Replies: 50 comments 7 replies
-
|
Great observation. Actually, they are the same file according to md5sum.
I think probably they got copied someplace without the -p flag when being
transcribed
…On Sat, Jun 10, 2023 at 1:48 AM Shasetty ***@***.***> wrote:
Hi Sir/Madam.
Shankar from Bangalore.
I found, that the file englishPCFG.ser.gz, in version 4.5.3 & 4.5.4 was
saved on 3 Nov 2020.
Whereas other version from 4.2.2 till 4.5.2 has file englishPCFG.ser.gz,
saved on 14 May 2021.
1. 1 issue what I found, is punctuation issue. (text pasted below)
Text:
If an unforeseen event occurs or business conditions change, we may
use the proceeds of this offering differently than as described in this
prospectus. See “Risk Factors.”
------------------------------
Can you please inform, if there are any other issues?
—
Reply to this email directly, view it on GitHub
<#1363>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWIKS74TF7HW2JQU64LXKQYGDANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
There is a shift reduce parser which is more accurate in CoreNLP, and
Stanza's constituency parser is significantly more accurate. If you're
specifically trying to use the PCFG, we can hopefully provide you an
upgraded one using a silver dataset we'd been working on as part of the
Stanza project.
…On Sat, Jun 10, 2023 at 9:21 AM John Bauer ***@***.***> wrote:
Great observation. Actually, they are the same file according to md5sum.
I think probably they got copied someplace without the -p flag when being
transcribed
On Sat, Jun 10, 2023 at 1:48 AM Shasetty ***@***.***> wrote:
> Hi Sir/Madam.
> Shankar from Bangalore.
> I found, that the file englishPCFG.ser.gz, in version 4.5.3 & 4.5.4 was
> saved on 3 Nov 2020.
> Whereas other version from 4.2.2 till 4.5.2 has file englishPCFG.ser.gz,
> saved on 14 May 2021.
>
> 1. 1 issue what I found, is punctuation issue. (text pasted below)
> Text:
> If an unforeseen event occurs or business conditions change, we may
> use the proceeds of this offering differently than as described in this
> prospectus. See “Risk Factors.”
>
> ------------------------------
>
> Can you please inform, if there are any other issues?
>
> —
> Reply to this email directly, view it on GitHub
> <#1363>, or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/AA2AYWIKS74TF7HW2JQU64LXKQYGDANCNFSM6AAAAAAZBRRUPI>
> .
> You are receiving this because you are subscribed to this thread.Message
> ID: ***@***.***>
>
|
Beta Was this translation helpful? Give feedback.
-
|
As i am using "englishPCFG.ser.gz", i would be happy , if you provide me the upgraded PCFG. |
Beta Was this translation helpful? Give feedback.
-
|
Well... the dev set performance goes from 85.45 to 85.74. I'll see if
fiddling around with the weight a little improves that result, but I'm not
too hopeful. In fact, one would wonder if this is a sign that the English
silver dataset I built is not that great. If it were any good, I would
have expected it to be a big help to an older style model like the PCFG.
If you plan to continue using Java, you might consider using the shift
reduce parser, as its accuracy is 90 or a little higher, and it's also a
lot faster. There's also the Stanza constituency parser, where we have a
model which gets 96, but I get the impression you want to stick with Java
…On Sun, Jun 11, 2023 at 12:38 AM Shasetty ***@***.***> wrote:
As i am using "englishPCFG.ser.gz", i would be happy , if you provide me
the upgraded PCFG.
EMail id is ***@***.***
—
Reply to this email directly, view it on GitHub
<#1363 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWILZTKXDLQMI7B7HWTXKVYYVANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Thank you for, informing the Genuine status. As you are the right person with good skills on handling the empty space in the programme of PCFG. I will be waiting for your inputs. Stanza Constituency Parser As I am working on Stanford dependency parser. if I use Dependency parsing of Stanza, I will only get Universal Dependencies output. (https://stanfordnlp.github.io/stanza/depparse.html) Are there any possibilities to get basic Stanford dependency output using Stanza? If yes, please guide me. Shift-Reduce Constituency Parser It has old version, dated: stanford-srparser-2014-10-23-models, are there any recently updated version. if yes can you provide me. |
Beta Was this translation helpful? Give feedback.
