Re: question about sentence segmentation
Very pleased to see so many people offer suggestions! Comparing some of these different methods might make an interesting student project. Sean: Just an fyi. Does that make sense? Haven't had my coffee ... Makes perfect sense, the downside is it requires some kind of higher level understanding during sentence segmentation to understand what the hierarchy is. You could imagine something that looks similar but with a different logical structure. Long term, some big joint model that does all things simultaneously is definitely something I'm interested in. Steve: Seems like rather than specifying a set of candidate characters, we want to specify a candidate boundary regular expression. This might be something that would be possible with minimal changes to the model. John: why not just split sentences with regex's off a small list of defined onc physical exam terms? My preference for vanilla ctakes is always to do basic linguistic things like tokenization and sentence segmentation without reference to context-specific rules, just because it makes them less portable. Obviously for specific use cases or applications (like what Britt is probably doing) you will use whatever information makes sense for your domain. But I think we could get maybe 75% of the remaining cases (which are probably only 5% of the total # of cases) by using smarter boundary conditions like Steve suggested. Thanks again, Tim On 08/02/2014 01:26 PM, John Green wrote: I was thinking the same thing as Steve. Thats a pretty regular onc physical exam, why not just split sentences with regex's off a small list of defined onc physical exam terms? The interesting case would be breast, as this term may appear in the body of a sentence (rather than just a term), but u could use a regex sub match where u conditionally match breast first then one or more key physical findings to correctly identify THAT breast word token as the term, eg beginning of the sentence. I would recommend red flag physical findings as they are more likely to always been in the body of the sentence, for example, Breast: no lumps or masses palpable. I have a few other ideas if thats barking up the right tree. JG — Sent from Mailbox for iPhone On Sat, Aug 2, 2014 at 8:58 AM, Steven Bethard steven.beth...@gmail.com wrote: On Sat, Aug 2, 2014 at 7:43 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: PE: Lymphnodes: neck and axilla without adenopathy Lungs: normal and clear to auscultation CV: regular rate and rhythm without murmur or gallop , S1, S2 normal, no murmur, click, rub or gal*, chest is clear without rales or wheezing, no pedal edema, no JVD, no hepatosplenomegaly Breast: negative findings right/left breast with mild swelling, warmth, mild erythema, slightly tender, no seroma or hematoma Abdomen: Abdomen soft, non-tender. It would be preferable to me to put sentence breaks in between the sections, so the first two sentences would be: 1) PE: Lymphonodes... 2) Lungs: normal... [snip] Another example that breaks our model in a different way (truncated): 1. Baseline labwork including tumor markers 2. Start DD AC on Friday 8/1 with RN chemo teach 3. S U parent study [snip] Here it would be preferable to get: 1. Baseline labwork... 2. Start DD... 3. S U parent study Seems like rather than specifying a set of candidate characters, we want to specify a candidate boundary regular expression. Something like, \p{P}|\b\p{Lu}|\b\p{N}, should cover all of the above cases: sentence boundaries may appear at punctuation marks, at uppercase letters after word boundaries, and at numbers after a word boundaries. Steve
Re: question about sentence segmentation
One method I use for finding headings is term followed by either 2 or more instances of white space or a symbol (colon, comma, dash) followed by 1 or more instances of white space. Its really naive but works well because the term is from a controlled set. Thats not super helpful in your first example above unless those sections can be predefined. Your second example seems a lot harder. Especially when there are valid number/period patterns at the end of the line. Patient presented with fever of 102. or other measurements. On Sat, Aug 2, 2014 at 7:43 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: I'm annotating some oncology notes from SHARP right now, and they are basically a nightmare for our current sentence segmentation model. Mainly because they eschew explicit markers between sentences. I thought I'd ping the list with some interesting examples just in case it stimulates ideas. But it seems to me that at some point we'll have to augment the opennlp module (preferable) or roll our own to handle cases like these. In this example a bunch of background is on one line with no punctuation between logical breaks: PE: Lymphnodes: neck and axilla without adenopathy Lungs: normal and clear to auscultation CV: regular rate and rhythm without murmur or gallop , S1, S2 normal, no murmur, click, rub or gal*, chest is clear without rales or wheezing, no pedal edema, no JVD, no hepatosplenomegaly Breast: negative findings right/left breast with mild swelling, warmth, mild erythema, slightly tender, no seroma or hematoma Abdomen: Abdomen soft, non-tender. It would be preferable to me to put sentence breaks in between the sections, so the first two sentences would be: 1) PE: Lymphonodes... 