7 ms·

It's actually interesting that frameworks provided by platforms such as .NET, Java, etc., don't include an abstraction for the representation of names. Such ab

It's actually interesting that frameworks provided by platforms such as .NET, Java, etc., don't include an abstraction for the representation of names. Such abstractions exist for dates, times, calendars, currencies, calculations with money, and so forth, but not names. On the one hand I can understand it, because names are so complicated, and how would you sit down and come up with something good enough to represent all of them? On the other hand they're prevalent in such a high percentage of line of business and consumer facing apps that it's almost ridiculous that every single developer on the face of the earth at one time or another has to come up with their own half-baked implementation. It's especially ridiculous when you consider that so many of these home-rolled implementations, if not all of them, are rife with terrible flaws that constantly cause frustration and inconvenience to a small but significant number of users.

maxander 7y ago
A dedicated, well-maintained name abstraction is certainly something that needs to happen. More than interesting, it's a bit bizarre that this hasn't been done yet (AFAIK.) In terms of developer-facing complexity, this could be a laughably simple thing to use- just a type that supports equality, perhaps ordering, and conversion to string. Only the constructor would need to be complex. :) I guess the reason this hasn't been done is simply that the implementation would never be "correct"- there is no formal specification of human names out there, and there would always be cases where some poor individual with an unusual name falls afoul of the system. Strictly fewer cases than we have today, where everyone rolls their own name system, but still some; it's not a solvable problem the way timezone conversion is. But, on the other hand, that's the way the world is- messy. Developers are going to have to learn better best practices for communally doing our best in cases where there is no perfect answer, because there's only going to be more of such cases as tech continues to eat the world.
- sargram01 7y ago
  I’m pretty sure it’s solvable, the main problem is that we break up names into first and last to identify the parts and we do bad data quality checks. Let’s say we did just have one field and a service that was trained on each counties’s variations that could return the parts of the name you wanted. So some database and detection system to understand the pattern. It’s definitely possible since we humans do read and understand names just fine in our own locales.
  - smegger001 7y ago
    It would help if he quit using the First|Middle|Last| terminology an used Surname(if-any)|Personal-name #N|Personal-name #N+1(ifany)...
    - b3orn 7y ago
      It would help if we quit using that and had one field for the full name and then another one for "how do you want to be addressed".
      - deathanatos 7y ago
        I've in the past written stuff that generated an index of names; this was sorted. While you can certainly sort on a free-form text field, the culture here is that name indexes are generally sorted by family name. So, to do that, you have to have some understanding of what the family name is, which a free-form text field does not give you. But a lot of software has no need for the breakdown, and would be better served by a free-form field.
        vidarh 7y ago
        Then at least be upfront with the user about why you're asking, because they might well answer differently (e.g. include a different number of parts of their last name) if they know your purpose is to sort the name than if they think it's being used for a different purpose. They might even give totally different names. That's the most important part of the comment above: The concept of a name is so overloaded that unless you ask about the string to use for the specific purposes you intend to use it, then there is very little you can do with it.
  - vidarh 7y ago
    I'm not convinced it is solvable. I don't think the general case is reliably solved even by humans. E.g. is Carlos the same person as Karl? Well, that depends. Was one of them localized, or are these the actual given names. Just this weekend I was on an offsite and saw a Spanish book about Karl Marx, or Carlos Marx as they had written it in the title, in the library of the house we stayed in. Clearly in this instance the names are the same, but that requires knowing that Carlos Marx maps to Karl Marx and that Karl Marx is a famous name; otherwise you can't assume the name was translated. There were many other names on books in that library. I don't know which one of them - if any - also maps to someone known under another name, because that requires me to know which person they are about. Is Curt, Curth, Kurt the same person? My uncle had all three on different documents, and delighted in telling people about it. What about countries like the UK, where there is no legal requirement to notify anyone of a change of name, and where a the legal way of formally changing your name - a "deed poll" is just a document structured in a certain way where you assert that you are known under a certain name? My ex is known under at least three different name