-
|
I trained an updated SRParser a couple years ago - it's available here
https://search.maven.org/remotecontent?filepath=edu/stanford/nlp/stanford-corenlp/4.4.0/stanford-corenlp-4.4.0-models-english.jar
I don't think there's a way to get Stanford dependencies instead of
Universal from Stanza, unfortunately.
…On Mon, Jun 12, 2023 at 3:43 AM Shasetty ***@***.***> wrote:
Thank you for, informing the Genuine status.
As you are the right person with good skills on handling the empty space
in the programme of PCFG.
I will be waiting for your inputs.
------------------------------
*Stanza Constituency Parser*
As I am working on Stanford dependency parser.
if I use Dependency parsing of Stanza, I will only get Universal
Dependencies <https://universaldependencies.org/> output. (
https://stanfordnlp.github.io/stanza/depparse.html)
Are there any possibilities to get basic Stanford dependency output using
Stanza?
If yes, please guide me.
------------------------------
Shift-Reduce Constituency Parser
(https://nlp.stanford.edu/software/srparser.html)
It has old version, dated: stanford-srparser-2014-10-23-models, are there
any recently updated version.
if yes can you provide me.
—
Reply to this email directly, view it on GitHub
<#1363 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWN6CVGZVBEIZPBUTITXK3XENANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
We tried to run SRparser , we were not able to get the out put. Followed method as per : https://nlp.stanford.edu/software/srparser.html (Calling Parsing from Java) As this file was not available (stanford-postagger-3.5.0.jar) we downloaded http://www.java2s.com/Code/Jar/s/Downloadstanfordpostaggerjar.htm We are getting Dependency Parse (enhanced plus plus dependencies) : we don’t want this. We want basic dependency (as per Stanford). Please provide the command line to get basic dependency (as per Stanford) |
Beta Was this translation helpful? Give feedback.
-
|
Things are going to be a bit hectic here for another week, but I'm happy to
help how I can. Would you remind me the command line you are using to get
the SD in the first place? And, ideally, the command line you tried for
the SRParser
…On Wed, Jun 14, 2023 at 10:48 AM Shasetty ***@***.***> wrote:
We tried to run SRparser , we were not able to get the out put.
Followed method as per : https://nlp.stanford.edu/software/srparser.html
(Calling Parsing from Java)
As this file was not available (stanford-postagger-3.5.0.jar) we
downloaded
http://www.java2s.com/Code/Jar/s/Downloadstanfordpostaggerjar.htm
We are getting Dependency Parse (enhanced plus plus dependencies) : we
don’t want this.
We want basic dependency (as per Stanford).
Please provide the command line to get basic dependency (as per Stanford)
—
Reply to this email directly, view it on GitHub
<#1363 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWLH34XH4S7UGBDY323XLH2QXANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
-------------------------------- edu.stanford.nlp.parser.shiftreduce.ShiftReduceParser (Must specify a treebank to train from with -trainTreebank or a parser to load with -serializedPath) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.shiftreduce.ShiftReduceParser -parse.model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models-english\edu/stanford/nlp/models/srparser/englishSR.ser.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txt -------------------------------- edu.stanford.nlp.parser.nndep.DependencyParser (not working) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.nndep.DependencyParser -parse.model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models-english\edu\stanford\nlp\models\srparser\englishSR.ser.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txt -------------------------------- edu.stanford.nlp.parser.lexparser.LexicalizedParser (not working) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.lexparser.LexicalizedParser -parse.model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models-english\edu\stanford\nlp\models\srparser\englishSR.ser.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txt ----------------------------- edu.stanford.nlp.pipeline.StanfordCoreNLP(working) Dependency Parse (enhanced plus plus dependencies): java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.pipeline.StanfordCoreNLP -parse.model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models-english\edu\stanford\nlp\models\srparser\englishSR.ser.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txtI want out put of "basic dependency" |
Beta Was this translation helpful? Give feedback.