2) Lungs: normal... but without any candidate characters to split the sentence I don't think it is possible. Another example that breaks our model in a different way (truncated): 1. Baseline labwork including tumor markers 2. Start DD AC on Friday 8/1 with RN chemo teach 3. S U parent study Our model will break on the period after the number, so we'd probably get: 1. Baseline labwork including tumor markers 2. Start DD 3. S U parent study So the number is going in exactly the wrong place. Here it would be preferable to get: 1. Baseline labwork... 2. Start DD... 3. S U parent study Anyways, just something to think about! The problem is much more complex in clinical data than in edited text, but I'm sure we all knew that already :) Tim From: Miller, Timothy [timothy.mil...@childrens.harvard.edu] Sent: Monday, July 28, 2014 2:38 PM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation Yes, you're right about that Britt. I've been doing some annotations side by side with a treebank viewer and think I have a pretty good handle on the actual rules. Basically, if a header or list identifier is followed by a period or a newline it is considered a sentence break and otherwise it is part of the sentence. e.g. 1. 20 mg flomax is two sentences, while: 1 - 20 mg flomax is one sentence. For headings: Allergies: Pt is allergic to aspirin. is one sentence, while: Allergies: Pt is allergic to aspirin. is two sentences. I'm planning to follow these guidelines. Tim On 07/28/2014 01:53 PM, britt fitch wrote: Thanks for the document, Tim. It seems to not be explicit about how to handle sentences occurring in lists. Are you still considering having the list number as outside of the sentence? Thanks Britt On Jul 25, 2014, at 7:09 AM, Miller, Timothy timothy.mil...@childrens.harvard.edumailto: timothy.mil...@childrens.harvard.edu wrote: Checking with Guergana and other colleagues here the advice is to have the sentence segmenter follow the treebank guidelines for sentence segmentation: http://clear.colorado.edu/compsem/documents/treebank_guidelines.pdf They are a bit light on detail but fortunately we have some treebanked data so I will use that for the training data and hopefully that will illuminate the tricky cases. Tim From: Masanz, James J. [masanz.ja...@mayo.edumailto:masanz.ja...@mayo.edu ] Sent: Tuesday, July 15, 2014 4:39 PM To: 'dev@ctakes.apache.orgmailto:dev@ctakes.apache.org' Subject: RE: question about sentence segmentation Sorry, I don't know if there was a reason. If you haven't checked with Guergana, you might want to ask her if she had a reason or if it was just the way it had been since that corpus was created. -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 3:34 PM To: dev@ctakes.apache.orgmailto:dev@ctakes.apache.org Subject: Re: question about sentence segmentation Thanks James, I was hoping to hear from you. I'll probably go ahead and change the data to split sentences between the list header
RE: question about sentence segmentation
Hi Tim, It would be preferable to me to put sentence breaks in between the sections, so the first two sentences would be: 1) PE: Lymphonodes... 2) Lungs: normal... The punctuation is (always) after the logical break, being Term: for a Term:Definition list. I think that the first three sentences should be 1) PE: 2) Lymphnodes: neck and ... 3) CV: regular and ... Where the first line is an overarching Term: sentence (tree root), because each Term:Definition line that follows is within the physical exam. Just an fyi. Does that make sense? Haven't had my coffee ... Sean -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Saturday, August 02, 2014 7:44 AM To: dev@ctakes.apache.org Subject: RE: question about sentence segmentation I'm annotating some oncology notes from SHARP right now, and they are basically a nightmare for our current sentence segmentation model. Mainly because they eschew explicit markers between sentences. I thought I'd ping the list with some interesting examples just in case it stimulates ideas. But it seems to me that at some point we'll have to augment the opennlp module (preferable) or roll our own to handle cases like these. In this example a bunch of background is on one line with no punctuation between logical breaks: PE: Lymphnodes: neck and axilla without adenopathy Lungs: normal and clear to auscultation CV: regular rate and rhythm without murmur or gallop , S1, S2 normal, no murmur, click, rub or gal*, chest is clear without rales or wheezing, no pedal edema, no JVD, no hepatosplenomegaly Breast: negative findings right/left breast with mild swelling, warmth, mild erythema, slightly tender, no seroma or hematoma Abdomen: Abdomen soft, non-tender. It would be preferable to me to put sentence breaks in between the sections, so the first two sentences would be: 1) PE: Lymphonodes... 