Gender is not
about meaning

Three separate headwords in the corpus mean, in whole or in part, "wife" — and the corpus assigns each of them a different gender, with no semantic logic behind the assignment at all:

दार is grammatically masculine — and, distinctly, it has no singular or dual at all: the corpus's own paradigm is blank in those columns, and only the plural is ever attested, so "my wife" is necessarily , plural in form, one person in fact. कलत्र is neuter — and the corpus's own gloss lists a second, unrelated sense first (, "the waist" and "the wife"), so even within one headword, gender does not track meaning consistently. भार्या is the one that is feminine, declining exactly like रमा on page 5 — but it is one choice out of three the language actually makes, not the only one available.

Gender is not decoration. It decides two separate things, and getting it wrong breaks both:

• Which paradigm the noun itself takes. Page 5 showed six stem families; within the largest of them, अ-stems split by gender alone — देव (m.) and फल (n.) share a stem-shape but decline on two different rows from the very first case, and a feminine in that family does not stay अ-final at all — it takes (4.1.4) and becomes an आ-stem, a different pattern entirely.
• What every word agreeing with it must do. An adjective, a participle, a demonstrative or relative pronoun modifying a noun copies that noun's gender — not once, but at every point in the sentence where it appears. Get the noun's gender wrong and every agreeing word downstream is wrong with it; this is not a small, local error, it is one that propagates through the whole clause.

The pronoun page leans on this directly, and so will the sentence page later in the series: a demonstrative or relative pronoun's form is read off the gender of the noun it points to, not the other way round.

(4.1.4) attaches टाप् to an अ-final stem to mark it feminine — रमा, page 5's example, is the result. It is by far the strongest signal this page has:

The exceptions are not counterexamples to टाप् — they are words टाप् never had a chance to apply to, because they are not adjectives or ordinary nouns taking the affix at all. Most are agent-noun compounds built directly on the root पा "to drink / to protect" with क्विप्, not on टाप्:

गोपा "cow-herd" and सोमपा "soma-drinker" are masculine agent nouns that happen to end in the same long आ टाप् produces — a coincidence of shape, not a counterexample to the rule. हाहा, a गन्धर्व's name, is simply a proper noun that keeps its own gender regardless of its ending.

(1.4.3) names every long-ई word denoting a female नदी-class; a separate, closed list of consonant stems takes ङीप् to reach the same ending (, e.g. दर्वी "ladle" from दर्वि, listed पक्षे, "optionally," alongside the plain form). Together:

Here the exceptions cluster into two real patterns rather than scattering randomly — which is itself worth knowing, because it means "-ई, therefore feminine" fails in predictable places:

One cluster is agent nouns in -नी from the root नी "to lead" (अग्रणी, ग्रामणी, सेनानी — "the one who leads," masculine by what it names, an agent). The other is बहुव्रीहि adjectives built on धी "intellect" (सुधी, दुर्धी, प्रधी — "of good/bad/ first-rate intellect") — these are not exceptions to gender assignment at all, they are the three-way-adjective pattern this page ends on, just caught here because their masculine declension happens to end in the same ई.

Page 5 already showed that अ-stems divide by gender rather than staying one pattern. Counted:

Zero feminine अ-stems is not a coincidence, it is the flip side of the आ-stat above: every feminine that would otherwise end in अ takes टाप् instead and leaves this bucket entirely. Within what remains, the neuter half carries its own visible signal — every neuter अ-stem's own citation form (nominative singular) ends in -म्:

Abstract nouns built with क्तिन् — गति, मति, कृति, जाति — are feminine. But "-ति words are feminine" as a blanket rule fails about a third of the time:

The masculine third is not noise — it clusters almost entirely around पति "lord, husband" itself and its compounds:

पति was never built with क्तिन् at all — it just happens to end in the same two syllables. Strip the compounds built on it (गणपति, अहर्पति, उमापति…) and the "-ति, therefore feminine" rule is far stronger than the raw count suggests; keep them in, and the honest number is what's printed above.

Sanskrit builds abstract ("-ness") nouns from adjectives two ways, and the choice of suffix alone decides the gender — the underlying idea doesn't change at all. वृद्धत्व and a hypothetical वृद्धता would mean the identical thing, "old age"; only the suffix's own gender differs:

There is no ambiguity to report here — of the suffixes on this page, these are the two cleanest. The interesting fact is not the exception count (there barely is one); it's that two different word-formation strategies for the same semantic job were assigned two different genders, arbitrarily, and both stuck.