-
working codes of other parsers-------------------- original dependency(englishPCFG.ser.gz) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.lexparser.LexicalizedParser -retainTmpSubcategories -originalDependencies -outputFormat "typedDependencies" -outputFormatOptions "basicDependencies" G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models\edu\stanford\nlp\models\lexparser\englishPCFG.ser.gz G:\shiftreduce-corenlp-4.4.0\file.txt -------------------- Ud dependency(englishPCFG.ser.gz) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.lexparser.LexicalizedParser -retainTmpSubcategories -outputFormat "typedDependencies" -outputFormatOptions "basicDependencies" G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models\edu\stanford\nlp\models\lexparser\englishPCFG.ser.gz G:\shiftreduce-corenlp-4.4.0\file.txt -------------------- nndep(english_SD.gz) java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.parser.nndep.DependencyParser -model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models\edu\stanford\nlp\models\parser\nndep\english_SD.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txt -------------------- sr(StanfordCoreNLP) working Dependency Parse (enhanced plus plus dependencies): java -mx2048m -cp "G:\shiftreduce-corenlp-4.4.0/*" edu.stanford.nlp.pipeline.StanfordCoreNLP -parse.model G:\shiftreduce-corenlp-4.4.0\stanford-corenlp-4.4.0-models-english\edu\stanford\nlp\models\srparser\englishSR.ser.gz -textFile G:\shiftreduce-corenlp-4.4.0\file.txt |
Beta Was this translation helpful? Give feedback.
-
|
Two things - for CoreNLP to use the SRParser, you'll need
currently it's loading the neural dependency parser in CoreNLP, which doesn't do Stanford dependencies AFAIK To switch back to Stanford dependencies, you can do this
so here's what I ran and what I got: It should be noted that, for whatever improvements have been made to the dependency conversion from constituencies over the years, there aren't really any recent improvements to the SD, since everyone has moved on to UD. Still, you seem pretty determined to get SD and it is possible with CoreNLP, so this is how, I believe. There are output options which will hopefully make the output file in whatever format you need... let us know if you get stuck with that. Also, there is actually a converter which uses the Stanza constituency parser (again, it is much more accurate), but it ONLY does UD. If you want, we can connect it with the SD constituency -> dependency converter so that you can get SD, but that is the kind of change which will probably have to wait until after deadlines. |
Beta Was this translation helpful? Give feedback.
-
|
==========================(working sr parser witch basic dependency) =========================(working sr parser witch (enhanced plus plus dependencies) using 1st command line basic dependency, output was received. Thank you sir, for the command line. ============================================== |
Beta Was this translation helpful? Give feedback.
-
|
After using SRparser, i found, in SR parser "dep: dependent" outcomes are more , compared to 3.9.2 basic SD or NNDEp of 4.5.2. any possibility to reduce the "dep: dependent" outcomes in the text.Sample text attached below: This offering is being conducted on a firm commitment basis. The underwriter is obligated to take and pay for all of the ordinary shares if any such shares are taken. We have granted to the underwriter an option for a period of 45 days from the date of this prospectus to purchase up to 192,000 additional ordinary shares from us at the We plan to use the net proceeds from this offering for (i) general working capital (30%); (ii) business and team expansion by recruiting more professional consultants across With respect to the Company’s industry focused acquisitions, the Company plans to (i) purchase at least a majority interest in businesses it targets and not the assets of We have not currently identified any targets for acquisition. Pending use of proceeds from this offering, we intend to invest the proceeds in bank accounts, short-term, interest-bearing, investment-grade instruments, or hold as cash. The foregoing represents our current intentions based upon our present plans and business conditions to use and allocate the net proceeds of this offering. Our management, however, will have significant flexibility and discretion to apply the net proceeds of this offering. If an unforeseen event occurs or business conditions change, we may use the proceeds of this offering differently than as described in this prospectus. See “Risk Factors.” |
Beta Was this translation helpful? Give feedback.