2) Lungs: normal... but without any candidate characters to split the sentence I don't think it is possible. Another example that breaks our model in a different way (truncated): 1. Baseline labwork including tumor markers 2. Start DD AC on Friday 8/1 with RN chemo teach 3. S U parent study Our model will break on the period after the number, so we'd probably get: 1. Baseline labwork including tumor markers 2. Start DD 3. S U parent study So the number is going in exactly the wrong place. Here it would be preferable to get: 1. Baseline labwork... 2. Start DD... 3. S U parent study Anyways, just something to think about! The problem is much more complex in clinical data than in edited text, but I'm sure we all knew that already :) Tim From: Miller, Timothy [timothy.mil...@childrens.harvard.edu] Sent: Monday, July 28, 2014 2:38 PM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation Yes, you're right about that Britt. I've been doing some annotations side by side with a treebank viewer and think I have a pretty good handle on the actual rules. Basically, if a header or list identifier is followed by a period or a newline it is considered a sentence break and otherwise it is part of the sentence. e.g. 1. 20 mg flomax is two sentences, while: 1 - 20 mg flomax is one sentence. For headings: Allergies: Pt is allergic to aspirin. is one sentence, while: Allergies: Pt is allergic to aspirin. is two sentences. I'm planning to follow these guidelines. Tim On 07/28/2014 01:53 PM, britt fitch wrote: Thanks for the document, Tim. It seems to not be explicit about how to handle sentences occurring in lists. Are you still considering having the list number as outside of the sentence? Thanks Britt On Jul 25, 2014, at 7:09 AM, Miller, Timothy timothy.mil...@childrens.harvard.edumailto:timothy.mil...@childrens.harv ard.edu wrote: Checking with Guergana and other colleagues here the advice is to have the sentence segmenter follow the treebank guidelines for sentence segmentation: http://clear.colorado.edu/compsem/documents/treebank_guidelines.pdf They are a bit light on detail but fortunately we have some treebanked data so I will use that for the training data and hopefully that will illuminate the tricky cases. Tim From: Masanz, James J. [masanz.ja...@mayo.edumailto:masanz.ja...@mayo.edu] Sent: Tuesday, July 15, 2014 4:39 PM To: 'dev@ctakes.apache.orgmailto:dev@ctakes.apache.org' Subject: RE: question about sentence segmentation Sorry, I don't know if there was a reason. If you haven't checked with Guergana, you might want to ask her if she had a reason or if it was just the way it had been since that corpus was created. -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 3:34 PM
Re: question about sentence segmentation
On Sat, Aug 2, 2014 at 7:43 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: PE: Lymphnodes: neck and axilla without adenopathy Lungs: normal and clear to auscultation CV: regular rate and rhythm without murmur or gallop , S1, S2 normal, no murmur, click, rub or gal*, chest is clear without rales or wheezing, no pedal edema, no JVD, no hepatosplenomegaly Breast: negative findings right/left breast with mild swelling, warmth, mild erythema, slightly tender, no seroma or hematoma Abdomen: Abdomen soft, non-tender. It would be preferable to me to put sentence breaks in between the sections, so the first two sentences would be: 1) PE: Lymphonodes... 2) Lungs: normal... [snip] Another example that breaks our model in a different way (truncated): 1. Baseline labwork including tumor markers 2. Start DD AC on Friday 8/1 with RN chemo teach 3. S U parent study [snip] Here it would be preferable to get: 1. Baseline labwork... 2. Start DD... 3. S U parent study Seems like rather than specifying a set of candidate characters, we want to specify a candidate boundary regular expression. Something like, \p{P}|\b\p{Lu}|\b\p{N}, should cover all of the above cases: sentence boundaries may appear at punctuation marks, at uppercase letters after word boundaries, and at numbers after a word boundaries. Steve
Re: question about sentence segmentation
