In preparation for next week's quarterly planning, I'd like to restate some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter. (n.b. These are more like "devil's advocate" thoughts, lest I make it sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
Of course the counterargument to this is that the current state of auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
The other thing (a more obvious one) that makes Wikidata descriptions redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary. On top of all that, it creates an unnecessary synchronization cost, fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
This means that we would need to implement the same kind of moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration. If we really want to get it right, we have to figure this out before proceeding.
-Dmitry
Both of those are very good points.
I am no expert on Wikidata so I'm unqualified to assess how much weight to give those concerns.
I do wonder if it would be better to focus on encouraging more mobile app users to become Wikipedia and/or Commons contibutors instead of Wikidata contributors. Thoughts?
Pine On Mar 22, 2015 8:57 AM, "Dmitry Brant" dbrant@wikimedia.org wrote:
In preparation for next week's quarterly planning, I'd like to restate some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter. (n.b. These are more like "devil's advocate" thoughts, lest I make it sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
Of course the counterargument to this is that the current state of auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
The other thing (a more obvious one) that makes Wikidata descriptions redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary. On top of all that, it creates an unnecessary synchronization cost, fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
This means that we would need to implement the same kind of moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration. If we really want to get it right, we have to figure this out before proceeding.
-Dmitry
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
On 22 March 2015 at 09:53, Pine W wiki.pine@gmail.com wrote:
I am no expert on Wikidata so I'm unqualified to assess how much weight to give those concerns.
I do wonder if it would be better to focus on encouraging more mobile app users to become Wikipedia and/or Commons contibutors instead of Wikidata contributors. Thoughts?
I've spun this off into a separate thread to keep this conversation focussed: [Apps] Encouraging Wikipedia/Commons contributions (was: Wikidata descriptions).
Dan
Responses inline...
On Mar 22, 2015, at 8:57 AM, Dmitry Brant dbrant@wikimedia.org wrote:
In preparation for next week's quarterly planning, I'd like to restate some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter. (n.b. These are more like "devil's advocate" thoughts, lest I make it sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
I disagree with the premise that the description being "data" means it is missing its promise if it is human curated. I am more concerned with the quality of the description.
Of course the counterargument to this is that the current state of auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
It's surprisingly hard to create auto generated descriptions that rival the quality of user generated descriptions.
Deeply hard, in fact, because it's complicated not only by language syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc).
This already complicated situation then becomes many orders of magnitude more difficult because these qualitative factors can differ between languages.
The other thing (a more obvious one) that makes Wikidata descriptions redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary.
The description needs to be able to be shorter than the first sentence in the article.
On top of all that, it creates an unnecessary synchronization cost, fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
I predict this won't be any worse than what happened when we enabled section editing.
This means that we would need to implement the same kind of moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration.
I've been editing descriptions from the Wikidata site directly for months and only one, of dozens I've added or edited were reverted.
If we really want to get it right, we have to figure this out before proceeding.
-Dmitry
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
I agree with Monte Hurd and would add that my personal volunteer time on Wiki projects, though unique, is not irreplaceable, and the idea that I and others interested in my area of editing could be "overworked" by some new technology is just silly. I think you need to have a little faith in the whole concept of crowd-sourcing. It really does seem to work. Automated descriptions sounds like a terrible idea and I have seen time and again on all sorts of subjects that the main claim to fame switches across languages. An example is a language -pedia that has added a town because it is the location of a castle that is notable in that language -pedia for whatever reason, while in the language -pedia of the town itself, the town may be better known as a hub on a railway network or some such thing.
On Sun, Mar 22, 2015 at 8:08 PM, Monte Hurd mhurd@wikimedia.org wrote:
Responses inline...
On Mar 22, 2015, at 8:57 AM, Dmitry Brant dbrant@wikimedia.org wrote:
In preparation for next week's quarterly planning, I'd like to restate
some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter.
(n.b. These are more like "devil's advocate" thoughts, lest I make it
sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a
machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to
manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
I disagree with the premise that the description being "data" means it is missing its promise if it is human curated. I am more concerned with the quality of the description.
Of course the counterargument to this is that the current state of
auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
It's surprisingly hard to create auto generated descriptions that rival the quality of user generated descriptions.
Deeply hard, in fact, because it's complicated not only by language syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc).