combinations, all of which are present on different sets of legal documents. Some subsets of the issue is solvable, but for example there is no way of taking a full name and returning the "name this person prefers to be known by" because the name does not contain that information. You can make a pretty good guess. But you'll fail dramatically for people from different countries. And don't think for a second you can guess correctly based on where a name is from - many names are used in different countries, and often as different elements (e.g. firstname one country, lastname another; feminine name one place, masculine another), and many people have names that combine different nationalities (e.g. my son has a name that combined an English firstname, a Nigerian middle name and a Norwegian last name). The only reliable solution is to not assume any one single string can be used as a generic name - you need to ask what to use within a given context and within whatever constraints you have.
- rectangletangle 7y ago
  I actually attempted to do this. I used ML techniques to help smooth over some of the difficult parts (there are many difficult parts). The hardest cases are ambiguous names, for instance delineating Hispanic vs. Puerto Rican naming conventions (they're different). The fundamental approach involved pushing all ambiguity up to the end user, so they always have the option to correct the system. https://www.alphanym.com/demo/?jm2 https://www.alphanym.com/demo/?jm2
dkersten 7y ago
We already have an abstraction for names: a text field. Trying to be more clever than that will break for someones name.
- hirsin 7y ago
  Which is great until the client asks for a friendly name (given name) identifier. GitHub uses name, we use first/last name. So we just shove the GitHub name in the first name spot and ask people to organize what looks right. Our partners suggested string.split(' '), which produced interesting results against the sample list of github users.
  - clarry 7y ago
    A lot of software would be much better without stupid clients asking for the wrong things.
    - hirsin 7y ago
      Eh, I understand it. They want to put some name in the identity control in the header. Putting the full name in is guaranteed to go wrong. We might start asking for a nickname for that purpose.
      - gbear605 7y ago
        Why is the full name guaranteed to go wrong? A better assumption is that there is no reasonable nickname and you have to use their full name.
  - dkersten 7y ago
    Use two fields: name and display name. Anything else will break for someone. In most cases where something asks for my name its really not even necessary and certainly not necessary to split it into first/last/display/friendly etc.
    - 7y ago
wbl 7y ago
It's a string. Dont even try anything else.
- axilmar 7y ago
  Exactly this, and I really don't see what the fuss is all about in the above comments.
legulere 7y ago
Because it’s extremely diverse between cultures and countries how names work. Here in Germany it’s typical to have several first names and it’s legal to use any of it, even though it might be just the name of your godfather/godmother. FHIR has a relatively general definition for names, but multiple general and country-specific extensions exist for it: https://www.hl7.org/fhir/datatypes.html#HumanName https://www.hl7.org/fhir/datatypes.html#HumanName
nradov 7y ago
This is a solved problem from a modeling standpoint. The HL7 Reference Information Model allows any entity (such as a person) to have multiple names. Each name can be tagged with a type (legal, maiden, alias, etc) and validity date range. A name can contain multiple parts in any order, optionally tagged as prefix / suffix / family / given. Names can also be explicitly marked as null if unknown or not assigned. There are open source RIM implementations in several languages.
- deathanatos 7y ago
  Interesting; I didn't think I'd see an HL7 reference in this thread. I work somewhat with FHIR, which also has a HumanName[1] data type, and I think it handles most of the cases in this thread. For those not familiar: FHIR is a standard that covers health and patient data. IMO, it's a pretty good model. (HL7 is the organization, and there are few other standards under it.) I'm less familiar with RIM; could you link to it's definition of a name? (The best I could find suggested that it was nothing more than an unconstrained piece of text.) [1]: https://www.hl7.org/fhir/datatypes.html#humanname https://www.hl7.org/fhir/datatypes.html#humanname
  - a-priori 7y ago
    What's the purpose of structured modelling of names? Why does it matter which parts are family vs given vs whatever else? Lots of the W3C International Examples they give use a `<text>` field. Why not just use that? https://www.hl7.org/fhir/datatypes-examples.html#2.24.1.13.1 https://www.hl7.org/fhir/datatypes-examples.html#2.24.1.13.1
    - nradov 7y ago
      It matters for collation order, if you want to show a list of patients sorted by family name.
      - Piskvorrr 7y ago
        Is this a current use case, or did that just make sense 100 years ago? (As in these assumptions: "tracking actual relationships in data is Hard, and family name correlates strongly with real-world relationships") I mean, it sort of works even today, but neither of the assumptions are as strong any more.

