Lost in
Transitive

A grammatical category is a knife with a fixed angle. It cuts one dimension of a situation and ignores the rest. Every language carries some of these knives as compulsory and leaves the others optional. What follows is the working vocabulary for the rest of the page; the icons reappear later as shorthand.

The situation: something is wrong, now, with the speaker. Below are eight ways of saying it. The vertical lines are word boundaries. These are not stylistic variants of one underlying sentence. They are different decisions about how many participants exist and what relation each has to the state.

English can suppress the person, but it charges for the service: a placeholder in the subject slot, and the conversion of a state into a noun. Tamil pays neither. That is the technical content of the observation that the English impersonal sounds clinical.

Experiencer deletion is not a Tamil quirk

Dropping the person entirely from a report of one's own condition is available across South Asia, in both language families. Malayalam is the limiting case: no person, no bearer, no tense, one word.

Marathi and Bengali pattern the same way. Those forms need a native check before they are printed as evidence.

The Sabbāsava Sutta is about how the āsavā — the taints, the things that leak — come to an end. It names seven ways they are abandoned: by seeing, by restraint, by use, by endurance, by avoidance, by dispelling, by cultivation. The entire text is built on pahātabbā, "to be abandoned", a gerundive, which is a form with no agent slot in it at all.

This paper uses only the first of the seven, abandoning by seeing. That section is where the discourse describes how a self gets manufactured, and it is the section English translation damages most.

Pali from the Chaṭṭha Saṅgāyana edition. Tamil from நவாலியூர் சோ. நடராசன், on SuttaCentral. English renderings from Bodhi, Sujato, and Words of the Buddha, paraphrased rather than quoted.

The passage is a causal chain, and it runs in both directions. A person who has not been taught does not know what is worth attending to. Attending from the wrong place, sixteen questions arise. Out of those questions, one of six views. Out of the views, what the discourse calls a thicket of views, a wilderness of views, and no release from birth, ageing and death.

Run the other way: a person attends from the source to the four truths, and three fetters fall away. The first of the three is sakkāyadiṭṭhi, identity view — the belief that there is a self here.

The passage is an account of how a self is assembled out of a certain kind of questioning, and of what dismantles it.

Take the word here. It picks out wherever the speaker happens to be standing. It has no career of its own: it was not somewhere else last year and it will not move tomorrow. "Was here here in the past?" is not a difficult question but a malformed one, because here is a device for indexing an utterance, not a name for a place.

The sixteen questions treat I as though it were a name. Each asks after a property of something across time: whether it existed, what it was, how it was, what it changed into. Those are reasonable questions about Devadatta, who is a person with a history. "Was Devadatta in the past?" and "Was I in the past?" have the same grammatical shape, so the second looks equally answerable. But I is not a name. It picks out whoever is speaking, the way here picks out wherever they are standing. Asking what it was doing before the sentence started is asking after the history of a pointer.

This is why the discourse calls the result a thicket rather than a mistake. A mistake is corrected by finding the right answer. A malformed question generates answers indefinitely, and the six views are where the generating stops: six positions a person settles into after trying to answer something that has a grammatical subject and no semantic one.

A language that obliges its speakers to pronounce that pointer in every clause is a poor vehicle for this diagnosis.

Pali, Sanskrit and Tamil are pro-drop languages: the verb ending already identifies the person, so the pronoun can be left unsaid, and normally is. Ahosiṁ means "I was" on its own. English is not pro-drop — it cannot leave the subject slot empty, and inserts a placeholder where there is no participant at all: it is raining, there is a body.

So when ahaṁ, "I", is written in Pali alongside a verb that already carries it, it is doing no grammatical work. It is surplus, and surplus in a pro-drop language is emphasis. It appears in five of the fourteen questions, always at a question about bare existence or about becoming, and never at a "what" or a "how".

Below, the past and present sets. The five future questions repeat the past pattern with bhavissāmi for ahosiṁ, and are omitted to keep the table readable.

Across all fourteen questions: Pali writes the pronoun five times. Natarajan's Tamil writes it thirteen times. English writes it fourteen times, because it cannot do otherwise.

The Tamil count is the interesting one. Tamil could have tracked the Pali exactly — it is pro-drop, and dropping the pronoun costs nothing. Natarajan supplies it almost everywhere instead, and the one place he omits it is a place where the Pali has it. His two parallel "having been what" lines disagree with each other, the past dropping நான் and the future keeping it, which indicates the choice was made for flow rather than for the pattern.

This is the forced-and-chosen distinction from section 4, applied to a translator. For English the pronoun is forced. For Tamil it was chosen, and the capacity went unused.