-
|
its not 4.5.2 , it is 4.5.4 |
Beta Was this translation helpful? Give feedback.
-
|
I'm not experiencing the same issue. For example, this is what I get for a couple of the sentences you gave. What are you getting? |
Beta Was this translation helpful? Give feedback.
-
|
Typically a |
Beta Was this translation helpful? Give feedback.
-
|
sr parser 4.5.4 ; dep = 17 nndep 4.5.4 ; dep = 9 + (other modifiers are not identifed properly by the parser) lexparser 4.5.4 ; dep = 15 lexparser 3.9.2 ; dep = 9 any possibility to fix the left out dependency in the converters, and provide the best parser. |
Beta Was this translation helpful? Give feedback.
-
|
there is a section where it produces a dep: this looks like a bad parse. the second NML should be an NP, inside a large NP, so something like gonna skip. presumably with the parse tree the parser produced, one could make the dep there a conj, but it still wouldn't be right anyway |
Beta Was this translation helpful? Give feedback.
-
|
this section looks like we should be able to come up with a dependency for it: looking for similar annotations in EWT there are some better comparables in Craft, but those don't have gold dependencies (do they?) is there any reason this isn't just appos, like in EWT? also from that sentence: this is more clearly an I will have to check with my PI if either of these are fixable. For the first, we could decide |
Beta Was this translation helpful? Give feedback.
-
|
Anyway, that's the summary for what I have so far. There's still going to be some It might also be worth taking a step back and asking why the use of SD instead of UD - the gold UD training data available makes it much easier to build a direct to dependencies parser, meaning the output dependencies will be much more accurate. |
Beta Was this translation helpful? Give feedback.
-
a) Considering the inputs of : b) what is available in SD and what is merged UD (UD precision reduced) vmod : acl neg : advmod prep :case nn :compound xcomp :xcomp c) The dependency relationship completely changes between SD and UDPIPE. (I will provide the text with examples later.) |
Beta Was this translation helpful? Give feedback.
-
We finished early today. early jj rb |
Beta Was this translation helpful? Give feedback.
-
|
https://nlp.stanford.edu/software/stanford-corenlp-4.5.4b.zip |
Beta Was this translation helpful? Give feedback.
-
|
I guess what I'm saying is the direct-to-UD parser in Stanza might be better at everything, but you know your requirements better than I do. |
Beta Was this translation helpful? Give feedback.
-
|
Many Thanks for the release in short duration, Sir. Surely i will go through UD parser in Stanza. If i find any issues , i will update you sir, do the best, what you can. |
Beta Was this translation helpful? Give feedback.
-
|
I need to, have your opinion on one subject, can you share your personal mail id, please. |
Beta Was this translation helpful? Give feedback.
-
|
It should be available on my profile, but if it's NLP related, you could
consider posting it here
…On Mon, Jul 10, 2023, 8:27 AM Shasetty ***@***.***> wrote:
I need to, have your opinion on one subject, can you share your personal
mail id, please.
—
Reply to this email directly, view it on GitHub
<#1363 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWJKRF2MBSRJPYIEFP3XPQNNPANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
content for the opinion , is getting prepared, it will take some time, sir.NLP related: "This" from below text, in 1) obj and 2) nsubj. which one is correct.
UD4.5.4case(point-4, From-1) SD4.5.4prep(find-9, From-1)
|
Beta Was this translation helpful? Give feedback.
