Audio description against general English

The Lancaster and Winter norms say where a word sits in the lexicon, where every word counts once. That is not where it sits in use, where a handful of common, abstract words take up most of the room. Against the lexicon, audio description is unremarkable in iconicity (d = +0.03); against running text it is clearly more iconic. The same pipeline is used throughout: spaCy lemmas, content words only, norms looked up per lemma and weighted by how often the lemma actually occurs.

One Billion Word benchmark

News wire: a single register, so this is a genre contrast rather than a comparison with English at large.

dimension	AD	reference	d	P(AD higher)
sight	3.58	2.98	+0.63	0.69
touch	1.59	1.11	+0.50	0.62
iconicity	3.81	3.53	+0.34	0.60
smell	0.47	0.42	+0.08	0.48
taste	0.32	0.32	+0.00	0.49
bodily / inner	0.85	1.13	-0.32	0.39
hearing	1.08	1.81	-0.72	0.28

British National Corpus

Balanced by design — fiction, academic prose, newspapers, letters, transcribed speech — so this is the fairer reference, and the one whose effect sizes are worth quoting.

dimension	AD	reference	d	P(AD higher)
sight	3.58	2.95	+0.64	0.69
touch	1.59	1.18	+0.41	0.61
iconicity	3.81	3.53	+0.34	0.60
smell	0.47	0.45	+0.03	0.47
taste	0.32	0.37	-0.08	0.46
bodily / inner	0.85	1.18	-0.37	0.38
hearing	1.08	1.73	-0.65	0.29