What happens to a naive English rendering, and what a translator can do about it.

Case 3 is the practical one. A teacher reading these questions aloud can put the weight where the Pali puts it — on "Am I?" and "Will I be?", and off "what" and "how" — and the passage recovers its shape without a word being changed.

A case ending is a suffix on a noun saying what part it plays in the event: the doer, the thing done to, the means, the place. English mostly uses word order and prepositions instead. Pali, Sanskrit and Tamil use endings, which lets them distinguish roles English blurs.

Here are the six, with the grammar shown.

Six views, three grammatical shapes.

In 1 and 2, attā is in the nominative as the subject of an existential verb, with me in the genitive. A self exists, or fails to, and it is owned. In 6, attā is again nominative, but now as the subject of finite verbs: it speaks, it feels, it experiences the results of good and bad deeds, and it is permanent. Not a thing that exists — a thing that acts. In 3, 4 and 5, attā is not the subject at all. The subject is the unpronounced "I" inside sañjānāmi, and attā has been demoted into the machinery of perceiving.

Those three are one clause with two slots. The first slot takes the instrumental: what the perceiving proceeds from. The second takes the accusative: what gets perceived. Each slot accepts either attā or anattā, and three of the four possible fillings are listed.

View five is the one worth sitting with. Anattanā va attānaṁ sañjānāmi — proceeding from not-self, perceiving a self. Read quickly it sounds like accomplished practice: someone who has seen through the self and is now observing from that vantage. The discourse lists it as a wrong view.

The reason is in the case ending. The instrumental turns whatever fills that slot into something the perceiving comes from — a standing position. Not-self used as a standing position is no longer functioning as not-self. It has become the place you are looking out from, which is the job a self does. The view is not wrong because not-self is false. It is wrong because the grammar has made it into somewhere to stand.

That is a near-enemy, and it is the kind a practitioner can occupy for years. It is also invisible in English, where "with not-self", "by not-self" and "in what is not-self" all appear in published translations and point at three different things.

This reading of view five follows from the case marking. It is not something the discourse spells out, and the commentary may take a different line — see below.

The missing fourth cell

Anattanā va anattānaṁ sañjānāmi, proceeding from not-self and perceiving not-self, is not in the list. Three explanations, in decreasing order of confidence.

The plainest is that the absence may mean nothing. The six are an enumeration of views that arise, not a matrix someone filled in. That three of them happen to occupy three-quarters of a clean square may be an artefact of our drawing the square. This possibility should be held before the other two, and it is the reason the grid above is presented as an observation rather than as the text's own scheme.

If the list is systematic, the structural explanation is sufficient. All six are views about a self — the passage is describing how sakkāyadiṭṭhi, identity view, gets built. A sentence with no attā anywhere in it is not a view about a self, so it has no business in a list of them. This accounts for the absence without any claim about whether the sentence is true.

A third reading is more speculative and more interesting. The sentence may be close to accurate, and holding it as a position is what spoils it. Sañjānāmi still carries a first-person ending, so somebody is claiming it, and a claim about the absence of a self made by a claimant has smuggled one back in. On this reading the cell is empty for the same reason emptiness is said not to be a view. This is attractive, and the discourse does not say it.

The commentary, Papañcasūdanī, reads views 3–5 in terms of which of the aggregates is taken as self and which as not-self, and would probably settle the question. It has not been checked for this paper.

One sentence, five renderings. The Pali is a genitive absolute with a passive verb: the fetters are the grammatical subject, the practitioner sits in the genitive at the edge of the clause, and nobody abandons anything.

Tassa evaṁ yoniso manasikaroto tīṇi saṁyojanāni pahīyanti.
3SG.GEN — thus — from-the-source — mind.LOC-doing.PTCP.GEN — three fetters.NOM.PL — abandon.PASS.3PL

Tamil does not win everywhere. Yoniso is an ablative, from yoni, womb or origin: attention from the source. Every English rendering converts it to an adjective of quality — wise, appropriate, radical, rational. So does the Tamil: சாதுரியமான, sāduriyamāṉa, "skilful", from Sanskrit cāturya, dexterity. And Natarajan renders manasi karoti, "does in the mind", as சிந்திக்கிறான், "he thinks", a finite verb with an agent in the ending. Where Tamil does hold the line is in the table below.

The discourse is built on pahātabbā, a gerundive. A gerundive has no agent slot in the grammar at all — not an omitted agent, an absent one. The seven means are case-marked action nouns, ablative in Pali and instrumental in Tamil, and none of them names a doer either.