This already complicated situation then becomes many orders of magnitude more difficult because these qualitative factors can differ between languages.
The other thing (a more obvious one) that makes Wikidata descriptions
redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No
problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary.
The description needs to be able to be shorter than the first sentence in the article.
On top of all that, it creates an unnecessary synchronization cost,
fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a
missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if
we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
I predict this won't be any worse than what happened when we enabled section editing.
This means that we would need to implement the same kind of
moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative
mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration.
I've been editing descriptions from the Wikidata site directly for months and only one, of dozens I've added or edited were reverted.
If we really want to get it right, we have to figure this out before
proceeding.
-Dmitry
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
As I have stated before, some items can benefit from a manual description, but the vast majority could be described sufficiently by a machine, based on the item's statements. Having people type in "German composer (X-Y)" is a waste of time, and not scalable with 14M items times 300 or so languages, even with our large volunteer base. Adding a single statement (e.g. "award received") would instantly improve an automatic description in all languages; but if we rely on manual descriptions alone, all descriptions in all languages, would have to be updated manually. Descriptions will be missing, incomplete, out of date, and full of vandalism on the "obscure" languages. This is not Wikipedia, where Tanalog has 64K articles, but Wikidata, where Tanalog has 15M items with few people to patrol them.
Automatic descriptions may not (yet!) be as eloquent as some humans, but in most cases, they could be "good enough". Let machines do the grunt work where they can, so humans can focus on the work only they can do.
On Sun, Mar 22, 2015 at 7:29 PM Jane Darnell jane023@gmail.com wrote:
I agree with Monte Hurd and would add that my personal volunteer time on Wiki projects, though unique, is not irreplaceable, and the idea that I and others interested in my area of editing could be "overworked" by some new technology is just silly. I think you need to have a little faith in the whole concept of crowd-sourcing. It really does seem to work. Automated descriptions sounds like a terrible idea and I have seen time and again on all sorts of subjects that the main claim to fame switches across languages. An example is a language -pedia that has added a town because it is the location of a castle that is notable in that language -pedia for whatever reason, while in the language -pedia of the town itself, the town may be better known as a hub on a railway network or some such thing.
On Sun, Mar 22, 2015 at 8:08 PM, Monte Hurd mhurd@wikimedia.org wrote:
Responses inline...
On Mar 22, 2015, at 8:57 AM, Dmitry Brant dbrant@wikimedia.org wrote:
In preparation for next week's quarterly planning, I'd like to restate
some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter.
(n.b. These are more like "devil's advocate" thoughts, lest I make it
sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a
machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to
manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
I disagree with the premise that the description being "data" means it is missing its promise if it is human curated. I am more concerned with the quality of the description.
Of course the counterargument to this is that the current state of
auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
It's surprisingly hard to create auto generated descriptions that rival the quality of user generated descriptions.
Deeply hard, in fact, because it's complicated not only by language syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc).
This already complicated situation then becomes many orders of magnitude more difficult because these qualitative factors can differ between languages.
The other thing (a more obvious one) that makes Wikidata descriptions
redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No
problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary.
The description needs to be able to be shorter than the first sentence in the article.
On top of all that, it creates an unnecessary synchronization cost,
fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a
missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if
we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
I predict this won't be any worse than what happened when we enabled section editing.
This means that we would need to implement the same kind of
moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative
mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration.
I've been editing descriptions from the Wikidata site directly for months and only one, of dozens I've added or edited were reverted.
If we really want to get it right, we have to figure this out before
proceeding.
-Dmitry
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of course I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of Wikipedia articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
1) People wouldn't need to edit Wikidata descriptions, and would instead focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor only the article, instead of the article plus its corresponding Wikidata description.
-Dmitry
On Sun, Mar 22, 2015 at 3:29 PM, Jane Darnell jane023@gmail.com wrote:
I agree with Monte Hurd and would add that my personal volunteer time on Wiki projects, though unique, is not irreplaceable, and the idea that I and others interested in my area of editing could be "overworked" by some new technology is just silly. I think you need to have a little faith in the whole concept of crowd-sourcing. It really does seem to work. Automated descriptions sounds like a terrible idea and I have seen time and again on all sorts of subjects that the main claim to fame switches across languages. An example is a language -pedia that has added a town because it is the location of a castle that is notable in that language -pedia for whatever reason, while in the language -pedia of the town itself, the town may be better known as a hub on a railway network or some such thing.
