? viz. what do people say is the best way of living?=kathaṁ jīvaṁ jīvati āhu seṭṭhaṁ (Sutta Nipāta), ākāse-m-abhipūjaye=ākāse abhipūjaye (Pay. S.).
48. When a word ending in a vowel is followed by iva, occasionally the consonant r is inserted for euphony. Similarly we sometimes hear the vulgarism Indiar Office, but the vulgarisms of one generation sometimes become the grammar of the next. Ex. nakkhattarājā-r-iva tārakānaṁ=nakkhattarājā iva tārakānaṁ. (Pay. S.), āragge-r-iva=āragge iva, usabho-r-iva=usabho iva (Pay. S.), sāsapo-r-iva=sāsapo iva.
49. In one or two instances final ā is shortened. Ex. yatha-r-iva=yathā eva, tatha-r-iva=tathā eva.
50. Rarely other consonants besides y, v, m, and r are inserted euphonically: e.g. n in ito-n-āyati=ito āyati.
51. In a great many cases a lost consonant is revived for the sake of euphony. Ex. manasād-aññavimuttānaṁ (Sen. 26), yasmād-apeti [S. yasmād apeti]=yasmā apeti, It is usual to quote tasmātiha as an instance of a revived consonant, tasmāt iha standing for tasmād iha; but I have shown that the phrase should really be analysed tasmā ti ha (Dict. p. 480). tasmād-eva [S. tasmād eva]=tasmā eva, kenacid-eva karaṇīyena=kenaci eva karaṇīyena, kocid-eva [S. kaścid eva]=koci eva, ahud-eva [S. abhūd eva]=ahu eva, tāvad-eva [S. tāvad eva]=tāva eva, puthag-eva [S. prithag eva]=putha eva, pageva [S. prāg eva]=pa or pā eva, tuṇhīm-āsīnaṁ [S. tūṣṇīm]=tuṇhī āsīnaṁ, vuttir-esā [S. vṛttir eṣā]=vutti esā (Kasībhāradvāja S.), sabbhir-eva [S. sadbhir eva]=sabbhi eva, paṭhavidhātur-eva [S. dhātur eva]=paṭhavidhātu eva (Pay. S.), punar-eva [S. punar eva]=puna eva, bhattur-atthe [S. bhartur arthe]=bhattu atthe (Pay. S.), chaḷeva=cha eva [ṣaḍ evā, Mah.].
These revived euphonic consonants must be carefully distinguished from the euphonic consonants of rules 43-50, which are parasitic.
52. In a few curious examples a wrong consonant is revived. Ex. sammā eva should become sammageva, but we have always sammad-eva; AJ: This d for g seems to be a hypercorrection, likely influenced by the tad/etad pattern (Rule 56). similarly dhi atthu becomes [110] dhir-atthu instead of dhig-atthu; vijju eva becomes vijjur-eva when the S. would be vidyud eva (Pay. S. gives this example), while on the other hand in Sela S. we have sāvako satthud-anvayo when we ought to have satthur anvayo [S. śāstur anvayo]. Unless we look upon it as a compound for satthu-anvayo, in which case the d would be parasitic. AJ: CST4 has the expected satthur-eva.
II. Mixed Sandhi
53. When a word ending in a vowel is followed by a word beginning with a consonant, the latter, if it represents an original conjunct consonant, is sometimes doubled. In prose, this doubling is most frequent in cases in which the proclitic particle na is immediately followed by a verb qualified by it. In verse, the doubling is pretty frequent for the purpose of lengthening a short syllable. Examples from prose: idhaṭṭhātuṁ=idha ṭhātuṁ (Dh. 212), punadivasatoppabhuti=punadivasato pabhuti, nakkhamati=na khamati.
Examples from verse: yatraṭṭhitaṁ=yatra ṭhitaṁ (Dh. 23), nabhasiṭṭhitā=nabhasi ṭhitā (Mah. 108), kariṭṭhātu=karī ṭhātu, the final vowel being shortened to avoid a long vowel before a double consonant (Mah. 106), cajjhānaphalo=ca jhānaphalo, citrāhi gāthāhi munippakāsayi=citrāhi gāthāhi muni pakāsayi (Āmagandha Sutta), kāsippiyadasī=kāsi Piyadasī (Alw. I. xiii), nappasahetha=na pasahetha (Dh. 23), nappamajjeyya=na pamajjeyya (Dh. 31).