Thanks for the document, Tim. It seems to not be explicit about how to handle sentences occurring in lists. Are you still considering having the list number as outside of the sentence? Thanks Britt On Jul 25, 2014, at 7:09 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: Checking with Guergana and other colleagues here the advice is to have the sentence segmenter follow the treebank guidelines for sentence segmentation: http://clear.colorado.edu/compsem/documents/treebank_guidelines.pdf They are a bit light on detail but fortunately we have some treebanked data so I will use that for the training data and hopefully that will illuminate the tricky cases. Tim From: Masanz, James J. [masanz.ja...@mayo.edu] Sent: Tuesday, July 15, 2014 4:39 PM To: 'dev@ctakes.apache.org' Subject: RE: question about sentence segmentation Sorry, I don't know if there was a reason. If you haven't checked with Guergana, you might want to ask her if she had a reason or if it was just the way it had been since that corpus was created. -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 3:34 PM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation Thanks James, I was hoping to hear from you. I'll probably go ahead and change the data to split sentences between the list header and list element. You don't happen to know if there is any principled reason for the original style or whether it was just an arbitrary convention? The only thing I can think of is it might be hard to learn when to separate when there is no period after the list header (as in your examples). I think it's worth empirically checking on that point, but there might be other reasons that I'm not thinking of. Thanks Tim On 07/15/2014 03:27 PM, Masanz, James J. wrote: I don't have an opinion about how it should work. But I can verify that the clinical notes from Mayo Clinic that were used in the initial cTAKES sentence detector model had the list markers included in the first sentence, so, for example, the following would be two sentences, with each line a separate sentence. #1 Dilated esophagus. #2 Adenocarcinoma -- James -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 6:04 AM To: dev@ctakes.apache.org Subject: RE: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. I can get behind this, I think it makes the issue a bit cleaner, to either have the list header as non-sentential or it's own sentence. As far as I can tell, this is not the current default behavior. At least in my runs the list header seems to get attached to the first following sentence, even in cases where it starts with a digit and a period (3. Magnesium oxide 400 mg p.o. daily. is all one sentence). This behavior is probably strongly dependent on the annotations we give the sentence detector so as I'm prepping new training data I should have a default in mind. Does anyone have any objections to changing the sentence detector behavior to break list headers (things like 3. or A or #5) as their own sentence? Tim From: Britt Fitch [britt.fi...@gmail.com] Sent: Monday, July 14, 2014 8:29 AM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. That won't be as straightforward as splitting on a period in cases like 2. Magnesium oxide 400 mg p.o. daily. In cases where there are more than 1 written sentence like your example in the original email, I'd prefer those were each a sentence rather than making the entire list line a single sentence. My feeling is that each line without terminating punctuation would be a single sentence and would exclude the list number. As an aside, I have encountered several issues with numbered lists being interpreted differently depending on 1. what number is included at the start for example: 2. Magnesium oxide 400 mg p.o. daily. vs 12. Magnesium oxide 400 mg p.o. daily. (This appears to be a chunking issue where the line starting with 12. Magnesium is identified as starting with chunks [O, O, B-NP, B-NP, I-NP, B-NP, B-ADVP, O] even though the parts of speech appear to be correct) 2. whether there is a period at the end of a list for example: 4. CHF vs 4. CHF. (This appears to be an issue with the chunker though which produces [O,O] in the first case and [B-VP, B-NP, O] in the second. Cheers, Britt
Re: question about sentence segmentation
Yes, you're right about that Britt. I've been doing some annotations side by side with a treebank viewer and think I have a pretty good handle on the actual rules. Basically, if a header or list identifier is followed by a period or a newline it is considered a sentence break and otherwise it is part of the sentence. e.g. 1. 20 mg flomax is two sentences, while: 1 - 20 mg flomax is one sentence. For headings: Allergies: Pt is allergic to aspirin. is one sentence, while: Allergies: Pt is allergic to aspirin. is two sentences. I'm planning to follow these guidelines. Tim On 07/28/2014 01:53 PM, britt fitch wrote: Thanks for the document, Tim. It seems to not be explicit about how to handle sentences occurring in lists. Are you still considering having the list number as outside of the sentence? Thanks Britt On Jul 25, 2014, at 7:09 AM, Miller, Timothy timothy.mil...@childrens.harvard.edumailto:timothy.mil...@childrens.harvard.edu wrote: Checking with Guergana and other colleagues here the advice is to have the sentence segmenter follow the treebank guidelines for sentence segmentation: http://clear.colorado.edu/compsem/documents/treebank_guidelines.pdf They are a bit light on detail but fortunately we have some treebanked data so I will use that for the training data and hopefully that will illuminate the tricky cases. Tim From: Masanz, James J. [masanz.ja...@mayo.edumailto:masanz.ja...