On Sun, Mar 22, 2015 at 8:08 PM, Monte Hurd mhurd@wikimedia.org wrote:
Responses inline...
On Mar 22, 2015, at 8:57 AM, Dmitry Brant dbrant@wikimedia.org wrote:
In preparation for next week's quarterly planning, I'd like to restate
some of my concerns regarding Wikidata descriptions and flesh them out more comprehensively, since we're featuring them more prominently in the upcoming quarter.
(n.b. These are more like "devil's advocate" thoughts, lest I make it
sound like the Apps team isn't unified in its vision, which it certainly is.)
My reservations fall under two categories:
== Philosophical ==
Wikidata is a superbly valuable repository of *data* -- data that a
machine can use to generate all kinds of results that us humans can consume. The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
To allow users to manually fill in the Wikidata description (i.e. to
manually duplicate the contents of Wikipedia) is to miss the point of the true potential of Wikidata, which is to be able to *use* the data to generate the description automatically!
I disagree with the premise that the description being "data" means it is missing its promise if it is human curated. I am more concerned with the quality of the description.
Of course the counterargument to this is that the current state of
auto-generated descriptions is not quite good (they often sound strange or nonsensical), but that's only because the tools we have at our disposal for generating descriptions are still in their infancy. I don't deny that this will be a hard problem to solve, but in my view, this is ultimately the *correct* problem to solve.
It's surprisingly hard to create auto generated descriptions that rival the quality of user generated descriptions.
Deeply hard, in fact, because it's complicated not only by language syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc).
This already complicated situation then becomes many orders of magnitude more difficult because these qualitative factors can differ between languages.
The other thing (a more obvious one) that makes Wikidata descriptions
redundant is the first sentence of every Wikipedia article which, on its own, is intended to provide a concise description of the article (and many articles already do this with rather good consistency). In fact, as we speak, we're working on programmatically "cleaning up" the first sentence to make it even more concise. Why not simply use this as the description?
Is the first sentence sometimes too long to be a good description? No
problem: create a markup annotation that will denote the *portion* of the first sentence that will serve as the description. In any case, making users manually copy the content from the first sentence (which is from where most of the current Wikidata descriptions appear to be derived) seems extraordinarily unnecessary.
The description needs to be able to be shorter than the first sentence in the article.
On top of all that, it creates an unnecessary synchronization cost,
fulfillable only by a human contributor, between the two sources of data.
So, what I mean to say is: every edit to the Wikidata description is a
missed opportunity to edit the Wikipedia article in such a way that the description could be auto-generated correctly. (or, similarly, a missed opportunity to edit the *data* of the Wikidata entry in such a way that the description could be auto-generated correctly)
== Practical ==
If we open the floodgates to editing the Wikidata description (i.e. if
we make it too easy to edit the description), I predict that we'll be very disappointed by the quality of the contributions we'll get. I can see it quickly devolving into a whole lot of noise, spam, and vandalism.
I predict this won't be any worse than what happened when we enabled section editing.
This means that we would need to implement the same kind of
moderation/administration schemes that currently exist on Wikipedia itself. I'm by no means qualified to speak for the Community, but I doubt that many Wikipedians will want to double their workload by having to "watch" the Wikidata description of their favorite articles, in addition to the articles themselves.
I'll also point out that we do not yet expose any administrative
mechanisms in the mobile apps. This means that users will routinely see their edits disappear or be reverted without any notification or explanation. This is already the case for the general editing of article content in the apps, but since the description is featured much more prominently, any edits (or reverts) to it will be much more noticeable, and will surely add to the confusion and frustration.
I've been editing descriptions from the Wikidata site directly for months and only one, of dozens I've added or edited were reverted.
If we really want to get it right, we have to figure this out before
proceeding.
-Dmitry
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of course I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of Wikipedia articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor only the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now), which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description? This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article. The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher < lydia.pintscher@wikimedia.de> wrote:
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of
course
I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of
Wikipedia
articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor only the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Responses inline...
On Mar 22, 2015, at 1:53 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now),
They are also used in "Recent" and "Nearby" and Vibha wants them in "Saved Pages" list as well.
which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description?
