Live CEOing Ep 764: Design Review of Word and Language Data

Okay. Hello, everyone. We are here to do a Wolfram language design review about language. Human language. That is so. Or maybe it's also of nafa. Should we start off with phonetic symbols?

I think that's probably a good place. This was the thing that we, we had talked about. We didn't have anything really to look at last time. Christian has put in a very, very preliminary packlet with some properties. I think it's. It's just going to be more helpful for us to kind of look at this early on and figure out what we want to do with it. Think about naming of things and how they're going to be useful. I mean, this sort of alpha forward. But we also are going to want these things in the language.

Okay. So, I mean, obviously that shouldn't be capitalized, but.

Right. Yeah, that's lots of design things and capitalization of values that we need to fix. But there's some data.

So I thought that with Fanatic. I mean, there are all these things about the classification of the thing. I guess that's. And then there's also, you know, what the mouth shape is for that and so on.

Right. Manner and place of articulation and those kinds of things.

Yeah, right. And then. But I think, I mean, this entity is something that, for example, from a. We're supposed to have a classifier that will go from a, you know, a waveform to break it into phonemes. Right. That was one of the things that was supposed to be coming from the audio team.

Right.

And so then one of the questions.

Is, I mean, my assumption is what that will output is just a string of IPA symbols. Right.

I think so. I think it's basically like, just like we will have speech. We have speech recognized that gives us words. This will be. I mean, perhaps we should put it into speech recognize with a second argument or something that says phonetic or something like that. That just gives the phonemes.

Yeah.

Then there's the question of can we go from the phonemes to the, to the words? Right. And. Yeah, well, you know, or maybe we need a thing that is something like, you know, phonetic. I don't know whether phonetic form is quite the right thing. That could take, you know, elephant and that. I mean, and could either take that or it could take audio. But either way what it's doing is. I mean, I don't know if I say word, data, elephant and then I say.

Right. I mean, we have a. I can't remember. Yeah.

Okay, so is that.

But then that would be, I mean, to go phonetic form of A string is going to be dependent on us having having a stored IPA representation of a word in a given language. Or are you imagining.

Okay, right, but this is what, what should happen is if we say speech synthesize of elephant, Then. Okay, now we say now I'm imagining.

Does that actually do. Right now I'm just realizing I don't know what it is. Is this beep, synthesize, go from anything we've got stored or is that. That's machine learning stuff to do.

That's machine learning. It has nothing to do with the fanatic.

All right.

Oh, dear. That's unfortunate. We should give that to the family. Machine learning and audio teams. It's a good test. And qa, it's a good test to do the round trip for speech and speech recognize. I mean, it's very hard, but we can imagine taking that. That speech form and saying phonetic form of that. Or phonetic. You know, maybe I don't know if it's speech recognized. You know, I don't know whether speech recognized should take a second argument which says. Yeah, I think it could just literally take a second argument that just says phonetic. Maybe that's the right thing. But I'm not quite sure what this should do. I'm not sure. But something.

Why would that. Yeah, why. Why is that more uncertain than the phonetic form of a string?

Where are we putting this?

I don't think you had saved a notebook last time. Do we have a word data? Could.

Well, let's call this word.

I mean, you had, you had a word data down below, but yeah, that works.

Okay. What were you saying?

I was saying why. Why is phonetic form of audio.

Less.

Certain for you than phonetic form of a string?

Oh, because this. I, I like the recognized word because it's much less clear that you can do it.

Oh, I got it. Okay. You mean just the naming of the.

Okay, yeah, got it. Okay. But. But then once we have this, you know, I don't even know. I don't know what that, what that means. I don't know what, you know, for each of these letters. I mean, it's sort of interesting because what we want to do here, if I say characters 72. Okay. We. We have, you know, a two entity concept.

Right. Which right now is just characters. You think that could be going to the. Recognizing those as phonetic symbols and giving you things that would have the phonetic property?

I think that would be the first.

Yeah.

Interpretation for the string.

Yeah.

Except that here, that just says Latin, small letter F. So that's super confusing. So in the phonetic form here, surely Unicode has in it all the phonetic. You know, that's just an F. It's not a. It's not some kind of phonetic F.

Is. I mean, I don't know if I don't recall offhand, is there a specific F to ipa or does IPA use, I don't know, Latin characters? Does anybody. Francisco or Christian. Do you guys know offhand?