Rarely ch preceded by a vowel is doubled. Ex. esa cchecchati (Dh. 63).
54. In a few cases a lost final consonant is revived before a consonant. Thus we have yāvañ-cidaṁ=yāva ca idaṁ, where the modified anusvāra represents the lost t of yāvat; yāvakīvañ-ca.
When a word ending in anusvāra is followed by a vowel, anusvāra is often changed to m. Ex. ekam-idāhaṁ=ekaṁ idāhaṁ, kim-etaṁ=kiṁ etaṁ (Dh. 207).
In one instance m is added to a short vowel before a consonant metri causā. Ex. taṁ bahuṁ yam-pi jūrasi.
55. In one case of mixed sandhi an aspirated consonant loses [111] its aspirate to avoid the harsh concurrence of two aspirated consonants. Ida bhikkhave=idha bhikkhave. AJ: this would seem not to occur, and is contradicted by the fact bhikkhave itself has two aspirates, without it being felt as harsh.
56. There are at least two instances of ṁ being changed to n instead of m before a vowel. The first is cirannāyati AJ: In CST4 this is given as the expected ciraṁnāyati. adduced by Kaccāyana (Sen. 26), when it is doubled to reproduce the heavy sound of the Pāḷi anusvāra. See p. 112. It is impossible to say whether the words occur in prose or verse. The second, from Jayaddisa Jātaka, is “satanan esa dhammo,” AJ: In CST4 this occurs as satānesa dhammo. ‘This is the practice of good people, where satanan stands for satānaṁ, both the syllables being shortened metri causā.
Sometimes in Pāḷi final anusvāra stands for an original final consonant, and when this is the case, it is liable before a vowel to be replaced by the original consonant. Thus the Sanskrit sakṛt is sakiṁ in Pāḷi, but if eva follows we have sakid-eva, the Sanskrit sakṛd eva. Again taṁ, yaṁ, and etaṁ, are the Pāḷi representatives of the Sanskrit tad, yad, etad, and when followed by a vowel the original d is restored; AJ: This is one of the few truly morphological sandhi rules in Pāḷi, restoring the original Sanskrit pronominal stem before a vowel. It is almost without exception in prose for taṁ, yaṁ, etaṁ, imaṁ. e.g. tad-avasari, tad-apaviddhaṁ (Dh. 52), tad-eva, yad-āyasaṁ (Dh. 62), yañ-yad-eva, etad-avoca, etad-ahosi.
57. In a few instances a wrong consonant is revived. Thus the Sanskrit punar is sometimes in Pāḷi punaṁ, which should be changed before a vowel to punar; yet in Brāhmaṇadhammika Sutta we have punam-upāgamuṁ, and punad-eva is by no means an uncommon form. So we have aññad-atthu for aññam-atthu, and bahud-eva rattiṁ for bahum-eva rattiṁ. There can be little doubt that in this case the change of ṁ to d is on the false analogy of etaṁ and etad.
58. In verse, anusvāra is sometimes elided before a vowel. Ex. no ce muñceyyacandimaṁ=no ce muñceyyaṁ candimaṁ (Gog. Ev. 28), maccānajīvitaṁ=maccānaṁ jīvitaṁ (Das. 34), dasasassīnacūbhayaṁ=dasasassīnaṁ cūbhayaṁ (Jāt. 19), etaṁ buddhānasāsanaṁ=etaṁ buddhānaṁ sāsanaṁ, phuseyyutaṁ=phuseyyuṁ taṁ (Dh. 24).
59. In verse, sometimes both anusvāra and the vowel which precedes it are elided. Ex. imettaṁ=imaṁ ettaṁ (Dh. 35), malitthiyā=malaṁ itthiyā (Dh. 43), maggānaṭṭhaṅgiko=maggānaṁ atthaṅgiko (Dh. 48), pabbajayattano=pabbajayaṁ attano (Dh. 69), nipajjahaṁ=nipajjiṁ ahaṁ (Jāt. 13) [112], ovaditussahe=ovadituṁ ussahe (Jāt. 160).
60. In prose, as well as in verse, we sometimes have the combination aṁ-a=ā, as dhuvāhaṁ=dhuvaṁ ahaṁ (Jāt. 19). Here a double process has taken place, the ṁ being first elided, which gives dhuvaahaṁ, and then the two as blended into ā. Other examples are, ekam-idāhaṁ samayaṁ=ekaṁ idaṁ ahaṁ samayaṁ, kathāhaṁ=kathaṁ ahaṁ (Dh. 192), tāsāhaṁ santīke=tāsaṁ ahaṁ santīke (Pay. S.), evāyaṁ=evaṁ ayaṁ (Das. 27), labheyyāhaṁ=labheyyaṁ ahaṁ (Par. 59).
