There is a kind of speech that has no listener and is not private.
Talking to a photograph. Arguing, out loud, in the car, with somebody who is not in it. Addressing a dead parent while doing the washing up. Everybody does some version of this and almost nobody counts it as speech, because speech is supposed to have a recipient and this does not.
It also has recognisable shapes. The rehearsed argument. The plea. The recrimination. The sudden turn to sarcasm. These recur across people who have never compared notes, which is the signature of a form rather than of an accident.
Antya-līlā and Madhya 23 supply a tradition that noticed this, gave it a name, and divided it into ten.1
Citra-jalpa — literally variegated talk — is classified in the corpus's taxonomy as a division of mohana, the separation half of the highest ecstasy.2
It has ten named divisions, and its worked example is Rādhā's bhramara-gīta: the bumblebee song, in which she addresses a bee that has arrived from Kṛṣṇa's direction and conducts, through it, an entire argument with an absent person.3
Note what the classification asserts. Deranged speech is not the residue of the state; it is one of the state's two named components, the other being physical unsteadiness. A taxonomy that divides the summit into a body-register and a speech-register is claiming that the talking is not a symptom of the condition but part of what the condition is.
And ten divisions is a lot. That is not a gesture toward variety; it is somebody having sat with a body of material long enough to sort it.
The choice of interlocutor is doing the work and it is easy to enjoy without noticing.
A bee is the minimum viable addressee. It arrived from the right direction, which makes it a plausible messenger; it will not answer, which means every turn of the argument is hers; and it cannot be embarrassed, contradicted, or made to carry a reply back.
Which is exactly the specification for the kind of speech being catalogued. The form requires an addressee — you cannot conduct a recrimination at nothing — and requires that the addressee supply no resistance. A bee satisfies both and a person satisfies neither.
That is the same operation as questioning a garden about where somebody went, and the corpus performs both in the same stretch of chapters.4 The world is being staffed with things that can be addressed, and the staffing is not delusion in any simple sense — it is the provision of the one component the speech genre cannot run without.
The corpus does not say why it classified this, and the answer is visible in what classification does.
Unsorted, the speech of somebody in extremity is evidence about them. It is what a person in that state produces, and its features — the repetition, the address to absent parties, the sudden reversals — are read as the disorder showing through.
Sorted into ten named divisions, it becomes a genre. And a genre has conventions, which means an instance can be well or badly formed, which means it can be appreciated rather than only observed.
That is the whole transformation and it is not small. The tradition is not tolerating this speech. It is reading it as literature — and the bhramara-gīta is treated as a composition, quoted, studied and admired, rather than as a transcript of somebody in distress.
Which is either the most humane thing a community can do with the speech of a person in extremity, or a way of not hearing it. The corpus supplies no way to tell, and both readings survive everything on the page: the material is genuinely beautiful, it was genuinely produced by somebody at the end of their capacity, and the community responded by anthologising it.
There is a sampling question underneath the taxonomy and it is worth asking, because the answer changes what the ten divisions are.
A genre requires more than one author. Conventions are what recur across people, and a form identified from a single speaker's output is not a form — it is a description of that speaker.
The corpus's worked example is one composition, the bhramara-gīta, attributed to one figure. And the material the classification was applied to in practice was the recorded speech of one man over twelve years, through one channel, written down by one secretary.10
So the ten divisions are doing one of two very different things. Either they codify a genre with many instances, of which the bhramara-gīta is the exemplar — in which case the corpus is drawing on a literary tradition wider than itself. Or they are a sorting of one man's speech, given the dignity of a taxonomy.
The corpus does not say, and the two are not distinguishable from the text. What can be said is that the classification is presented as general and its evidence, in the range read, is singular.
The same taxonomic material supplies three further terms and they belong here, because they describe the mechanics the speech runs on.5
Bhāvodaya — an emotion arising. Bhāva-sandhi — two emotions meeting, at a junction. Bhāva-śābalya — emotions in collision, mixed and contending.
That third term is the useful one. It names a condition in which several affects are present simultaneously and are not resolving into one — which is a real state, is extremely common, and has almost no vocabulary in English beyond mixed feelings, a phrase so weak it implies mild ambivalence rather than contention.
And it explains the speech's characteristic feature. Sudden reversal — plea into accusation into tenderness — is not instability of the speaker. It is what a sequence of utterances looks like when the underlying affects are in collision and each surfaces in turn.