I mean, if you see the example in the scanner down below, you can see that, for example, that symbol is what we would call like an I, and it's a close from unrouted vowel. So I believe there are some characters that will be represented as vowels as, you know, letters.

No, look at that. That's not right. Look at that. See that? That's the Unicode for a standard communal garden I. So I'm saying that's not a special I in some kind of special part of Unicode space.

Right? That's. That's what I'm thinking. That. That IPA is some regular Latin symbols, plus, like the, you know, the schwa symbol. It's. That's used for other things. It's not unique to ipa.

Yes, well, so, okay, so that's telling us that arguably even this phonetic form isn't quite right, although we probably used it in other places. So we have. We'll need to have a different name for that. We'll need something which is the true IPA string here. Which. Okay. You know, it's making me think that we need some wrapper. I don't know. I don't know if this is right. I mean, you know that. That says. Oh, I don't know. I mean. Okay, so what functions are we going to have? We're going to have.

I mean, we. When you represent something phonetically, you usually have the string. You usually have it between backslashes or slashes or in brackets. So if the string had that, we could say that string wrapper. You could try to recognize those and those characters as phonetic symbols.

Okay. I mean, so that's a good point. Yeah. What's that? Yeah, it's a good point.

Yeah.

I mean, so you're saying. I mean, we could.

What.

What do you think the Unicode is for that Backslash? Maybe there's a special Unicode for that, and we could just literally parse the thing into that.

I'm not even sure it's the special kind of backslash. I think it's just that the, The. That the.

Yeah, I see a se.

A sequence of characters that happen to be in the family of things that are used in IPA are Presented between two backslashes. Yeah, I believe it's sequence of things that. That together says. Oh, that's. That's pronunciation.

Yeah, it's just a convention, basically.

I understand that. But if we wanted to, we could end up with some actual character that we've invented that is, you know, some. Is that a raw backslash? I don't know. What is that? No, that's just an ordinary backslash. Okay. I mean. Okay, what is one going to be using this for? Okay, Let's see. What are we doing here? I mean, the question. Okay, we want to be able to go. I mean, what operations do we want. Think we want to do on fanatic strings? You know, we might want to do. I mean, let's. Let's make a list of some operations. Okay. So we want to go audio to phonetic, fanatic to audio. We might want to do things like language or accent recognize. We might want to create. Create diagrams for how to say a word. So I'm saying. Or, you know, create a video. Okay.

What are you talking about? Using the thing like the. The side view of the whole.

Yeah, yeah, right.

Language apparatus.

Yeah, yeah. I'm trying to remember. I mean, if I pick a book about phonetics off my shelf and I have at least one. I mean, what's going to be in it? I'm going to do the experiment. Just a second. It shows that the stuff about language is actually active in my world because I have these books nearby. I have three of them here. Practical Introduction to Phonetics. Okay, so cardinal vowels. So that. They're diagrams of. Yeah, there are all kinds of stuff about sort of. It's. It's a lot about kind of how the translation between the physicality of air, motion and things and. Oh, and there are things about transitions from one vowel, you know, from one phoneme to another. Those types of things. First and second formants of the cardinal vowels. Yeah. Okay. And then there's. There's all the tonal stuff for, you know, languages like Chinese. All right. I have a thicker book here. It says more because it's thicker. A lot of stuff about. Yeah, there's a lot of stuff here about waveforms. I see things with different vowel systems for different. Okay, so there's the other one is the mapping from language to phonemes. That's another. That's another type of thing.

I mean, it's also intersects. I know you were. You were looking for vowels and consonants over the weekend. I wasn't sure context of that was. But that brought up some of the old. We're talking about English, that's pretty straightforward. As soon as we get away from English, it can get more complicated. But we hadn't really been the last time we looked at it, seriously. We weren't really thinking about phonemes and thinking about vowels, not as just, you know, this collection of written characters in the language that represent these things, but it sounds. And so that's. That's something that I think we can address here.

Right. Oh, there's rhyming stuff and so on. Yeah. Okay. All right. Well, I mean, I think. I think the most basic thing is just, you know, getting. I don't know whether. Right. I mean, look, we clearly got some distance here. We've got to have entities in the language that represent all the pieces of ipa. We've got to have entity classes for, you know, labels and things like this. All the usual stuff. Right. We got part of the way there 10 years ago. Let's finish it.