61. When a word ending in anusvāra is followed by eva, the letter y is often inserted for euphony.
In the same way, after anusvāra, iva often becomes viya, which is a metathesis for yiva.
In the case of eva we sometimes have a further change, the final ṁ and initial y blending into ññ, e.g tañ-ñeva=taṁ yeva.
In one case final anusvāra elides a following vowel: evaṁsa=evaṁ assa, alaṁthavā=alaṁ athavā (Dict. art. Peyyālam, from a very late text).
III. Consonant Sandhi
62. When a word ending in anusvāra is followed by a word beginning with a consonant of one of the second, fourth, or fifth classes, it is sometimes changed to the nasal of that class. Even Bālāvatāra recognizes this (see Cl. Gr. 13). If the second word begins with a consonant of the first class (k, kh, g, or gh), no change takes place, because in Pāḷi anusvāra is identical with the nasal of the first class. It is not, therefore, usual to write taṅ karoti. I have never met with a case of anusvāra changed to ṇ before ṭ or ḍ. Ex. evañ-ca=evaṁ ca, yañ-ce=yaṁ ce, kin-ti=kiṁ ti, kiñ-je=kiṁ je, tañ-jano=taṁ jano (Dh. 39), oran-tato=oraṁ tato (Mah. 62), param-maraṇā=paraṁ maraṇā, evam-me=evaṁ me.
63. When a word ending in anusvāra is followed by a word beginning with y, the final and initial consonants sometimes blend into ññ. Ex. ānantarikañ-ñam=ānantarikaṁ yaṁ, yañ-ñad eva=yaṁ yaṁ eva, saddhiñ-ñeva=saddhiṁ ñeva (Dh. 84). [113]
64. In one instance anusvāra is elided before a consonant, bahubhāsati=bahuṁ bhāsati (metri causā).
65. Sometimes anusvāra before h becomes ñ. Ex. evañ-hi=evaṁ hi, tañ-hi=taṁ hi, kathañ-hi=kathaṁ hi.
Rules of the Sandhi of Compounds
I. Vowel Sandhi
1. When a word whose base ends in a is compounded with a word beginning with a followed by a single consonant, the two as blend into ā. Ex. atītānāgata=atīta anāgata, pamādādhikaraṇa=pamāda adhikaraṇa, kāyānupassī=kāya anupassī, kappāvasesa=kappa avasesa, purisādhama=purisa adhama (Dh. 14), hatāvakāsa=hata avakāsa (Dh. 18).
2. But if the second a is followed by a conjunct consonant, the first a is elided. Ex. dhammanvaya=dhamma-anvaya, katañjalī=kata-añjalī, jātakaṭṭhakathā=jātaka-aṭṭhakathā, bhāvitattā=bhāvita-attā (Dh. 20), maraṇanta=maraṇa-anta, rattakkhī=ratta-akkhi.
3. Exceptions to the last rule are rare, but we have piyāppiya=piya-appiya (Dh. 38), sāṭṭhakathā=sa-aṭṭhakathā, upānta=upa-anta.
4. a-ā, ā-a, and ā-ā, all become ā. Ex. kāmāsava=kāma-āsava, dhammādāsa=dhamma-ādāsa, balavāghāto=balava-āghāto, pacchābhimukho=pacchā-abhimukho, yathā-bhirantaṁ=yathā-abhirantaṁ, pūjāraho=pūjā-araho, sadādara=sadā-ādara, paññāvudhaṁ=paññā-āvudhaṁ (Dh. 8), avijjāsavo=avijjā-āsavo.
5. But ā before a followed by a conjunct consonant is usually elided. Ex. parisantarā=parisā-antarā, mahaggha=mahā-aggha, mahaṇṇava=mahā-aṇṇava, mahatthika=mahā-atthika.
6. a or ā - i sometimes becomes e. Ex. uttaretara=uttara-itara, duddasetara=duddasa-itara (Abhidh. 998), mahesakkha=mahā-īsakkha.
7. a-i sometimes becomes ī. Ex. setībha=seta-ibha, itarītara=itara-itara.
