Adding new parallels to SuttaCentral


One main feature of SuttaCentral is to show parallels between Early Buddhist Texts. However, there are many more parallels than are yet shown on the site, simply because we have not had the time to add them. Adding parallels is rather time-consuming and with the launch of the new site they have not been updated lately. Some of our users like @lemon, @crizna and @seniya have posted parallels and those will be among the ones that I will look at in the next months.

A list of parallels to be added is here:

But first I have started adding the parallels as listed in the Udanavarga de Subaši, which is now online at and will come to the main site soon. @Aminah has done a great job to OCR the pdf (which you can download from the meta area), which was a very hard job.


Thanks so much Ayya, please keep us informed of your progress. The parallels are the backbone of SC.


Ayya, excuse my ignorance how do you locate parallels in SC. Neural networks (with human oversight) may provide some insights in this kind of work do you think?


The method I use for adding new parallels is mainly by looking through documents and research done by scholars who have identified parallels. Like for instance the book about the Udanavarga de Subasi. Then in checking those parallels, I sometimes find others which were unknown. SuttaCentral is a great help in this because I can key in a sentence in pali in the search and it will show me places where this sentence is found in the texts. Often I find new parallels that way.

Unfortunately, I can only compare Pali, Sanskrit, Prakrit and Gandhari but not the Chinese or Tibetan texts. This is where @lemon has been a great help to me to check the parallels of the various Dharmapadas.


This sounds hard. When you say parallels, are you producing a document discussing these? The reference to neural networks is not spurious. Someone in this forum suggested uniform translations of words is the gold standard and yet the meaning of words is context bound it seems from scholars in the field including Ven. Analyo. Translators essentially create a model and the use of artificial intelligence makes the model assumptions explicit. It seems if you are comparing translations of multiple documents the rules of engagement are even more complex.


Bhante, it is great that you’ve came back.
I’ll continue to proofread the parallels of Udanavagga and T213.

And I’d like to add 2 things. Firstly, according to Bhikkhu Ānandajoti (2003), almost all the Udanas(mostly gathas) from Udana can be found in Udanavarga:
A_Comparison_of_the_Pali_Udana_and_the_B.pdf (379.5 KB)
And according to Bhikkhu Anālayo, the Udanas(mostly gathas) of Udana are internally linked:
The_Development_of_the_Pali_Udana_Collec.pdf (869.4 KB)
So it seems that the Udana originally consisted of only short Udana senteces, mostly gathas, and it was expanded later in 2 ways. The Theravadins added back stories, and the Sarvastivadins added more gathas, changed the non-gatha Udanas into gathas, and called it “Udanavarga”.

Secondly, according to a master’s thesis by Ji Wenting (季文婷, 2003), Shanghai Normal University, the translation quality of Chinese Dhammapada is quite poor.
Just like the Tibetan guy’s thesis mentioned last time, its English abstract is also unreadable, so I translated the abstract by myself:

The First Exploration on the Pali-Chinese Translation Comparison of Dhammapada


The Chinese translation of Dhammapada(T210) hybridized both Pali and Sanskrit sources, and the origin of some verses are still unknown. The Chinese Dhammapada translation was in the early stage of Chinese Buddhist scriptures translation. Its grammar and vocabulary are not yet in a system, and its verses formats are not yet unified, which was in a relatively natural state.

During the translation, the Chinese Dhammapada added some contents, such as verses number; meanwhile it omitted much information to meet the rules and forms of classical Chinese poetry.

The biggest feature of the Chinese translation is the loss of the syntax. During the Chinese translation process, anything might be omitted, including words, sentences, verses, especially the syntax.

The second feature is that the Chinese translation rejected all repetitions in the Pali Dhammapada, including words, sentences, and verses.

Furthermore, except the vagga names translation, it was basically translated in a literal way, although it also includes many necessary explanations. But all in all, the Chinese translation of Dhammapada still had a high error rate due to the obscure Pali language.


The parallels are shown on SuttaCentral in a dropdown like for instance here (it will open up automatically):
And yes, Bhante Analayo’s work is also included in these as well as many others but I intend to check through those.

Thank you @lemon. I will have to come back to you later about all this. I’m still going through the Udanavarga de Subasi first and then get to all of your work. I can only work a limited time per day because we are still building the monastery.


what are the implications of the high error rate?


As I am going through the UVS parallels as listed in the research by Nakatani, I get quite a lot of chance-finds within the Pali canon. This happens when I key in a sentence in the search in order to find a specific text and then find many others with the same verse.