So, I mean, at least for now, we're all of the different symbols that mean something in ip. And the other thing we touched on before is that some of. Some of the symbols are diacritics, Right. They're. They're things that need to be paired with some other symbol and they modify it in some way. But in any. But in any case, I don't know the set of. The set of. The set of standard properties about.

Place.

Manner of articulation, backness, height, all these different properties that describe specific sounds sort of. There are, I think. Christian, do you. I don't know if you had looked at this. Are there kind of clear mappings for what phonemes are commonly used or not used in. In particular languages? Do we have that mapping between phonemes?

We certainly can get that. That's a known thing.

Can get that, I believe, from the specific Wikipedia pages of the. Of each of the symbols, there's a section called Occurrences, if I'm not mistaken.

Well, we can also go to, you know, dictionaries.

Yeah.

Go and figure out. I think it'd be a good thing to do to get the frequencies of different phonemes in certain languages.

Okay.

Because, you know, in English, there are weird phonemes that occur, but very rarely, I think. Okay, what else do we have to look at with this right now? I mean, so. So clearly in this thing, we're going to want to see the stylized. For a vowel, for example, we want to see the stylized formants that this corresponds to. We're going to need to have, certainly in the language, something represented in terms of audio generators that makes that phoneme.

We want to have the kind of. The standard you were talking about, the. Oh, what was it? The cardinal vowels and there's. There's that kind of standard table of vowels. Yeah, we want to be able to generate the.

We need the sort of table, the.

Standard graphic in some way that we can highlight where things sit in relation to other symbols.

Right. I think for wolf alpha, that's exactly what we need to do. We need to be able to show where in that diagram, you know, in the standard diagram, this particular object occurs. That makes sense.

Yeah.

Okay. All right. Should we go on?

Yeah, I think so. Language, language, family. This was.

We had touched.

We mostly had just looked at this classification pod because Christian was updating that last time. The. You had some comments about the layout of that. He's implemented some of those changes. So I think we wanted to just take a look at design and pod. And there's.

In the language. Is it a tree?

Sorry, no, it is not a tree. It's a graph. Because we have trees implemented. It's not yet implemented in 12.

But it should be in. When we make this an entity in the language, it should return a tree. Right?

Oh, okay, sure. That can be done.

Right? What it, what it does right now, implementationally, internal dwarf malfa is not so relevant. I mean, this look a lot better.

Yeah.

It's cool. Oh, boy. We need the.

Is the. Sorry, can you, can you go up. Is the.

The.

The handling of cases when we have many, many, many siblings at a given level is that you should do it.

The way trees do it? I mean, you know, it's a temporary thing that we're doing. Well, this is more or less how trees do it.

Okay.

Yeah, that's how we're doing it. The, the gray blob is basically the elision, the equivalent of an elision blob in the tree.

And fine. And eventually these should just be trees.

Yes.

You should be able to get the tree in the language.

Yes.

SAP.

Yes.

Okay, now if I look at these intermediate languages, Latino Feliskan, do we have properties for that?

All of these are entities, aren't they, Christian? They're either language. These are. Those are families, right?

Yes, those are families and we do have properties for those. But the packlet itself is still on default.

It's.

It's not yet in production.

I see. So if I go Latino feliskun here, I'm not going to win?

No. Okay.

But I will in the future.

Unless you go in def mode. And still we are trying to add not all of the Entities has half spellings yet. And we're trying to add that in PR that's open right now. Pr.

Sorry.

Did I just learn some other languages version of PR Pull request. Okay, the, the. How many languages are there of these, of these sort of intermediate. Because I mean, these are not in the current inventory and are we treating them as different? You know, Western Romance is a historical language, right?

Or no, in these cases they are language families and they are not languages by itself, but rather like grouping of languages, if it would be a way of saying them.

Right. Okay, but those entities, are they like our language entity? I mean, if I say, you know, if I have a language here that's a language entity, but something like Western Romance, what is it going to be? Is it. What, what entity type is it going to be?

Language family.

Okay, do we already have that?

It is in def mode.

That. That's exactly the thing that we're talking about. Oh, I see what you're saying.

What we, what we already.

Language family is for the most part things that previously were entity classes in language data.

Now what the heck are we going to do with that?

The plan right now is. Well, the idea is that when we implement language family, those classes will. Well, would not be classes, but instead would point to the language family entity.