8. a may be elided before ī and ū. Ex. sabbītiyo=sabba-ītiyo, ekūna=eka-ūna. [114]
But a or ā is elided before i if the latter is followed by a conjunct consonant. Ex. manujinda=manuja-inda, jīvitindriya=jīvita-indriya, appicchatā=appa-icchatā, icchiticchita=icchita-icchita, nāṭakitthi=nāṭaka-itthi, Laṅkinda=Laṅkā-inda, aññindriya=aññā-indriya, mahicchatā=mahā-icchatā, mahiddhika=ınahā-iddhika, mahissara=mahā-issara, yathicchita=yathā-icchita.
9. a or ā before u followed by a single consonant frequently becomes o. Ex. pallaṅkopari=pallaṅka-upari (Mah. 164), candodaya=canda-udaya, sopapada=sa-upapada, cittopacāra=citta-upacāra, senākāyopagatabhaṇḍakaṁ=senākāya-upagatabhandakaṁ (Mah. 151), paṇyopajīvinī=paṇya-upajīvinī (Jāt. 105), pacchimodadhi=pacchima-udadhi (Mah. 217), uṇhodaka=uṇha-udaka, nahānodaka=nahāna-udaka, candopama=canda-upama (Dh. 244), sunakhopama=sunakha-upama (Mah. 227), dakkhiṇodaka=dakkhiṇā-udaka, mahodadhi=mahā-udadhi, mahoraga=mahā-uraga, sahasopagā=sahasā-upagā (Mah. 52).
10. This crasis may even occur when a is followed by a double consonant, as dhammokkā=dhamma-ukkā (Jāt. 34), appossukka=appa-ussukka, noyāti=na-uyyāti (here a y is dropped to insure the long pronunciation of the o).
But far more frequently ū. Ex. pādūdara=pāda-udara, lābhūpanisā=lābha-upanisā, acirūpasampanna=acira-upasampanna, rūpūpajīvinī=rūpa-upajīvinī, uposathūpavāsa=uposatha-upavāsa, āhārūpaccheda=āhāra-upaccheda (Dh. 353), kāmūpādāna=kāma-upādāna, aññamaññūpaghaṭṭitā=aññamañña-upaghaṭṭitā (Jāt. 26).
vigatūpakkilesa=vigata-upakkilesa, parūpaghāta=para-upaghāta, ariyūpavāda=ariya-upavāda, sabbūpadhi=sabba-upadhi, pupphūpaga=puppha-upaga (and similarly phalūpaga=phala-upaga, gayhūpaga=gayha-upaga, brahmalokūpaga=brahmaloka-upaga), vassūpagata=vassa-upagata (and so rūpūpagata=rūpa-upagata), sotāpattiphalūpanissaya=sotāpattiphala-upanissaya (Das. 30, and so samatthūpadanissayo=samattha-upadanissayo).
laddhūpasampada=laddha-upasampada, paradārūpasevī=paradāra-upasevī, lakkhaṇūpanijjhāna=lakkhaṇa-upanijjhāna (so ārammaṇūpanijjhāna=ārammaṇa-upanijjhāna), attūpanāyika=atta-upanāyika, sabbākāravarūpeta=sabbākāravara-upeta, sabbūpakaraṇūpeta=sabba-upakaraṇa-upeta (Mah. 123), sabbūpakaraṇa=sabba-upakaraṇa (Mah. 123), pheṇūpama=pheṇa-upama, alagaddūpama=alagadda-upama, [115] aggikkhandhūpama=aggikkhandha-upama. Ex. of ā-u, jarūpaga=jarā-upaga, nidūpagata=niddā-upagata (and so pabbajjūpagata=pabbajjā-upagata).
11. But if u is followed by a conjunct consonant, final a or ā is elided. Ex. nīluppala=nīla-uppala, dumuppala=duma-uppala, khujjuttarā=khujja-uttarā, aṭṭhuttara=aṭṭha-uttara, pacchimuttara=pacchima-uttara, sippuggahaṇa=sippa-uggahaṇa, dīpujjalato=dīpa-ujjalato, byañjanujjalo=byañjana-ujjalo, rājuyyāna=rāja-uyyāna, kaṇṭhakuddharaṇa=kaṇṭhaka-uddharaṇa, phalikubbhava=phalika-ubbhava, vadhudyata=vadha-udyata, sukhuccāraṇa=sukha-uccāraṇa, pattuṇṇa=patta-uṇṇa, bhattuddesaka=bhatta-uddesaka, kāmummattaka=kāma-ummattaka, dvādasahatthucca=dvādasahattha-ucca, pubbutthāyī=pubba-utthāyī, mahussava=mahā-ussava.
