I have a blind computer science student who has defined a problem that he,personally, has with screen readers. He is currently taking my undergraduate computer science research course and would like to tackle a small piece of this problem for his research project.
The problem is that he would like the novels that he reads by way of his screen reader to differentiate between various speakers in a way that is obvious to him (for instance the change of voice). We understand how to do the backend of this process -- i.e. marking up the text so that the screen reader will change.
My problem is that I don't know how to advise this student as to how much or little to take on in terms of the text language processing. Would you fill me in on what part of this process he may be able to do with existing NLP tools. The problem is to be able to discern, with some degree of certainly, who is currently speaking as part of a dialouge.
Any help would be GREATLY appreciated!
Thanks,
Ann from Saint Mary's University
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Ann,
My experience in this area is somewhat limited and revolves around trying to identify speakers for quoted speech in newswire text. The main goal of this work was to resolve the "I" pronouns to these speakers but once you know the quote extent and the speaker then it's pretty straight forward. In that context it wasn't too difficult. The main issues are that the quotes can proceed the speaker, speaker continuation -- where the speaker isn't locally identified but is the same as a previous quote -- is pretty common and the editorial convention for quotes in that case are a little odd. In some newspapers a continuation doesn't have a close quote. You'll also have the issue of resolving pronouns for the "he said" cases. Nested quotes can be tricky and you may need to determine between real quotes and "special" quotes.
I think the majority of cases will be easy and the strange cases (like those below) have some standard conventions as well. You could probably do it right for most cases pretty easily and will probably never get them all (pretty much like every other NLP task). You could probably get pretty far with specific rules for finding the quote extents, a good list a verbs-of-saying, a pos-tagger for "special" quotes (these tend not have verbs) and a very simple pronoun resolution sceme . Hope this helps...Tom
"Let's just say this," said Larkin. "Bobby Bonilla plays in New York. Darryl Strawberry and Jose Canseco play in California. Those are bigger markets than Cincinnati, and I take that into consideration. The numbers don't have to be exactly the same, so that just because Barry Larkin is a better player, supposedly, I should be getting $5.8 million. But the numbers do have to be competitive. "'It would be beneficial to myself financially to sign next year after testing the market. . . . But that's not what I want. I want to play with the Reds on winning teams."
``I had seen a few cabs around the city that had some ornaments and they were not displayed correctly, as I would think they should have been,'' Ward said.
``So I decided to display them in the proper way,'' he said while cruising north on a Chicago expressway.
Ward serached the city's variety stores, picking up bundles of
tacky decorations for every occasion.
``I give away little hearts and candies for Valentine's Day,'' he said. ``St. Patrick's Day, I mostly have those four-leaf clovers and have the Irish green different things in the car.''
``Everyone loves it and nobody wants me to take the decorations down,'' he said. ``People are waving at me down the street. I had a police paddy wagon wave at me. State troopers wave at me. An ambulance tipped their lights at me.
``I had one lady that came to the cab and said, ```Ooh, the
Christmas cab is here!'''
Another woman took one look at his cab and said, ```That's just too Christmassy.' She decided that she'd wait for the next cab,'' he recalled.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have a blind computer science student who has defined a problem that he,personally, has with screen readers. He is currently taking my undergraduate computer science research course and would like to tackle a small piece of this problem for his research project.
The problem is that he would like the novels that he reads by way of his screen reader to differentiate between various speakers in a way that is obvious to him (for instance the change of voice). We understand how to do the backend of this process -- i.e. marking up the text so that the screen reader will change.
My problem is that I don't know how to advise this student as to how much or little to take on in terms of the text language processing. Would you fill me in on what part of this process he may be able to do with existing NLP tools. The problem is to be able to discern, with some degree of certainly, who is currently speaking as part of a dialouge.
Any help would be GREATLY appreciated!
Thanks,
Ann from Saint Mary's University
Hi Ann,
My experience in this area is somewhat limited and revolves around trying to identify speakers for quoted speech in newswire text. The main goal of this work was to resolve the "I" pronouns to these speakers but once you know the quote extent and the speaker then it's pretty straight forward. In that context it wasn't too difficult. The main issues are that the quotes can proceed the speaker, speaker continuation -- where the speaker isn't locally identified but is the same as a previous quote -- is pretty common and the editorial convention for quotes in that case are a little odd. In some newspapers a continuation doesn't have a close quote. You'll also have the issue of resolving pronouns for the "he said" cases. Nested quotes can be tricky and you may need to determine between real quotes and "special" quotes.
I think the majority of cases will be easy and the strange cases (like those below) have some standard conventions as well. You could probably do it right for most cases pretty easily and will probably never get them all (pretty much like every other NLP task). You could probably get pretty far with specific rules for finding the quote extents, a good list a verbs-of-saying, a pos-tagger for "special" quotes (these tend not have verbs) and a very simple pronoun resolution sceme . Hope this helps...Tom
"Let's just say this," said Larkin. "Bobby Bonilla plays in New York. Darryl Strawberry and Jose Canseco play in California. Those are bigger markets than Cincinnati, and I take that into consideration. The numbers don't have to be exactly the same, so that just because Barry Larkin is a better player, supposedly, I should be getting $5.8 million. But the numbers do have to be competitive. "'It would be beneficial to myself financially to sign next year after testing the market. . . . But that's not what I want. I want to play with the Reds on winning teams."
``I had seen a few cabs around the city that had some ornaments and they were not displayed correctly, as I would think they should have been,'' Ward said.
``So I decided to display them in the proper way,'' he said while cruising north on a Chicago expressway.
Ward serached the city's variety stores, picking up bundles of
tacky decorations for every occasion.
``I give away little hearts and candies for Valentine's Day,'' he said. ``St. Patrick's Day, I mostly have those four-leaf clovers and have the Irish green different things in the car.''
``Everyone loves it and nobody wants me to take the decorations down,'' he said. ``People are waving at me down the street. I had a police paddy wagon wave at me. State troopers wave at me. An ambulance tipped their lights at me.
``I had one lady that came to the cab and said, ```Ooh, the
Christmas cab is here!'''
Another woman took one look at his cab and said, ```That's just too Christmassy.' She decided that she'd wait for the next cab,'' he recalled.
Thanks for the help Tom!