Nominal Dependents

An adjective modifying a noun. The commonest member of the family, and the one that behaves most predictably.

How spaCy decides: an adjective sitting inside a noun phrase, attached to its noun. It makes no difference whether that noun phrase is the subject, the object, or inside a prepositional phrase.

Stacked adjectives do not chain. Each one attaches to the noun independently, so two adjectives produce two sibling arcs rather than a ladder.

A noun modifying another noun. English builds these freely — bread flour, chess club, kitchen window — and the first noun takes this label rather than amod.

How spaCy decides: the modifier is tagged as a noun rather than an adjective. That is the whole test, and it is why the label sometimes surprises you: if the parser reads a word as a noun, you get compound even where amod feels right.

An appositive — a second noun phrase renaming the first. Our guide and Amara are one person, named twice.

How spaCy decides: a nominal immediately following another nominal, usually set off by commas, with no verb or preposition joining them. The label goes on the second one and points back at the first.

A number modifying a noun. Straightforward, and behaves exactly like an adjective.

How spaCy decides: the modifier is a numeral. Spelled-out numbers and digits are treated identically.

The possessor. It attaches to the thing possessed — so in Priya’s telescope, Priya hangs off telescope rather than the other way round.

How spaCy decides: either a possessive pronoun (their, her) or a noun carrying ’s. The two look different in the sentence and identical in the tree.

The ’s is a separate word to the parser, with its own arc. A possessive pronoun has no clitic to split off, so it produces one arc instead of two.

A quantifier phrase modifier — the extra machinery inside a complex quantity like more than half. It is the rarest member of this family and does not occur at all in the app’s sentence corpus.

How spaCy decides: it labels the connecting word inside a multi-word quantity expression. A plain number is nummod; only the surrounding apparatus becomes quantmod.

nmod is in the taxonomy — it has a colour, a gloss and a place in this family — but the model Grammify runs produces it for exactly one construction. It does not appear once across the app’s corpus, and for ordinary noun-modifies-noun sentences it survived every attempt to elicit it deliberately. What does produce it is a currency sign: in $12 the number is the head and the sign hangs off it as nmod, reliably, for $, £ and € alike.

The reason is a difference of scheme. Universal Dependencies defines a general nominal modifier covering a noun that modifies another noun by any means. The English pipeline here uses the older, more specific labels instead: a bare noun modifier is compound, a possessor is poss, and anything joined by a preposition is prep plus pobj.