This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article.
For example, take the enwiki "Fish" article.
The first couple sentences are these:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. Included in this definition are the living hagfish, lampreys, and cartilaginous and bony fish, as well as various extinct related groups.*
So if the we reduce the description to its first sentence we have:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
Now, for the sake of argument, let's imagine the *bold *words below represent a best case scenario for a relevant *subset* of the first sentence:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
"A fish is a gill-bearing aquatic animal".
That's nice and short and descriptive and reads like a little sentence. It's arguably the best reduction of the first sentence possible.
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence. You have to know *what a fish is* to know what parts of the first sentence are *most* important.
In other words, the "best" description is much more qualitative than it is quantifiable.
type of living organism typified by living in water and having gills
The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher < lydia.pintscher@wikimedia.de> wrote:
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of
course
I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of
Wikipedia
articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor only the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
_______________________________________________ Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Ignore that last reply - I accidentally hit submit while mid-way through writing it.
On Sun, Mar 22, 2015 at 6:03 PM, Monte Hurd mhurd@wikimedia.org wrote:
Responses inline...
On Mar 22, 2015, at 1:53 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now),
They are also used in "Recent" and "Nearby" and Vibha wants them in "Saved Pages" list as well.
which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description?
This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article.
For example, take the enwiki "Fish" article.
The first couple sentences are these:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. Included in this definition are the living hagfish, lampreys, and cartilaginous and bony fish, as well as various extinct related groups.*
So if the we reduce the description to its first sentence we have:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
Now, for the sake of argument, let's imagine the *bold *words below represent a best case scenario for a relevant *subset* of the first sentence:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
"A fish is a gill-bearing aquatic animal".
That's nice and short and descriptive and reads like a little sentence. It's arguably the best reduction of the first sentence possible.
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence. You have to know *what a fish is* to know what parts of the first sentence are *most* important.
In other words, the "best" description is much more qualitative than it is quantifiable.
type of living organism typified by living in water and having gills
The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher < lydia.pintscher@wikimedia.de> wrote:
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of
course
I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of
Wikipedia
articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor
only
the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
My previous reply was partial and accidentally sent - here's my actual reply :)
On Sun, Mar 22, 2015 at 1:53 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now), which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
They are also used in "*Recent*" and "*Nearby*" and Vibha wants them in "*Saved Pages*" list as well.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description? This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article.
Why not query it to provide an automatic description? Because finding the best subset of the first sentence(s) isn't all there is to it.
For example, take the enwiki "Fish" article.
The first couple sentences are these:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. Included in this definition are the living hagfish, lampreys, and cartilaginous and bony fish, as well as various extinct related groups.*
So if the we reduce the description to its first sentence we have:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
Now, for the sake of argument, let's imagine the *bold* words below represent a best case scenario for a relevant subset of the first sentence:
*A fish is* any member of *a* paraphyletic group of organisms that consist of all *gill-bearing aquatic* craniate *animal*s that lack limbs with digits.
So, we have "*A fish is a gill-bearing aquatic animal*", or you could reduce it further to "a *gill-bearing aquatic animal*".
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence - it's almost entirely a matter of qualitative judgement. You have to know what a fish is to know what parts of the first sentence are most important and then you have to know how to contextually stitch these words together according to rules of the language's grammar and syntax so they "read" nicely (see the word "a" and the "s" on the end of "animal*s*").
Basically, great descriptions require a native speaker of the language with some skill at summarizing. This is such a low bar for humans that almost anyone could contribute quality descriptions.
But, If descriptions are not human editable, then we are stuck with the limitations of whatever heuristics are used to auto-generate the description.
The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher < lydia.pintscher@wikimedia.de> wrote:
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of
course
I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of
Wikipedia
articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor
only
the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Hi Monte! inline:
Deeply hard, in fact, because it's complicated not only by language
syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc).
This already complicated situation then becomes many orders of magnitude
more difficult because these qualitative factors can differ between languages.
Again, I agree that this is not an easy problem. However, in the case of language translations, automated descriptions have the potential of simplifying things tremendously. The algorithm for the grammar and syntax of a certain language needs to be written only once. And once it's written, it can be applied to every Wikidata item, past and future. Sure, there would likely be a different algorithm for each language, and maybe even different algorithms for various taxa of Wikidata items. But this kind of solution simply feels more scalable, and I'm surprised that researching methods of accomplishing this are of little interest.