maxander 7y ago

A dedicated, well-maintained name abstraction is certainly something that needs to happen. More than interesting, it's a bit bizarre that this hasn't been done yet (AFAIK.) In terms of developer-facing complexity, this could be a laughably simple thing to use- just a type that supports equality, perhaps ordering, and conversion to string. Only the constructor would need to be complex. :) I guess the reason this hasn't been done is simply that the implementation would never be "correct"- there is no formal specification of human names out there, and there would always be cases where some poor individual with an unusual name falls afoul of the system. Strictly fewer cases than we have today, where everyone rolls their own name system, but still some; it's not a solvable problem the way timezone conversion is. But, on the other hand, that's the way the world is- messy. Developers are going to have to learn better best practices for communally doing our best in cases where there is no perfect answer, because there's only going to be more of such cases as tech continues to eat the world.

sargram01 7y ago
I’m pretty sure it’s solvable, the main problem is that we break up names into first and last to identify the parts and we do bad data quality checks. Let’s say we did just have one field and a service that was trained on each counties’s variations that could return the parts of the name you wanted. So some database and detection system to understand the pattern. It’s definitely possible since we humans do read and understand names just fine in our own locales.
- smegger001 7y ago
  It would help if he quit using the First|Middle|Last| terminology an used Surname(if-any)|Personal-name #N|Personal-name #N+1(ifany)...
  - b3orn 7y ago
    It would help if we quit using that and had one field for the full name and then another one for "how do you want to be addressed".
    - deathanatos 7y ago
      I've in the past written stuff that generated an index of names; this was sorted. While you can certainly sort on a free-form text field, the culture here is that name indexes are generally sorted by family name. So, to do that, you have to have some understanding of what the family name is, which a free-form text field does not give you. But a lot of software has no need for the breakdown, and would be better served by a free-form field.
      - vidarh 7y ago
        Then at least be upfront with the user about why you're asking, because they might well answer differently (e.g. include a different number of parts of their last name) if they know your purpose is to sort the name than if they think it's being used for a different purpose. They might even give totally different names. That's the most important part of the comment above: The concept of a name is so overloaded that unless you ask about the string to use for the specific purposes you intend to use it, then there is very little you can do with it.
- vidarh 7y ago
  I'm not convinced it is solvable. I don't think the general case is reliably solved even by humans. E.g. is Carlos the same person as Karl? Well, that depends. Was one of them localized, or are these the actual given names. Just this weekend I was on an offsite and saw a Spanish book about Karl Marx, or Carlos Marx as they had written it in the title, in the library of the house we stayed in. Clearly in this instance the names are the same, but that requires knowing that Carlos Marx maps to Karl Marx and that Karl Marx is a famous name; otherwise you can't assume the name was translated. There were many other names on books in that library. I don't know which one of them - if any - also maps to someone known under another name, because that requires me to know which person they are about. Is Curt, Curth, Kurt the same person? My uncle had all three on different documents, and delighted in telling people about it. What about countries like the UK, where there is no legal requirement to notify anyone of a change of name, and where a the legal way of formally changing your name - a "deed poll" is just a document structured in a certain way where you assert that you are known under a certain name? My ex is known under at least three different name combinations, all of which are present on different sets of legal documents. Some subsets of the issue is solvable, but for example there is no way of taking a full name and returning the "name this person prefers to be known by" because the name does not contain that information. You can make a pretty good guess. But you'll fail dramatically for people from different countries. And don't think for a second you can guess correctly based on where a name is from - many names are used in different countries, and often as different elements (e.g. firstname one country, lastname another; feminine name one place, masculine another), and many people have names that combine different nationalities (e.g. my son has a name that combined an English firstname, a Nigerian middle name and a Norwegian last name). The only reliable solution is to not assume any one single string can be used as a generic name - you need to ask what to use within a given context and within whatever constraints you have.
rectangletangle 7y ago
I actually attempted to do this. I used ML techniques to help smooth over some of the difficult parts (there are many difficult parts). The hardest cases are ambiguous names, for instance delineating Hispanic vs. Puerto Rican naming conventions (they're different). The fundamental approach involved pushing all ambiguity up to the end user, so they always have the option to correct the system. https://www.alphanym.com/demo/?jm2 https://www.alphanym.com/demo/?jm2