12. Instead of the crasis ū (as in Rule 10), we pretty frequently have a elided before u, even when the latter is followed by a single consonant. Ex. pāṇupeta=pāṇa-upeta, atthaṅguposatha=atthaṅga-uposatha, paṇṇarasuposatha=paṇṇarasa-uposatha, upavatthuposatha=upavattha-uposatha, yathākammupaga=yathākamma-upaga (Gov. Ev. 10), domanassupāyāsa=domanassa-upāyāsa, addhamāsupasampanna=addhamāsa-upasampanna (Raṭṭh. S.), udaggudagga=udagga-udagga, cullupaṭṭhāka=culla-upaṭṭhāka, buddhupaṭṭhāna=buddha-upaṭṭhāna, rājupaṭṭhāna=rāja-upaṭṭhāna, hatthupagata=hattha-upagata (Jāt. 51), attavādupādāna=attavāda-upādāna, ācariyupācariyā=ācariya-upācariyā, dassanupacāra=dassana-upacāra.
Very rarely ā elides u. Ex. silāpassaya=silā-upassaya (Mah. 132).
13. a and ā are elided before e and o. Ex. ajelaka=aja-elaka, sukhesī=sukha-esī, kāmesanā=kāma-esanā, bhavogha=bhava-ogha, sissorasa=sissa-orasa, dhammoja=dhamma-oja, mahogha=mahā-ogha, avijjogha=avijja-ogha.
14. i and ī are sometimes elided by a following vowel. Ex. diṭṭhogha=diṭṭhi-ogha, hatthācariya=hatthi-ācariya, paṭaggi=paṭi-aggi, vidatthukkaṭṭha=vidatthi-ukkaṭṭha (Mah. 166), uttaruttari=uttari-uttari, cutupapāta=cuti-upapāta, hirottappa=hiri-ottappa, paṭhavissara=paṭhavī-issara, paṭhavoja=paṭhavī-oja, bhikkhunovāda=bhikkhunī-ovāda.
15. Occasionally i and ī elide u. Ex. satipaṭṭhāna=sati-upaṭṭhāna, dubbuṭṭhipaddava=dubbuṭṭhi-upaddava, diṭṭhijju=diṭṭhi-ujju, bhikkhunipassaya=bhikkhunī-upassaya (the ī being at the same time shortened).
16. i or ī-u may become ū. Ex. diṭṭhūpādāna=diṭṭhi-upādāna, cutūpapāta=cuti-upapāta, uparūpari=upari-upari, rattūparato=ratti-uparata, bhikkhunūpassaya=bhikkunī-upassaya.
17. u and ū may be elided by a following short or long vowel. Ex. mātupaṭṭhāna=mātu-upaṭṭhāna, mudindriya=mudu-indriya, cakkhāyatana=cakkhu-āyatana, abhibhāyatana=abhibhū-āyatana.
18. Rarely u elides a vowel. Ex. utukkhāna=utu-akkhāna, cakkhundriya=cakkhu-indriya.
19. u sometimes blends with u into ū. Ex. bahūpakāra=bahu-upakāra, bahūdaka=bahu-udaka, sūju=su-uju, catū-sabha=catu-usabha.
20. Rarely i preceded by t or tt becomes y before a vowel. Ex. atyappa=ati-appa, mutyapekha=mutti-apekha (Vaṅgīsa S.).
21. Rarely i or ī preceded by d becomes y before a vowel. Ex.
22. u occasionally becomes v before a vowel. Ex. cakkhvāpātham=cakkhu-āpātham, bhikkhvāsana=bhikkhu-āsana, bavhābādha=bahu-ābādha (this is a metathesis for bahvābādha, which does not occur).
23. ati and paṭi before a vowel generally become acc and pacc, standing for an older aty and paty. Ex. accuṇha=ati-uṇha, accokkaṭṭha=ati-okkaṭṭha, accodāta=ati-odāta, accagā=ati-agā, paccāroceti=paṭi-āroceti, paccaṅga=paṭi-aṅga, paccupaṭṭhita=paṭi-upaṭṭhita, paccaññāsi=paṭi-aññāsi.