I predict this won't be any worse than what happened when we enabled
section editing.
But when we enabled section editing, did we do it with a prominent call to action? I just feel a little hesitation about going full-on with something like this, without having a baseline level of administrative feedback in the apps (e.g. a notification for when a description is reverted, and the reason for it).
To be clear, of course I'm totally on board for experimenting with allowing users to contribute descriptions. Making bold moves is what makes our team so great. My goal is simply to point out various other solutions that, to me, make slightly more sense (and to welcome feedback on why they don't!).
But reducing the first sentence in this way is deceptively complicated to
do programmatically, precisely because of the word "arguably" in the preceding sentence - it's almost entirely a matter of qualitative judgement. You have to know what a fish is to know what parts of the first sentence are most important
That's almost convincing :) but still... why duplicate content when the essential information is already there? Maybe I didn't convey my idea of "markup" for extracting a description properly. For example, the description for the [[Fish]] article can be marked up as follows:
A fish is any member of a paraphyletic group of organisms that consist of all *<description>**gill-bearing aquatic craniate animal**</description>*s that lack limbs with digits.
The above markup would be done by a human editor, with the knowledge that the text within the <description> tag will end up as the Wikidata description. I would wager that a similar scheme could be applied to any number of articles. Let's try it for a few random articles:
[[Poland]] Poland (Polish: Polska; pronounced [ˈpɔlska] ( listen)), officially the Republic of Poland (Polish: Rzeczpospolita Polska; pronounced [ʐɛt͡ʂpɔˈspɔʎit̪a ˈpɔlska] ( listen)), is a *<description>**country in Central Europe**</description>* bordered by Germany to the west; the Czech Republic and Slovakia to the south...
[[Schadenfreude]] Schadenfreude (/ˈʃɑːdənfrɔɪdə/; German: [ˈʃaːdn̩ˌfʀɔɪ̯də] ( listen)) is *<description>**pleasure derived from the misfortunes of others* *</description>*.[1] This word is taken from German...
[[Ming dynasty]] The Ming dynasty, also Empire of the Great Ming, was the *<description>**ruling dynasty of China for 276 years (1368–1644)**</description>* following the collapse of the Mongol-led Yuan dynasty...
[[Homomorphism]] In abstract algebra, a homomorphism is a *<description>**structure-preserving map between two algebraic structures**</description>* (such as groups, rings, or vector spaces)...
^^ What would be the downside(s) of doing something like that?
On Sun, Mar 22, 2015 at 9:37 PM, Monte Hurd mhurd@wikimedia.org wrote:
My previous reply was partial and accidentally sent - here's my actual reply :)
On Sun, Mar 22, 2015 at 1:53 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now), which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
They are also used in "*Recent*" and "*Nearby*" and Vibha wants them in "*Saved Pages*" list as well.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description? This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article.
Why not query it to provide an automatic description? Because finding the best subset of the first sentence(s) isn't all there is to it.
For example, take the enwiki "Fish" article.
The first couple sentences are these:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. Included in this definition are the living hagfish, lampreys, and cartilaginous and bony fish, as well as various extinct related groups.*
So if the we reduce the description to its first sentence we have:
*A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. *
Now, for the sake of argument, let's imagine the *bold* words below represent a best case scenario for a relevant subset of the first sentence:
*A fish is* any member of *a* paraphyletic group of organisms that consist of all *gill-bearing aquatic* craniate *animal*s that lack limbs with digits.
So, we have "*A fish is a gill-bearing aquatic animal*", or you could reduce it further to "a *gill-bearing aquatic animal*".
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence - it's almost entirely a matter of qualitative judgement. You have to know what a fish is to know what parts of the first sentence are most important and then you have to know how to contextually stitch these words together according to rules of the language's grammar and syntax so they "read" nicely (see the word "a" and the "s" on the end of "animal*s*").
Basically, great descriptions require a native speaker of the language with some skill at summarizing. This is such a low bar for humans that almost anyone could contribute quality descriptions.
But, If descriptions are not human editable, then we are stuck with the limitations of whatever heuristics are used to auto-generate the description.
The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher < lydia.pintscher@wikimedia.de> wrote:
On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of
course
I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions
leans
more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of
Wikipedia
articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would
instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor
only
the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
An explicit "description" tag eliminates the heuristic problem, but it has other problems I think.
