a sentential form
Midway through building a string, you have a half-finished thing: part real symbols, part placeholders still to be expanded — like a sentence template with some words filled in and some blanks remaining. Each such intermediate string that shows up along a derivation is called a sentential form. It is a snapshot of the construction in progress.
Formally, a sentential form is any string of terminals and variables that is derivable from the start symbol: any w with S =>* w, where w may contain both kinds of symbols. The very first sentential form is the start symbol S by itself (zero steps). If, moreover, w contains ONLY terminals, it is called a sentence (or just a string of the language) — the special sentential forms with no variables left are exactly the members of the generated language. So 'sentential form' is the general notion, and 'sentence' is the finished case.
Sentential forms are the natural objects to talk about when describing a derivation step (each step turns one sentential form into the next) and when proving things by induction about what a grammar can produce. A refinement: a LEFT-sentential form is one that appears in a leftmost derivation, and a RIGHT-sentential form one that appears in a rightmost derivation; these restricted notions matter in parsing theory, where bottom-up parsers work with right-sentential forms. A common slip is to call an intermediate string with leftover variables a 'sentence' — it is only a sentential form until every variable is gone.
With S → A B, A → a, B → b, the derivation S => AB => aB => ab passes through the sentential forms S, AB, aB, and ab. Of these only 'ab' (all terminals) is a sentence; the others still contain variables.
A sentential form is any S =>* w with terminals and/or variables; if w has only terminals it is a sentence.
A sentential form may still contain variables; only a sentential form made entirely of terminals (a 'sentence') belongs to the language. The lone start symbol S is itself a sentential form (the zero-step one).