24. In one case dī followed by a vowel becomes jj, which represents an older dy: najjantara=nadī-antara.
25. abhi and adhi before a vowel generally become abbh and ajjh, which represent older forms abhy and adhy. Ex. abbhaññāsi=abhi-aññāsi, abbhattha=abhi-attha, abbhokāsa=abhi-okāsa, ajjhabhāsi=adhi-abhāsi, ajjhāvasatha=adhi-āvasatha, ajjhokāsa=adhi-okāsa, bojjhaṅga=bodhi-aṅga.
26. If a word ending in i or ī is compounded with a word beginning with a, ā, u, e, or o, a euphonic y is sometimes inserted, ī is always shortened. Ex. nadiyantike=nadī-antike, piṇḍiyālopa=piṇḍi-ālopa, ukkhaliyādīni=ukkhali-ādīni (Jāt. 30), nandiyāvaṭṭa=nandi-āvaṭṭa, AJ: these examples were written: piṇḍiyālopa=piṇḍ-ālopa, ukkhaliyādīni=ukkhal-ādīni, nandiyāvaṭṭa=nandi-āvarta, which must be mistakes. dviyāsīti=dvi-asīti (the lengthened ā has nothing to do with sandhi), pariyuṭṭhita=pari-uṭṭhita, pariyesati=pari-esati, pariyodāna=pari-odāna. [117] In two instances, in which a verb is compounded with pari, metathesis takes place: payirupāsati and payirudāha; pointing to transition forms pariyupāsati and pariyudāha.
27. In one instance, ativiya=ati-iva, we have y inserted between two is, for ativiya points to a transition form atiyiva.
28. In two or three cases v instead of y is inserted between i and a. Ex. tivaṅgika=ti-aṅgika, tivaṅgula=ti-aṅgula (Pay. S.).
Occasionally euphonic v is inserted between u and ā. Ex. puthuvāsana=puthu-āsana.
29. Occasionally euphonic v is inserted between a and u. Ex. savupādisesa=sa-upādisesa, pāguññavujutā=pāguñña-ujutā.
30. Sometimes a euphonic m is inserted between two vowels. Ex. aññamañña=añña-añña, ekameka=eka-eka, ekamantaṁ=eka-antaṁ (comp. ekamante, ekamantikaṁ), okamokato=oka-okato, adukkhamasukhā vedanā=adukkha-asukhā vedanā, samaṇamacala=samaṇa-acala, hīnamukkaṭṭha=hīna-ukkaṭṭha, bhāramoropana=bhāra-oropana, phalamuttama=phala-uttama.
There are two examples of a euphonic d being inserted: sadattha=sa-attha (Sansk. svārtha), and attadattha=atta-attha (Sansk. ātmārtha).
31. There are two examples of an inserted h: suhuju=su-uju, and suhuṭṭhita=su-uṭṭhita.
32. Very frequently a lost consonant is revived for euphony. Ex. yāvadattham=yāva-attham [Sansk. yāvad-artham], pātarāsa=pāto-āsa [Sansk. prātarāśa], caturaṅgula=catu-aṅgula, punarāgamana=puna-āgamana [Sansk. punarāgamana], pāturahosi=pātu-ahosi [the present tense is pātubhavati, the Sansk. is prādurabhaviṣīt], chaḷabhiññā=cha-abhiññā [Sansk. ṣaḍabhijñā or chaḷabhijñā].
33. In the case of sammā, a wrong consonant is revived, d instead of g. Ex. sammadakkhāta=sammā-akkhāta (when the Sanskrit would be samyagākhyāta), sammadañña=samma-añña (Dh. 11, 18). [118]
II. Mixed Sandhi
34. When a base ending in ā, ī, or u, is compounded with a word beginning with a single consonant which represents an original single consonant, the long vowel is very frequently shortened. Ex. paṇṇasālabhitti=paṇṇasālā-bhitti (Jāt. 8), mattikabhājana=mattikā-bhājana (Das. 34), yānikata=yānī-kata, Vedehiputta=Vedehī-putta, sabbaññubuddha=sabbaññū-buddha.