It is markup reliant which raises the contributory bar and complicates any description editing UX.
That is, when a user taps the edit pencil to the right of the description, instead of showing just the description in a simple editable text box with a small prompt to "Enter a concise description of 'article title'", you'd have to show the first section wikitext and explain the description markup.
It also conflates two concerns, that of a concise description and some sub-portion of the first section text. I can appreciate the desire to write descriptive information only once, but this comes at a cost - changes to improve the quality of the description would have to also be proofed to ensure the changes also work in the sub-portion context.
On Mar 22, 2015, at 7:28 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Monte! inline:
Deeply hard, in fact, because it's complicated not only by language syntax and grammatical rules, but also by qualitative factors (readability, meaning, context, relevance etc). This already complicated situation then becomes many orders of magnitude more difficult because these qualitative factors can differ between languages.
Again, I agree that this is not an easy problem. However, in the case of language translations, automated descriptions have the potential of simplifying things tremendously. The algorithm for the grammar and syntax of a certain language needs to be written only once. And once it's written, it can be applied to every Wikidata item, past and future. Sure, there would likely be a different algorithm for each language, and maybe even different algorithms for various taxa of Wikidata items. But this kind of solution simply feels more scalable, and I'm surprised that researching methods of accomplishing this are of little interest.
I predict this won't be any worse than what happened when we enabled section editing.
But when we enabled section editing, did we do it with a prominent call to action? I just feel a little hesitation about going full-on with something like this, without having a baseline level of administrative feedback in the apps (e.g. a notification for when a description is reverted, and the reason for it).
To be clear, of course I'm totally on board for experimenting with allowing users to contribute descriptions. Making bold moves is what makes our team so great. My goal is simply to point out various other solutions that, to me, make slightly more sense (and to welcome feedback on why they don't!).
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence - it's almost entirely a matter of qualitative judgement. You have to know what a fish is to know what parts of the first sentence are most important
That's almost convincing :) but still... why duplicate content when the essential information is already there? Maybe I didn't convey my idea of "markup" for extracting a description properly. For example, the description for the [[Fish]] article can be marked up as follows:
A fish is any member of a paraphyletic group of organisms that consist of all <description>gill-bearing aquatic craniate animal</description>s that lack limbs with digits.
The above markup would be done by a human editor, with the knowledge that the text within the <description> tag will end up as the Wikidata description. I would wager that a similar scheme could be applied to any number of articles. Let's try it for a few random articles:
[[Poland]] Poland (Polish: Polska; pronounced [ˈpɔlska] ( listen)), officially the Republic of Poland (Polish: Rzeczpospolita Polska; pronounced [ʐɛt͡ʂpɔˈspɔʎit̪a ˈpɔlska] ( listen)), is a <description>country in Central Europe</description> bordered by Germany to the west; the Czech Republic and Slovakia to the south...
[[Schadenfreude]] Schadenfreude (/ˈʃɑːdənfrɔɪdə/; German: [ˈʃaːdn̩ˌfʀɔɪ̯də] ( listen)) is <description>pleasure derived from the misfortunes of others</description>.[1] This word is taken from German...
[[Ming dynasty]] The Ming dynasty, also Empire of the Great Ming, was the <description>ruling dynasty of China for 276 years (1368–1644)</description> following the collapse of the Mongol-led Yuan dynasty...
[[Homomorphism]] In abstract algebra, a homomorphism is a <description>structure-preserving map between two algebraic structures</description> (such as groups, rings, or vector spaces)...
^^ What would be the downside(s) of doing something like that?
On Sun, Mar 22, 2015 at 9:37 PM, Monte Hurd mhurd@wikimedia.org wrote: My previous reply was partial and accidentally sent - here's my actual reply :)
On Sun, Mar 22, 2015 at 1:53 PM, Dmitry Brant dbrant@wikimedia.org wrote: Hi Lydia,
Indeed, there are many more Wikidata items than Wikipedia articles. However, the users of our mobile apps only see Wikipedia articles in our search results (at least for now), which means that they will only be able to contribute descriptions to Wikidata items for which a Wikipedia article exists.
They are also used in "Recent" and "Nearby" and Vibha wants them in "Saved Pages" list as well.