The image the corpus supplies for the physical version is the same: mad elephants entering a sugarcane field and fighting one another, flattening the crop.6 Collision, not intensity, doing the damage — stated once as a figure about the body and once as a technical term about the emotions, and never connected.
Layer A supplies the classification, the ten divisions, the bhramara-gīta and the three collision terms.7
Layer B expands the material into the gopīs' supreme position and into instruction on the reader's own devotional development.8
Worth marking the direction. Layer A is doing taxonomy — sorting, naming, dividing — and its interest is the structure of what was produced.
Layer B is doing aspiration. It is concerned with the reader's relation to the material rather than with the material's anatomy.
The ten divisions are Layer A's, and a reader arriving through the commentary meets a doctrine about exalted love without meeting the fact that somebody counted.
You talk out loud to somebody who is not there. That is a form, not a symptom. It has recurrent shapes across people who never compared notes, which is what a genre looks like, and treating it as a lapse is the only part of it that does any damage.
You are choosing who to say the unsayable thing to. Notice that the requirement is an addressee who supplies no resistance. That is why people rehearse to mirrors, pets and the dead, and why the substitute has to be present enough to be addressed and absent enough not to reply.
You cannot work out what you feel. Consider that the answer may be several things, contending. Bhāva-śābalya names a state English handles badly, and mixed feelings is a much weaker phrase than the condition it gets used for — the affects are not diluted, they are colliding.
You are receiving somebody's extremity as material. Ask whether classifying it is hearing it or replacing hearing it. This tradition anthologised the speech of a man at the end of his capacity, and the anthology is beautiful, and both of those are true at once.
The evidence is a technical classification with ten named divisions, one extended worked example, and three further terms for the arrival, junction and collision of affect.9
The tension the source doesn't close: the speech is catalogued as a genre and produced by somebody in extremity, and the corpus's response to it is literary rather than practical throughout.
A second one. Bhāva-śābalya names collision as the mechanism, and the corpus's figurative account of the physical damage — elephants fighting one another in a field — says the same thing. Neither passage refers to the other, and the convergence is the strongest evidence in the material that the mechanism was actually observed rather than inferred from doctrine.
Open: what the ten divisions are, individually. The extraction record names the count and the worked example and does not enumerate them, and nothing on this page rests on their content.
This page and Mādana and Mohana are a taxonomy and one of its two branches. That page holds the structure — the two divisions, the subdivisions, the ten transformations; this one holds what is inside the speech branch — and the corpus's development effort visibly went into this half.
Against Interrogating the Trees it is the same speech-form with a different staffing solution. There the addressees are sorted by grammatical gender and reasoned about; here a bee is enlisted because it arrived from the right direction — and in both the requirement is an interlocutor who cannot answer back.
And it belongs beside The Body That Changes Shape, which holds the collision image from the physical side. Elephants fighting one another in a sugarcane field and bhāva-śābalya are one claim in two registers, and the corpus never puts them together.
The Pennebaker Pronoun Diagnostic Framework — that page's mechanism is that the small words give the speaker away. Pronouns, prepositions and articles pass underneath editorial control: you can rehearse what you say and almost never rehearse the function-word skeleton it rides on, so the connective tissue reports what the content was arranged to conceal.
Citra-jalpa is a classification of the same material from the opposite end, and the pairing gets at what each is refusing to do.
The diagnostic tradition treats uncontrolled speech features as a channel to something behind them. The features are not the object; they are the route to a state, an intention or a status the speaker did not mean to disclose.
The taxonomy treats them as the object. Ten divisions of mad talk are ten kinds of talk, sorted by their own properties. Nothing is being inferred through them — no hidden state is being read off, no speaker is being assessed, and the classification would be identical whoever produced it.
Which gives the finding neither page reaches. The same uncontrolled speech supports two complete and incompatible projects: it can be read through, as evidence about a person, or read as, as a form with conventions. Each forecloses the other. A room that has decided the speech is a genre has stopped diagnosing the speaker — which is the whole of what this community did for twelve years, and it is why the corpus contains a magnificent anthology and no assessment of anybody's condition.
Sharpest implication. Uncontrolled speech can be read through, as evidence about the person, or read as, as a form with its own conventions — and the two foreclose each other. A community that classifies has stopped diagnosing. This one produced ten divisions and an anthology, and no assessment of anybody's state in twelve years, and whether that is the most humane available response or a way of not hearing is not decidable from anything on the page.
Generative questions.