Therefore, I want to write some code in python to go through the entire Pali canon (and maybe later Sanskrit) to harvest all first lines of verses, compare them and output all the verses with numbers that have the same first lines. It is not a fullproof algorithm because first lines are not always exactly the same and it says nothing about the next lines, but it will give a list with indications where to look for parallels and I think it will be able to find around 80% of verse parallels within the Pali canon.

This will only work for verses because there the rules for parallels are much better defined.


I can give you a simple example from her thesis:
T210 13-16:
报 有 印章
(His) retribution is sealed/stamped. (i.e. His retribution is confirmed.)

Pali: muddham assa vipātayaṃ.
His head would split.

Muddha(head) is mistaken as Muddā(seal/stamp).
There are lots of such mistakes in the Chinese translation.


That’s great. Thanks so much! I will get to your analysis of UV2 soon.

I think we already have those in there but I will check it out. In any case sources also need to be noted on the Methodology page on SuttaCentral.

Thank you so much for all this background information! It is very interesting.


Studying parallels at Tilorien.


I’ve had some private discussions with @lemon about various parallels and I want to share these here because there are some interesting comments that might be useful to other people too.

I compared all the parallels and found the full view:
Sahasra (11.0 KB)
Please check this zip for a detailed table.

Today I just happened to quote T210 13.16 on Sutta Central.
But your T210 13.15, which was actually Dhp 70, is an amazing original meme that duplicated several times.

The keyword in Dhp 70 (T210 13.15) was saṅkhātadhamma, and there was only one gatha.
But after it moved to patna dharmapada, the keywords became 4, and another format appeared, so the gathas became 8:

original: sākkhātadhaṁma (21.85,89)
variant: buddha (82,86), dhamma (83,87), saṁgha (84,88)

variant: sahasreṇa (21.82-85)
original: kuśāggreṇa (86-89)

After the 8 gathas of PDHP moved to Mahāvastu & udānavarga, the key words duplicated again separately, but the 2 formats stood the same.

For Mahāvastu
buddha (109#5,10)
dharma (6,11)
saṃgha (7,12)
svākhyātadharma (9,15)
dhyāna (13)
saṃpannaśīla (8,14)

sahasrāṇāṃ (109#5-9)
kuśāgreṇa (10-15)

For udānavarga
buddha (24.17,26)
dharma (18,27)
saṁgha (19,28)
svākhyātadharma (25,34)
tacchīle (20,29) – ( śīla ) accustomed to that
maitrasya cittasya (21,30)
satvānukampāyā (22,31) – compassion to all beings
prāṇānukampāyā (23,32) – compassion to breathing beings
bhūtānukampāyā (24,33) – compassion to born beings

And its order of two formats turned around:
kuśāgreṇa (24.17-25)
sahasreṇa (26-34)

For the Chinese parallels, as the translation of udānavarga, the paralleled gathas in T213 are not complete, some of them also had errors that were marked in red by me.
However, T213 24.30 might be a derived gatha, since it has the meaning of “hold the grudge” that could not be found in UV.

And T212 listed only the first gatha, but it did mentioned other gathas in the explanation, but also incomplete.

That’s all I’ve found out by now, please have a look.


Original meme:
Dhp 207 (Sukhavagga)
Bālasaṅgatacārī hi,
dīghamaddhāna socati;
Dukkho bālehi saṃvāso,
amitteneva sabbadā;
Dhīro ca sukhasaṃvāso,
ñātīnaṃva samāgamo.

It appeared twice in the UV:
Sukhavarga 30.26
bālasaṁsargacārī hi dīrghādhvānaṁ praśocati ।
duḥkho bālair hi saṁvāso hy amitrair iva sarvaśaḥ ।
dhīrais tu sukhasaṁvāso jñātīnām iva saṁgamaḥ ।।
rnam pa kun du dgra ’dra ba’i ||
byis daṅ ’grogs pa sdug bsṅal źiṅ ||
byis pa dag daṅ ’grogs byed pa ||
yun riṅ rjes su ’gyod par ’gyur ||
'Tis as great suffering to be in the company of fools as in that of enemies; he who associates with fools will repent him of it for a long time.
------Please note that the last two lines were missing.
T213 30.27
------The last two lines were not missing. And it was cloned in T212:
The numbering of T212 is confusing, because some gathas were repeated in their explanations.