dkersten 7y ago

We already have an abstraction for names: a text field. Trying to be more clever than that will break for someones name.

hirsin 7y ago
Which is great until the client asks for a friendly name (given name) identifier. GitHub uses name, we use first/last name. So we just shove the GitHub name in the first name spot and ask people to organize what looks right. Our partners suggested string.split(' '), which produced interesting results against the sample list of github users.
- clarry 7y ago
  A lot of software would be much better without stupid clients asking for the wrong things.
  - hirsin 7y ago
    Eh, I understand it. They want to put some name in the identity control in the header. Putting the full name in is guaranteed to go wrong. We might start asking for a nickname for that purpose.
    - gbear605 7y ago
      Why is the full name guaranteed to go wrong? A better assumption is that there is no reasonable nickname and you have to use their full name.
- dkersten 7y ago
  Use two fields: name and display name. Anything else will break for someone. In most cases where something asks for my name its really not even necessary and certainly not necessary to split it into first/last/display/friendly etc.
  - 7y ago

wbl 7y ago

It's a string. Dont even try anything else.

axilmar 7y ago
Exactly this, and I really don't see what the fuss is all about in the above comments.

legulere 7y ago

Because it’s extremely diverse between cultures and countries how names work. Here in Germany it’s typical to have several first names and it’s legal to use any of it, even though it might be just the name of your godfather/godmother. FHIR has a relatively general definition for names, but multiple general and country-specific extensions exist for it: https://www.hl7.org/fhir/datatypes.html#HumanName https://www.hl7.org/fhir/datatypes.html#HumanName

nradov 7y ago

This is a solved problem from a modeling standpoint. The HL7 Reference Information Model allows any entity (such as a person) to have multiple names. Each name can be tagged with a type (legal, maiden, alias, etc) and validity date range. A name can contain multiple parts in any order, optionally tagged as prefix / suffix / family / given. Names can also be explicitly marked as null if unknown or not assigned. There are open source RIM implementations in several languages.

deathanatos 7y ago
Interesting; I didn't think I'd see an HL7 reference in this thread. I work somewhat with FHIR, which also has a HumanName[1] data type, and I think it handles most of the cases in this thread. For those not familiar: FHIR is a standard that covers health and patient data. IMO, it's a pretty good model. (HL7 is the organization, and there are few other standards under it.) I'm less familiar with RIM; could you link to it's definition of a name? (The best I could find suggested that it was nothing more than an unconstrained piece of text.) [1]: https://www.hl7.org/fhir/datatypes.html#humanname https://www.hl7.org/fhir/datatypes.html#humanname
- a-priori 7y ago
  What's the purpose of structured modelling of names? Why does it matter which parts are family vs given vs whatever else? Lots of the W3C International Examples they give use a `<text>` field. Why not just use that? https://www.hl7.org/fhir/datatypes-examples.html#2.24.1.13.1 https://www.hl7.org/fhir/datatypes-examples.html#2.24.1.13.1
  - nradov 7y ago
    It matters for collation order, if you want to show a list of patients sorted by family name.
    - Piskvorrr 7y ago
      Is this a current use case, or did that just make sense 100 years ago? (As in these assumptions: "tracking actual relationships in data is Hard, and family name correlates strongly with real-world relationships") I mean, it sort of works even today, but neither of the assumptions are as strong any more.