-
|
In cases like this, I would recommend checking with gold standard datasets to see how they handle similar cases. For example, from UD (EWT), there is "who considered it ...", which is very similar to "I found this ..." https://github.com/UniversalDependencies/UD_English-EWT Here, the relation is https://universaldependencies.org/en/dep/xcomp.html
I will point out that I am only the 12th most important annotator for UD-EWT (out of 12) and therefore cannot be fully trusted when it comes to UD analysis. There is also the issue of, I can certainly ask my PI about these kinds of things, but I only speak to him for about an hour per week and I'm reluctant to use too much of that time exploring the corners of UD annotation. When it's not clear what the right answer is, you can always post a question about it on the UD website instead: https://github.com/UniversalDependencies/docs/issues/new If there are cases where a reasonable constituency parse gets converted into the wrong dependency graph, you can certainly post about it here and we'll try to improve the converter over time. |
Beta Was this translation helpful? Give feedback.
-
|
dobj(find-9, this-10) : for text 1 |
Beta Was this translation helpful? Give feedback.
-
|
next query case(fairness-3,In-1) | case(fairness-3, In-1) |
Beta Was this translation helpful? Give feedback.
-
|
Again I would ask, what is the standard annotation, and work from there. "Did X" is generally treated as an aux of X, as per the example in EWT of and the UD description of "aux" specifically covers "to do": |
Beta Was this translation helpful? Give feedback.
-
|
I will look into this, but being far from the expert on dependencies, I
will have to take it to my PI next Tuesday.
For reference, is there one produced by one version of the parser or the
other?
Also, when you compare SD to UD, they have different treatments of which
words should be the head in certain cases. Even the root of an entire
sentence can be different
…On Tue, Jul 11, 2023 at 5:53 PM Shasetty ***@***.***> wrote:
next query
text: *In all fairness, she did try to phone the police.*
which one is correct among the two.
case(fairness-3,In-1) | case(fairness-3, In-1)
det(fairness-3,all-2) | det(fairness-3, all-2)
obl(try-7,fairness-3) | obl(did-6, fairness-3)
nsubj(try-7,she-5) | nsubj(did-6, she-5)
aux(try-7,did-6) | root(ROOT-0, did-6)
root(ROOT-0,try-7) | ccomp(did-6, try-7)
mark(phone-9,to-8) | case(phone-9, to-8)
xcomp(try-7,phone-9) | obl(try-7, phone-9)
det(police-11,the-10) | det(police-11, the-10)
obj(phone-9,police-11) | obj(try-7, police-11)
—
Reply to this email directly, view it on GitHub
<#1363 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWK3JLGFNIQVYDYGZZLXPXYQXANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Please take the output for both sentences in UD 4.5.4 and check them. From a personal point of view, I find this. Further for text: In all fairness, she did try to phone the police. As you and your team are considered one of the best, I would like to know the team's opinion. |
Beta Was this translation helpful? Give feedback.
-
|
Again, if you come up with situations where the converter is messing up a good constituency parse, we can work on that. If the constituency parse starts off bad, there's only so much we can do - certainly we won't want to overcompensate in the converter in such a way that a pattern which is the correct constituency parse for a different sentence is now worsened. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, that's exactly what I mean - I do not believe the nsubj in text 2 fits
the pattern as shown in the sentences I cited in EWT or the UD docs
…On Wed, Jul 12, 2023 at 4:33 PM Shasetty ***@***.***> wrote:
dobj(find-9, this-10) : for text 1
nsubj(solution-13, this-10) : for text 2
—
Reply to this email directly, view it on GitHub
<#1363 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA2AYWM65FZONQBL57WTC4LXP4X6LANCNFSM6AAAAAAZBRRUPI>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
Hi Sir/Madam.
Shankar from Bangalore.
I found, that the file englishPCFG.ser.gz, in version 4.5.3 & 4.5.4 was saved on 3 Nov 2020.
Whereas other version from 4.2.2 till 4.5.2 has file englishPCFG.ser.gz, saved on 14 May 2021.
Text:
If an unforeseen event occurs or business conditions change, we may use the proceeds of this offering differently than as described in this prospectus. See “Risk Factors.”
Can you please inform, if there are any other issues?
Beta Was this translation helpful? Give feedback.
All reactions