35. When a base ending in a, i, or u, e or o, is compounded with a word beginning with a single consonant which represents an original conjunct consonant, the consonant is doubled. Ex. kayakkaya=kaya-kaya [krayakraya], dukkhakkhaya=dukkha-khaya [Sansk. duḥkakṣaya], parappavāda=para-pavāda [Sansk. parapravāda], dutiyajjhāna=dutiya-jhāna [dvitīyadhyāna], ratanattaya=ratana-taya [ratnatraya], pariccajati=pari-cajati [paritjajati], paṭiggahaṇa=paṭi-gahaṇa [pratigrahaṇa], aggikkhandha=aggi-khandha [agni-skhandha], Ānandatthera=Ānanda-thera [Ānanda-sthavira], atthittaco=atthi-taco [Raṭṭh. S.; Sansk. asthi-tvac], anuddhaṁseti=anu-dhamseti [anudhvaṁs], senāpatiṭṭhāna=senāpati-ṭhāna [senāpatisthāna], tamokkhandha=tamo-khandha (Ev. 6), tejokkhandha=tejo-khandha (Ev. 6), aṭṭhito=a-ṭhito, abhissanna=abhi-sanna, paripphosaka=pari-phosaka, utuppasevanā=utu-pasevanā, nijasissappaveṇi=nijasissa-paveṇi, assuta=a-suta, mahagghasa=mahā-ghasa. AJ: with shortening of the long vowel owing to being before a double consonant. It should really be in the next section.
36. When the base ends in ā, ī, or ū, those vowels are generally shortened. Ex. nānappakāra=nānā-pakāra [Sansk. nānāprakāra], silatthambha=silā-thambha, taṇhakkhaya=taṇhā-khaya (Ev. 6, Dh. 34), paṭisambhidappatta=paṭisambhidā-patta (Alw. I., cvii), paṭhavippadesa=paṭhavī-padesa, paṭhavippabhāsa=paṭhavī-pabhāsa (Ten J. 47), aggamahesiṭṭhāna=aggamahesī-ṭhāna (Das. 2), viññuppasattha=viññū-pasattha (Par. 7, 18), vedanakkhandha=vedanā-khandha (Ev. 68), vedappaccayā=vedā-paccayā (Ev. 66), sammuñjanippahāra=sammuñjanī-pahāra (Dh. 372).
37. But sometimes the long vowels are retained. Ex. saññākkhandha=saññā-khandha (Ev. 68), vedanākkhandha=vedanā-khandha (Ev. 42, Atth. 163), tadāppabhuti=tadā-pabhuti, divāṭṭhāna=divā-ṭhāna, sikkhāttayaṁ=sikkhā-tayaṁ (Dh. 358), yathākkamaṁ=yathā-kamaṁ, yathātthāne=yathā-thāne, āṇākkhetta=āṇā-khetta.
38. There are a great number of exceptions to the rule of doubling the consonant, some due to phonetic causes, others to metrical, etymological and euphonic causes. Ex. supaṭipanna, [119] ñāyapaṭipanna, AJ: in CST4 these given as –ppaṭipanna, but in the Sri Lankan texts often as here. lekhāṭhāna (Mah. 113), avijjāpaccayā (Ev. 38, 66), nibbānagāminīpaṭipadā (Dh. 269), anāgatapaccuppanna, pādaphoṭa, adhoṭhita, kankhāṭhāna, upaṭṭhitasati, athena, anuthera, vattapaṭivatta, uccārapassāva, chinnapapāta, sīghasotāya, daḷhapākāra, pariphandati, duphassa, abhiṭhāna, utupamāṇa, porāṇakapaveṇiyā, anupakhajja, abhisanna, nigaṇṭha, atipāto, asuta, bodhimaṇḍathitaṭṭhāna (Mah. 7).
39. If a base ending in a vowel is compounded with a word beginning with ch, the latter becomes cch. Ex. cakkacchinna=cakka-chinna, meghacchanna=megha-channa, hatthicchāpa=hatthi-chāpa, vyañjanacchāyāya=vyañjana-chāyāya, setacchatta=seta-chatta.
There are a few exceptions to this rule, as gamanachanda (Dh. 84), uddhachiddaka (Dh. 169).
40. Occasionally anusvāra is inserted between a base ending in a vowel and the initial consonant of the word with which it is compounded. Ex. abhiṇhaṁsannipātā (pl.) ‘holding frequent assemblies, for abhiṇha-sannipātā (abhiṇha can only be an adj. here), purimaṁjāti for purima-jāti, āgamanaṁmaggaṁ for āgamana-maggaṁ (here it is metri causā), sataṁsamaṁ for sata-samaṁ, andhantama for andha-tama.