Mitravarga 25.24
bālaṁ na paśyec chṛṇuyān na ca no tena saṁvaset ।
duḥkho bālair hi saṁvāso hy amitreṇaiva sarvaśaḥ ।
dhīrais tu sukhasaṁvāso jñātīnām iva saṁgramaḥ ।।
byis pa bsten pa sdug bsṅal źiṅ ||
rnam pa kun du dgra ’dra bas ||
byis pa mi thos mi mthoṅ źiṅ ||
ma bsten gyur pa ñid ruṅ ṅo ||
gñen daṅ phrad pa ji lta bar ||
brtan pa bsten na bde bar ’gyur ||
He who associates with a fool is in misery, as if he were with an enemy; one ought not to associate with fools, neither ought one to listen to or see them; associating with the steadfast is happiness, like meeting again one’s kinsfolk.
------For the Tibetan translastion, if you compare the gatha in Vagga 30 with its parallel in Vagga 25, you’ll found that the first 2 lines are basically the same but upside-down, and for 3-6 lines, only the first 2 characters are the same, the rest parts are all paraphrased with similar meanings.
T213 25.20
------It is so interesting that the last 2 lines were missing in T213 25.20, just like what happened in the above Tibetan Vagga 30.
But its parallel in T212 was complete.
This is unbelievable, because T212 was translated in AD399 by Samghabhuti from Kophen (an ancient middle Asia country near the a river with the same name ) and Zhu Fo Nian ( 竺佛念), and T213 was translated in about AD980~1000 by Tian Xi Zai (天息灾, probably translated from the name Deva Santika), a monk from Kashmir.
It is for sure that during the translation of T213, Deva Santika checked the previous translation of T212. So how could he miss the last 2 lines of 25.20? I have no idea.


Good morning, Bhante.
Thank you for quoting.
I think the variants should also be treated as parallels.
Maybe a 90-100% parallel could be called an “original parallel”, and a 70-85% parallel could be called a “variant parallel”.
A variant parallel is similar to a partial parallel, or could be treated as one kind of partial parallels.


Here’s another example of meme copy, I have to use colour to present its process, a brainstorm map would be much better:

For the gatha in Ud 1.3 (also in the Pali Vinaya), what happened here could be called reduction division.
When the gatha moved to SA, it became a little longer.
After it reached Udanavarga, this one gatha divided into two (33.81 & 33.83), and each gatha contained one line of the original one.
And in the middle of these two gathas, a thrid gatha (33.82) appeared, which was a hybrid of 81 & 83.

And that’s more parallels of this set of 3 gathas (Ud 1.1-3) in different Chinese Vinayas.
Maybe I should draw a brainstorm map to show all these variants in a much clearer way.


It is indeed a little vague. We do no longer call things a “partial” parallel. The list of parallel-types and how they are defined can be found on our Methodology page:

We only use these 4 classes and do not split things up further.

For verses, if they have all lines in common or practically the same they are “full”, otherwise they are “resembling”. I do not list the ones that only have 1 line in common unless in specific circumstances. That would create a whole lot of them without added value. An exception would be for instance if only one line in the UVS has been recovered, and it is practically certain that it refers to a gatha in the UV simply because of it’s position within the whole structure, I use that one as a full parallel. I follow Nakatani in this.

Note however that most verse-parallels have not been checked by me but simply added from our old database and other sources.


Things like this are very interesting and by adding them here to D&D they will be linked to that sutta as a commentary.

Sets of variant verses are indeed very common but currently we have no way to display these. At the moment I just have a full parallel to one of them but not to the variants. The same variants might appear in different places so they have their own parallels too. @Sujato - any thoughts?


Interesting. It could be classified simply under “resembling”, but the question would be if we gain anything by introducing a concept of “variant” parallels. Formally it is very reminiscent of the musical form called “theme and variations”. There are similar cases in the Pali, too, I wonder how we deal with them. For example, the famous verses on sabbe sankhara anicca ….

One of the deciding factors is whether our source data uses this kind of distinction, or else, are we going to put the time and effort into creating the data ourselves? There’s no real point to just listing a few cases, but if it can be done reasonably consistently it may be worth while.


Actually for the verses this is not done either if you have a set that are all displayed under each other with for instance one line difference. Then there is a full parallel to one of them and the rest is not marked as a resembling parallel (also because they are just shown under each other so anybody who goes there sees the various variants immediately).

To be honest, I think it will be a lot of work to go through the whole lot again. I’m not sure if that’s worth it considering that there are many other sources of parallels that need to be checked also. And like I said, these variants are all displayed under each other so anybody who goes there sees it immediately.