@mayo.edu] Sent: Tuesday, July 15, 2014 4:39 PM To: 'dev@ctakes.apache.orgmailto:dev@ctakes.apache.org' Subject: RE: question about sentence segmentation Sorry, I don't know if there was a reason. If you haven't checked with Guergana, you might want to ask her if she had a reason or if it was just the way it had been since that corpus was created. -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 3:34 PM To: dev@ctakes.apache.orgmailto:dev@ctakes.apache.org Subject: Re: question about sentence segmentation Thanks James, I was hoping to hear from you. I'll probably go ahead and change the data to split sentences between the list header and list element. You don't happen to know if there is any principled reason for the original style or whether it was just an arbitrary convention? The only thing I can think of is it might be hard to learn when to separate when there is no period after the list header (as in your examples). I think it's worth empirically checking on that point, but there might be other reasons that I'm not thinking of. Thanks Tim On 07/15/2014 03:27 PM, Masanz, James J. wrote: I don't have an opinion about how it should work. But I can verify that the clinical notes from Mayo Clinic that were used in the initial cTAKES sentence detector model had the list markers included in the first sentence, so, for example, the following would be two sentences, with each line a separate sentence. #1 Dilated esophagus. #2 Adenocarcinoma -- James -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 6:04 AM To: dev@ctakes.apache.orgmailto:dev@ctakes.apache.org Subject: RE: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. I can get behind this, I think it makes the issue a bit cleaner, to either have the list header as non-sentential or it's own sentence. As far as I can tell, this is not the current default behavior. At least in my runs the list header seems to get attached to the first following sentence, even in cases where it starts with a digit and a period (3. Magnesium oxide 400 mg p.o. daily. is all one sentence). This behavior is probably strongly dependent on the annotations we give the sentence detector so as I'm prepping new training data I should have a default in mind. Does anyone have any objections to changing the sentence detector behavior to break list headers (things like 3. or A or #5) as their own sentence? Tim From: Britt Fitch [britt.fi...@gmail.commailto:britt.fi...@gmail.com] Sent: Monday, July 14, 2014 8:29 AM To: dev@ctakes.apache.orgmailto:dev@ctakes.apache.org Subject: Re: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. That won't be as straightforward as splitting on a period in cases like 2. Magnesium oxide 400 mg p.o. daily. In cases where there are more than 1 written sentence like your example in the original email, I'd prefer those were each a sentence rather than making the entire list line a single sentence. My feeling is that each line without terminating punctuation would
RE: question about sentence segmentation
I don't have an opinion about how it should work. But I can verify that the clinical notes from Mayo Clinic that were used in the initial cTAKES sentence detector model had the list markers included in the first sentence, so, for example, the following would be two sentences, with each line a separate sentence. #1 Dilated esophagus. #2 Adenocarcinoma -- James -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 6:04 AM To: dev@ctakes.apache.org Subject: RE: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. I can get behind this, I think it makes the issue a bit cleaner, to either have the list header as non-sentential or it's own sentence. As far as I can tell, this is not the current default behavior. At least in my runs the list header seems to get attached to the first following sentence, even in cases where it starts with a digit and a period (3. Magnesium oxide 400 mg p.o. daily. is all one sentence). This behavior is probably strongly dependent on the annotations we give the sentence detector so as I'm prepping new training data I should have a default in mind. Does anyone have any objections to changing the sentence detector behavior to break list headers (things like 3. or A or #5) as their own sentence? Tim From: Britt Fitch [britt.fi...@gmail.com] Sent: Monday, July 14, 2014 8:29 AM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. That won't be as straightforward as splitting on a period in cases like 2. Magnesium oxide 400 mg p.o. daily. In cases where there are more than 1 written sentence like your example in the original email, I'd prefer those were each a sentence rather than making the entire list line a single sentence. My feeling is that each line without terminating punctuation would be a single sentence and would exclude the list number. As an aside, I have encountered several issues with numbered lists being interpreted differently depending on 1. what number is included at the start for example: 2. Magnesium oxide 400 mg p.o. daily. vs 12. Magnesium oxide 400 mg p.o. daily. (This appears to be a chunking issue