Written English handles the gerundive well, because the modal passive is also agentless. The slide happens in three steps, as the same content moves from canon to retreat hall.

The third line is where the damage is done, and it is done by a single grammatical operation: converting a gerundive into an imperative. The listener now has a role — abandoner — and will begin monitoring their own performance in it. That monitoring is the ayoniso manasikāra the discourse opens by diagnosing. The sentence delivering the remedy installs the condition.

The fix is available and costs nothing: "these fall away in the seeing."

That sentence assigns one thing. The practitioner sees. The falling away is what follows from seeing, not a second task performed afterwards — dassanā is ablative, a source, and the taints proceed from it. Even the seeing is not grammatically attributed to anyone, since dassanā is an action noun with no agent in it.

"Abandon these by seeing clearly" assigns two things: see, and then abandon. The second acquires a success condition, and a practitioner given a success condition will check whether they are meeting it. That checking is the extra work, and the Pali never asked for it.

Ranked by distance between the English construction and the source.

Divergence is highest for the instructions beginners hear first and lowest for the doctrinal formulas. Any effect is concentrated in early practice.

A grammar with no subject

Pāṇini's grammar has no category corresponding to "subject". It has six kārakas: roles defined by their relation to the action, which sits at the centre. The agent is defined at Aṣṭādhyāyī 1.4.54 as svatantraḥ, "the independent one" — a position in an event, assignable to inanimate things, not a substrate.

The category of grammatical subject entered European grammar around the twelfth century as suppositum. Neither Apollonius, nor Priscian, nor Pāṇini, nor the Arab grammarians had it. It is late and local, and English builds every clause around it.

The middle voice, with an example

Sanskrit marks on the verb itself whether the fruit of an action returns to the one acting. One syllable does it.

English has no morphological middle. It covers part of the same ground with the anticausative — the door opens, the jaw relaxes — and the rest with reflexives and bare intransitives. The anticausative is the closest match and still differs in three ways. It deletes the initiator instead of merging initiator and endpoint, so the speaker ends up outside the event rather than as its site. It is lexical rather than inflectional, available for relax and soften but not for know, dwell or contemplate, which are the verbs the source texts use. And it cannot be made imperative. Sanskrit forms middle imperatives — yajasva, vartasva — so an instruction can be issued in the middle voice. English instruction has only "relax your jaw", where you act on a part you own, and "let your jaw relax", which adds a permission-granting agent rather than removing one.

One limitation. Pali's middle is largely vestigial, so this argument holds for Sanskrit and Greek and applies only weakly to the Pali of MN 2. The middle-voice case is about the shape of the gap in English, not about a feature the source texts of section 5 were exploiting.

What a lone participant is grouped with

Every language decides how to treat the single participant of an intransitive clause. English groups it with agents: "I sat" patterns with "I broke the vase", so sitting is a kind of doing. Ergative languages group it with patients: "I sat" patterns with "the vase broke", so sitting is a kind of happening, and agency becomes the marked option requiring an extra morpheme. Hindi does this in the perfective — मैंने, "I-ERG" — after which the verb agrees with the object rather than the speaker.

• MN 2 Sabbāsavasutta. Pali and English at wordsofthebuddha.org and SuttaCentral. Tamil translation by நவாலியூர் சோ. நடராசன், SuttaCentral.
• Jakobson, R. (1959). On linguistic aspects of translation. Languages differ in what they must convey, not in what they may.
• Slobin, D. (1996). Thinking for speaking. The framing of experience for the purpose of putting it into words, which is what a guided instruction imposes from outside.
• Verma & Mohanan, eds. (1990). Experiencer Subjects in South Asian Languages. CSLI.
• Subbarao, K. V. (2012). South Asian Languages: A Syntactic Typology. Cambridge.
• Kemmer, S. (1993). The Middle Voice. Benjamins.
• Bhartṛhari, Vākyapadīya. The division of an event into participants as vikalpa, conceptual construction. The oldest version of this argument.
• Barthakuria, A. C. The Philosophy of Sanskrit Grammar: A Critical Study of Kāraka.
• Anālayo (2003). Satipaṭṭhāna: The Direct Path to Realization.
• Collins, S. (1982). Selfless Persons. Cambridge.
• Fausey & Boroditsky (2011). Cross-linguistic differences in eyewitness memory. The closest existing experimental paradigm. Read with Liberman's critique of the effect sizes on Language Log.
• Nietzsche, Genealogy of Morality I.13. There is no substratum behind the doing; the doer is added to the deed by grammatical habit.
• Bohm, D. (1980). Wholeness and the Implicate Order, ch. 2. The rheomode, and its instructive failure.