No doubt, the description field is an important component of each Wikidata entry. But, when there is a corresponding Wikipedia article, why not query it to provide an automatic description? This could be based on the first sentence of the article, or a subset of the first sentence, or some other kind of metadata within the article.
Why not query it to provide an automatic description? Because finding the best subset of the first sentence(s) isn't all there is to it.
For example, take the enwiki "Fish" article.
The first couple sentences are these:
A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits. Included in this definition are the living hagfish, lampreys, and cartilaginous and bony fish, as well as various extinct related groups.
So if the we reduce the description to its first sentence we have:
A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits.
Now, for the sake of argument, let's imagine the bold words below represent a best case scenario for a relevant subset of the first sentence:
A fish is any member of a paraphyletic group of organisms that consist of all gill-bearing aquatic craniate animals that lack limbs with digits.
So, we have "A fish is a gill-bearing aquatic animal", or you could reduce it further to "a gill-bearing aquatic animal".
But reducing the first sentence in this way is deceptively complicated to do programmatically, precisely because of the word "arguably" in the preceding sentence - it's almost entirely a matter of qualitative judgement. You have to know what a fish is to know what parts of the first sentence are most important and then you have to know how to contextually stitch these words together according to rules of the language's grammar and syntax so they "read" nicely (see the word "a" and the "s" on the end of "animals").
Basically, great descriptions require a native speaker of the language with some skill at summarizing. This is such a low bar for humans that almost anyone could contribute quality descriptions.
But, If descriptions are not human editable, then we are stuck with the limitations of whatever heuristics are used to auto-generate the description.
The key is that the description would stay with the article, which would eliminate the need for duplication and synchronization.
So, in a sense, I would look at it the other way: descriptions within Wikipedia articles would be useful for Wikidata entries.
-Dmitry
On Sun, Mar 22, 2015 at 4:17 PM, Lydia Pintscher lydia.pintscher@wikimedia.de wrote: On Sun, Mar 22, 2015 at 9:10 PM, Dmitry Brant dbrant@wikimedia.org wrote:
Hi Jane,
Perhaps my comments came off as more pessimistic than I intended. Of course I believe in the power of crowdsourcing, and I would never want to make anyone feel like their contributions are being marginalized.
I'll agree for now that the idea of "fully" automated descriptions leans more towards science fiction than reality. :)
However, my whole point has more to do with the apparent duplication of content that seems to be happening between the first sentence of Wikipedia articles and the corresponding Wikidata description. There's something about it that seems unnecessary. If we can figure out a way to automatically extract the description from the first sentence of the article, it would simplify things in two ways:
- People wouldn't need to edit Wikidata descriptions, and would instead
focus on improving the Wikipedia article. 2) People who monitor changes made to articles would need to monitor only the article, instead of the article plus its corresponding Wikidata description.
There are a lot more items on Wikidata than articles on Wikipedia. And not every language has a Wikipedia article for each item. Don't just look at descriptions on Wikidata as something useful for Wikipedia. They're much more than that.
Cheers Lydia
-- Lydia Pintscher - http://about.me/lydia.pintscher Product Manager for Wikidata
Wikimedia Deutschland e.V. Tempelhofer Ufer 23-24 10963 Berlin www.wikimedia.de
Wikimedia Deutschland - Gesellschaft zur Förderung Freien Wissens e. V.
Eingetragen im Vereinsregister des Amtsgerichts Berlin-Charlottenburg unter der Nummer 23855 Nz. Als gemeinnützig anerkannt durch das Finanzamt für Körperschaften I Berlin, Steuernummer 27/681/51985.
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
Mobile-l mailing list Mobile-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mobile-l
On Mar 22, 2015 11:36 PM, "Monte Hurd" mhurd@wikimedia.org wrote:
It is markup reliant which raises the contributory bar and complicates
any description editing UX.
On this point I can imagine a simple "highlight description text and click an associated button" UX that would eliminate the need for showing source markup to those who don't want it.
Thanks, Jeff Hobson
Follow-up to my earlier comments:
What do you mean by this:
The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.
How is the description any less usable than any other field?
On Mar 22, 2015, at 8:57 AM, Dmitry Brant dbrant@wikimedia.org wrote:
The "description" field, on the other hand, is the only thing that is *not* data, and is not usable by a machine in any way.