At Abhidharmakośa VI .3, Vasubandhu analyses the phrase sandhāya ... bha $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{s} $$ ita $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{m} $$ as used in the sūtras. Here bhā $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{s} $$ ita $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{m} $$ mentions an utterance, to which a figurative sense is ascribed by the gerundive (not noun) sandhāya. The audience is split: some are intended to understand the literal, others the figurative sense. Vasubandhu's analysis works well for sandhābhā $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{s} $$ a etc. in the Saddharmapu $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{n}$$ $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{d}$$ arīka and the Guhyasamājatantra. (The Hevajratantra is (...) different and we consider it separately). Poussin's “arrière-pensée”, Edgerton's “real meaning” and “esoteric meaning”, and Wayman's “twilight language” are attractive in some cases but not all. Vasubandhu's analysis combined with Edgerton's “specially intended” (for ābhiprāyika) is much better; compare V. S. Bhattacharya's old paraphrase of this word “intended to imply or suggest something different from what is expressed by the words”. But Bhattacharya's widely accepted translation “intentional” for ābhiprāyika is misleading and misrepresents Vasubandhu, and we suggest ways of improving it in the light of the need for coherent translation, with the help of some considerations from Pā $$\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle\cdot}$}}{n}$$ ini and from Montague semantics. (shrink)