There needs to be an overriding exclusion list for tag verbs. Some writers place internal dialogue i.e. thoughts in quotes, tagged with thought or thinks
“It is well I drew the curtain,” thought I.
The verb closest to quoted dialogue should take precedence:
“Barbara,” she said to the servant who answered it…
The tag here should be said, not answered.
Bessie… sighed and said, “Poor Miss Jane.”
The tag here should be said, not sighed.
I thought they were saying, “Thank you”—Prof did all talking.
The tag here should be saying, not thought.
“Tell it to bear,” I said and ordered a liter…
The tag here should be said, not ordered.
“Yet,” I agreed; I would have argued, but…
The tag here should be agreed, not argued.
“Breakfast time,” he said eventually, casually — to prove, I’m sure, that he remembered all my human frailties.
The tag here should be said, not remembered.
Bilbo would have liked to say: “A warm bath and late breakfast on the lawn afterwards;” but he thought it better to say nothing at all.
The tag here should be say, not thought.
Interrupted speech or prose may not actually be tagged, or speech at all:
“You are not a servant at the hall, of course. You are–” He stopped, ran his eye…
“Well, now—” I scratched head and grinned.
Then decided he had to pay for his “crime”—but do it correctly.
“Uh—” She fisted me solidly in ribs, grinned.
She called his clarinet a “licorice stick.”
Clearly a wider set of training corpuses need to be run through the tag detection algorithm. Currently too much is being (mis)labeled.
- For UK-style dialogue, the single quote is often used:
‘Where are we, Gandalf?’ he asked.
‘Shoo!’ said Mr. Dursley loudly.
These fail to register as dialogue.