41. Occasionally when a word ending in a vowel is compounded with a word beginning with a consonant, a consonant which originally belonged to the base of the first word is revived, and if necessary assimilated to the initial consonant of the second word. Thus sammā-paññā becomes sammappaññā, which probably represents the Sanskrit samyakprajñā; anto is the Sanskrit antar, but in composition we sometimes have the original r revived, e.g. antaraghara=Sanskrit antargṛha. Again, the Sanskrit base catur is catu in Pāḷi, e.g. catuvagga (catumāsaṁ), but compounds like catugguṇa, catubbagga, catummukha, point to Sanskrit forms caturguṇa, caturvarga, caturmukha retaining the final r. So also we have cha=Sansk. ṣaḍ, but chammāsa points to an original ṣaḍmāsa. And puna compounded with bhava and with puna gives punabbhava and punappuna, in Sansk. punarbhava and punaḥpunar. [120]
42. When a word ending in anusvāra is compounded with a word beginning with a vowel, anusvāra is changed to m. Ex. evam-ucca=evaṁ-ucca, evam-āhāra, evam-ādi, alam-ariya, sāyam-āsa, aham-ahamikā.
43. In a few instances anusvāra stands for an original final consonant, and is replaced before a vowel by the original consonant. Ex. sakadāgāmī=sakiṁ āgāmī [Sansk. sakṛdāgāmin], īsadatthe=īsaṁ atthe (Abhidh. 1169), etad-atthāya=etaṁ atthāya [S. etadarthāya], tad-anurūpa=taṁ-anurūpa [tadanurūpa].
III. Consonant Sandhi
44. When a word ending in anusvāra is compounded with a word beginning with a consonant, it is generally changed to the nasal of the class to which the consonant belongs. In the case of a consonant of the first class, practically no change takes place, because anusvāra in Pāḷi is identical with the nasal of the first class, that is, the pronunciation is the same whether we write taṅkhane or taṁkhane. Ex. sabbañjaha=sabbaṁ-jaha, mahindhara=mahiṁ-dhara, jarantapa=jaraṁ-tapa, sayampabha=sayaṁ-pabha, rattindiva=rattiṁ-diva, evannāmaka=evaṁ-nāmaka, orambhāgiya=oraṁ-bhāgiya, muhumbhāsā=muhuṁ-bhāsā.
45. But there are instances in which anusvāra remains unchanged before a consonant. And it is important to bear in mind that when this is the case, there is a real difference of pronunciation; for instance, kathaṁ-bhūta is differently pronounced from what kathambhūta would be, the ṁ in the first case being pronounced like the ng in our “sing.” Ex. saddhiṁcara, kathaṁbhūta, evaṁmahānubhāva, evaṁnāma, evaṁdhamma.
In one instance m remains unchanged before 1: uccamlatā.
46. In muhumbhāsā, and perhaps two or three others, the nasal is retained, where, on the analogy of compounds like punabbhava, we might expect the original consonant to be revived and assimilated (e.g. muhubbhāsā).
There are a certain number of anomalous sandhis not coming under any of the above rules, which it will be well to mention here. There are three cases of de-aspiration coming [121] more or less under the head of sandhi. The first is ida bhikkhave, given in all the native grammars, where idha is changed to ida to avoid the harsh concurrence of two aspirate letters in consecutive syllables. AJ: see my note to § 55. AJ: Childers’ claim that this appears in all the native grammars is likely a repetition of a traditional rule that may be purely prescriptive or based on a lost reading. CST4 and all standard editions have idha bhikkhave. At Kh. vii. 6 we have an exactly similar change, only that the syllable affected is the second instead of the first, gorakkhetta for gorakkhettha. The third instance is a very curious example of compound-sandhi given in Pay. S., saṁvidāvahāro=saṁvidhāya-avahāro. In the phrase ekaṁ idāhaṁ samayaṁ, the native grammars generally assert idāhaṁ to be a sandhi for idha-ahaṁ, but there can be no doubt that they are mistaken, and that ida is the Sanskrit idam. DPD gives it as Sansk. iha+ahaṁ. There are two cases, one of compound-sandhi, one of word-sandhi, in which ya is elided before a vowel. The first being saṁvidāvahāro already mentioned, and the second vācā-udacetasā vā for vācāya uda vā. In tad-aminā for tad-iminā, and in atippago kho for atippage kho, we probably have examples of vowel assimilation. Bālāvatāra quotes the phrase kacci no tvaṁ for kacci nu tvaṁ, where u may be changed to o metri causā, but no clue is given as to whether the passage from which it is taken is prose or verse. AJ: the phrase can be found in prose at Pāyāsi Sutta, DN 23.