Any root can form an agent noun with तृच् — कर्तृ "doer," ज्ञातृ "knower." Counted by actual gender in this corpus:

The neuter thirteen are not a second exception — they are the very same agent nouns, simply declined to modify a neuter head noun; page 5's ऋ-stem family already shows this pattern (कर्तृ as m.). The real exception is the feminine handful, and the Liṅगानुशासन names it directly:

Five kinship words are named — मातृ, दुहितृ, स्वसृ, यातृ, ननान्दृ — and this corpus's own ऋ-final feminine list matches it almost exactly, compounds included (लोकमातृ, वीरमातृ…). Two of the five, स्वसृ and ननान्दृ, don't even share the surface -तृ that the other three do — a reminder that "ends in तृ" and "is one of the five kin exceptions" are related but not identical questions.

पुत्र "son" is masculine for the plainest possible reason (it names a male), sitting right beside six neuters that share its ending and mean everything from "field" to "friend." A quarter of this bucket being the "wrong" gender is too high to treat -त्र as a signal at all.

आत्मन् "self" and राजन् "king" are masculine; कर्मन् "action" and नामन् "name" are neuter; सीमन् "boundary" is one of only two attested feminines. Nothing about the ending itself leans either way — this is the cleanest example on the page of "learn the word," not "learn the pattern."

Neuter ब्रह्मन् and masculine ब्रह्मन् are not "the same word, either gender fine" the way गो below is. The corpus's own glosses diverge sharply:

Neuter ब्रह्म is the impersonal Absolute, the Veda, even ascetic penance; masculine ब्रह्मा is the personal creator-god. Say the wrong gender and you have named a different concept, not inflected the same one incorrectly.

— the Liṅगानुशासन names गो among a short list of words that are simply स्त्रीपुंसयोः, "usable for either," with its own commentary noting ("this — or that — [is] गौः"). Unlike every suffix-based signal above, गो does not change shape at all between genders:

One paradigm, गौः/गावौ/गावः throughout, serving "cow" and "ox" alike — which gender is meant is read off context and agreement, never off the noun's own form. This corpus's own headword happens to be a spectacularly polysemous one besides — its gloss lists eleven distinct senses, from "cow" to "ray of light" to "the four cardinal directions" — which is one more reason not to reason about gender from meaning.

Masculine, "one born again"; feminine, "a woman remarried" — two real, distinct social categories the tradition names, sharing one stem purely because both involve पुनः + भू, "happening again."

Sūtras 176–182 open an entire section — , "in [the case of] masculine-and-neuter" — devoted to nouns attested in both genders with, unlike गो, no meaning distinction at all: gives ("this [is] घृतः; this [is] घृतम्" — one substance, either gender). Thirty-seven words are named across the six sūtras; this corpus independently attests of them with exactly the two genders predicted:

(Liṅगानुशासन 189) — every सर्वादि pronoun is अविशिष्टलिङ्ग, "gender-unspecified": it takes whatever gender the noun it stands for or points to has, rather than owning a fixed one of its own the way देव owns masculine. The corpus marks this literally — अस्मद् "I/we" and युष्मद् "you" carry a fourth linga code, distinct from पुं./स्त्री./नपुं., that simply means "not applicable":

"I" and "you" refer to whoever is speaking or spoken to — asking "what gender is अहम्?" is a category error, and the corpus's own indexing agrees: it doesn't force an answer.

(174) singles out उ-final गुणवचन (quality-denoting) words as त्रिलिङ्ग, illustrating with — पटु as masculine, feminine and neuter, three shapes for one adjective. 186 repeats the point for the whole class, not just उ-final words: — ("शुक्लः cloth [m.], शुक्ला [f.], शुक्लं garment [n.]" — one meaning, "white," three obligatory shapes).

This corpus's own कटु "pungent, harsh" is exactly that pattern, attested in all three genders:

Notice the feminine here declines identically to the masculine (कटुः, कटू, कटवः throughout) rather than taking a separate ङीप् form the way an ordinary feminine noun would — id 174's own gloss allows both पटुः and पट्वी as valid feminines for this exact class; this corpus's headword records the first option. Whichever shape a given गुणवचन word's feminine takes, the underlying point survives: the adjective supplies no gender of its own. Every cell above is "correct" only in agreement with a noun that is already masculine, feminine or neuter on its own account — exactly the fact page 5 deferred to this page, and the fact the whole guessing game below is built to test.