where the line starting with 12. Magnesium is identified as starting with chunks [O, O, B-NP, B-NP, I-NP, B-NP, B-ADVP, O] even though the parts of speech appear to be correct) 2. whether there is a period at the end of a list for example: 4. CHF vs 4. CHF. (This appears to be an issue with the chunker though which produces [O,O] in the first case and [B-VP, B-NP, O] in the second. Cheers, Britt On Mon, Jul 14, 2014 at 7:50 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: Just curious about an edge case regarding headers/lists and wondering what people think the correct behavior and annotation are. In cases like this: #1 Dilated esophagus. #2 Adenocarcinoma my intuition is that each whole line is one sentence. But then there are cases where the number may be followed by multiple sentences on one line. 1. EGD as a complex procedure. If there is an abnormality, obtain biopsies. For this example my intuition is not as clear. Should there be a break after the 1. or should the first sentence be 1. EGD as a complex procedure.? Again, my intuition leans towards the latter but it seems a bit odd since the 1. kind of distributes over all the following sentences (i.e. it's like a paragraph descriptor.) Does the period after the 1 matter? The number of sentences after the list header? The fact that it's all on one line? Anything else? Tim
Re: question about sentence segmentation
Thanks James, I was hoping to hear from you. I'll probably go ahead and change the data to split sentences between the list header and list element. You don't happen to know if there is any principled reason for the original style or whether it was just an arbitrary convention? The only thing I can think of is it might be hard to learn when to separate when there is no period after the list header (as in your examples). I think it's worth empirically checking on that point, but there might be other reasons that I'm not thinking of. Thanks Tim On 07/15/2014 03:27 PM, Masanz, James J. wrote: I don't have an opinion about how it should work. But I can verify that the clinical notes from Mayo Clinic that were used in the initial cTAKES sentence detector model had the list markers included in the first sentence, so, for example, the following would be two sentences, with each line a separate sentence. #1 Dilated esophagus. #2 Adenocarcinoma -- James -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 6:04 AM To: dev@ctakes.apache.org Subject: RE: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. I can get behind this, I think it makes the issue a bit cleaner, to either have the list header as non-sentential or it's own sentence. As far as I can tell, this is not the current default behavior. At least in my runs the list header seems to get attached to the first following sentence, even in cases where it starts with a digit and a period (3. Magnesium oxide 400 mg p.o. daily. is all one sentence). This behavior is probably strongly dependent on the annotations we give the sentence detector so as I'm prepping new training data I should have a default in mind. Does anyone have any objections to changing the sentence detector behavior to break list headers (things like 3. or A or #5) as their own sentence? Tim From: Britt Fitch [britt.fi...@gmail.com] Sent: Monday, July 14, 2014 8:29 AM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. That won't be as straightforward as splitting on a period in cases like 2. Magnesium oxide 400 mg p.o. daily. In cases where there are more than 1 written sentence like your example in the original email, I'd prefer those were each a sentence rather than making the entire list line a single sentence. My feeling is that each line without terminating punctuation would be a single sentence and would exclude the list number. As an aside, I have encountered several issues with numbered lists being interpreted differently depending on 1. what number is included at the start for example: 2. Magnesium oxide 400 mg p.o. daily. vs 12. Magnesium oxide 400 mg p.o. daily. (This appears to be a chunking issue where the line starting with 12. Magnesium is identified as starting with chunks [O, O, B-NP, B-NP, I-NP, B-NP, B-ADVP, O] even though the parts of speech appear to be correct) 2. whether there is a period at the end of a list for example: 4. CHF vs 4. CHF. (This appears to be an issue with the chunker though which produces [O,O] in the first case and [B-VP, B-NP, O] in the second. Cheers, Britt On Mon, Jul 14, 2014 at 7:50 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: Just curious about an edge case regarding headers/lists and wondering what people think the correct behavior and annotation are. In cases like this: #1 Dilated esophagus. #2 Adenocarcinoma my intuition is that each whole line is one sentence. But then there are cases where the number may be followed by multiple sentences on one line. 1. EGD as a complex procedure. If there is an abnormality, obtain biopsies. For this example my intuition is not as clear. Should there be a break after the 1. or should the first sentence be 1. EGD as a complex procedure.? Again, my intuition leans towards the latter but it seems a bit odd since the 1. kind of distributes over all the following sentences (i.e. it's like a paragraph descriptor.) Does the period after the 1 matter? The number of sentences after the list header? The fact that it's all on one line? Anything else? Tim
RE: question about sentence segmentation
Sorry, I don't know if there was a reason. If you haven't checked with Guergana, you might want to ask her if she had a reason or if it was just the way it had been since that corpus was created. -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 3:34 PM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation Thanks James, I was hoping to hear from you. I'll probably go ahead and change the data to split sentences between the list header and list element. You don't happen to know if there is any principled reason for the original style or whether it was just an arbitrary convention? The only thing I can think of is it might be hard to learn when to separate when there is no period after the list header (as in your examples). I think it's worth empirically checking on that point, but there might be other reasons that I'm not thinking of. Thanks Tim On 07/15/2014 03:27 PM, Masanz, James J. wrote: I don't have an opinion about how it should work. But I can verify that the clinical notes from Mayo Clinic that were used in the initial cTAKES sentence detector model had the list markers included in the first sentence, so, for example, the following would be two sentences, with each line a separate sentence. #1 Dilated esophagus. #2 Adenocarcinoma -- James -Original Message- From: Miller, Timothy [mailto:timothy.mil...@childrens.harvard.edu] Sent: Tuesday, July 15, 2014 6:04 AM To: dev@ctakes.apache.org Subject: RE: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. I can get behind this, I think it makes the issue a bit cleaner, to either have the list header as non-sentential or it's own sentence. As far as I can tell, this is not the current default behavior. At least in my runs the list header seems to get attached to the first following sentence, even in cases where it starts with a digit and a period (3. Magnesium oxide 400 mg p.o. daily. is all one sentence). This behavior is probably strongly dependent on the annotations we give the sentence detector so as I'm prepping new training data I should have a default in mind. Does anyone have any objections to changing the sentence detector behavior to break list headers (things like 3. or A or #5) as their own sentence? Tim From: Britt Fitch [britt.fi...@gmail.com] Sent: Monday, July 14, 2014 8:29 AM To: dev@ctakes.apache.org Subject: Re: question about sentence segmentation My preference is to treat the list row number as outside of the sentence of interest. Or if it is necessary to be included in a sentence, have it be a sentence on its own. That won't be as straightforward as splitting on a period in cases like 2. Magnesium oxide 400 mg p.o. daily. In cases where there are more than 1 written sentence like your example in the original email, I'd prefer those were each a sentence rather than making the entire list line a single sentence. My feeling is that each line without terminating punctuation would be a single sentence and would exclude the list number. As an aside, I have encountered several issues with numbered lists being interpreted differently depending on 1. what number is included at the start for example: 2. Magnesium oxide 400 mg p.o. daily. vs 12. Magnesium oxide 400 mg p.o. daily. (This appears to be a chunking issue where the line starting with 12. Magnesium is identified as starting with chunks [O, O, B-NP, B-NP, I-NP, B-NP, B-ADVP, O] even though the parts of speech appear to be correct) 2. whether there is a period at the end of a list for example: 4. CHF vs 4. CHF. (This appears to be an issue with the chunker though which produces [O,O] in the first case and [B-VP, B-NP, O] in the second. Cheers, Britt On Mon, Jul 14, 2014 at 7:50 AM, Miller, Timothy timothy.mil...@childrens.harvard.edu wrote: Just curious about an edge case regarding headers/lists and wondering what people think the correct behavior and annotation are. In cases like this: #1 Dilated esophagus. #2 Adenocarcinoma my intuition is that each whole line is one sentence. But then there are cases where the number may be followed by multiple sentences on one line. 1. EGD as a complex procedure. If there is an abnormality, obtain biopsies. For this example my intuition is not as clear. Should there be a break after the 1. or should the first sentence be 1. EGD as a complex procedure.? Again, my intuition leans towards the latter but it seems a bit odd since the 1. kind of distributes over all the following sentences (i.e. it's like a paragraph descriptor.) Does the period after the 1 matter? The number of sentences after the list header? The fact that it's all