Okay. It seems to make sense to me. Although it still might be useful. Go ahead.

No, no, no. I mean we can, we can preserve but downscore the entity. This is for some time, I think, but it's, it's much more useful to have them as entities because they do.

Have properties, specific properties. But what about the possibility of thinking about them as hierarchical entities? I mean, this is very similar to, well, bylam species etc, isn't it?

No, because there's no, there's no like finite canonical set of levels here. Like there's nothing in that graph on the left that corresponds to kingdom. Right.

So we haven't had this, I mean, for. This is worth bringing up with the hierarchical entity people that, I mean, this is a case we haven't seen before where as you say, we don't have known meanings to different levels on the tree.

Yeah, I mean hierarchical entities work when each, you know, each level is, is going to be a different entity type. And this is not, this is just.

Yeah, I understand.

A different relationship than what that's meant to, to con.

I understand. Although it sort of reminds me. I see. So if, if we're thinking about organization like an army, it's got, you know, I don't know what, you know, divisions, regiments, Whatever. And each one has a troops and level. Each level has a name.

Yeah.

Whereas here you claim that this is a pure sort of containment story without names, without names for the levels.

Yeah. At most you could have in the language family, you could differentiate the ones that are root families and the ones that aren't, but that would be the only hierarchical differentiation that you could make.

All right, okay, I retract my thought that the. So, so I think then this, it's an interesting case. It is worth pointing out to the hierarchical anti folk that there is this case because maybe there's a way that they could handle this. I worry a bit that these names are really obscure, you know, Northwestern Shifted Romance. Sounds like. Okay, all right. But anyway. All right, this looks good. So what else? Let's see. Let's look at language, family properties.

Right. I mean, this is the stuff that Christian was saying. This is still in dev, but just to be able to see right now what's showing up in, in the language. If we want to talk about property naming, how things are represented here.

Well, I mean, this is sort of an interesting concept. I mean, to me, if you can get out the tree, I mean, one of the properties should just be. Something about. Well, I see. So there's a single path from the root to this particular thing and the question is to generate the tree that is kind of a sibling type tree. Boy, I mean, what you really want is that tree and you want to know where this particular language is in that tree.

I mean, you would, you would be able to see that in the full tree property. I believe it's in others. I'm not sure though. Yeah. This ancestry tree, it's only really going to give you a straight line of.

Well, I know, I know, I know, I understand that. Okay, I understand. But I mean, so somewhere I see there are disconnected parts to the tree of all languages because there are different routes for different root families. Yes, I see. So you can see the full tree for the root family. Yeah, that makes sense. I mean, in the end we're going to want something which gives all the root, all those trees, the full.

Is there a property, Christian, that indicates whether something is a root family or not? A class of entities that's root families.

No, it could be easily implemented though. I mean, I believe there is.

I mean, that's what you would want, right? If you would want to be able to get all the root families and show all the trees of. That's what you're saying, I think, Stephen.

Sure. That would be implemented like a root Family classes. Yeah, Root family class. Sorry.

Yes, yes, yes. Okay.

Yeah. I do believe there is like a root family cube property. And so the class would be use. We just. We would just have to use that.

I see. Okay, so this is aggregating the individual languages. So I'm a little bit unclear. If somebody speaks both, you know, English and Welsh, do they get double counted in. I don't know whether English and Welsh come from the same root family. Let's assume they do. Do they get double counted or not? If I say, what's the total number of native speakers of whatever the root family? Indo European. Right. If they speak English and French and you're counting the total number of Indo European speakers, do you correctly count them only once between the English and French?

No, I do believe it would be duplicated. Yes. Or how it's implemented.

Okay. I don't know whether we can do any better than that. I mean, that seems like a bug.

The question is, do. Does it make sense to try to aggregate. If we don't. If we can't.

Know what's happening at.

The individual language level, it might not make sense to do that.

Right. We should best. Best just get the derived languages from a language family. So presumably one of the properties here is languages in that language family. Where is that?

You mean what are the.

What are all the languages.

What are the leaves in. Under. Under that node?

Yeah, you have descendant languages.

Okay, okay, that's.

And that languages are. That's not the families that might be under a given family. It's following the branches all the way out to the end. Those are all the things that are at the end of the branches.

I mean, these names are not obviously bad, but. But they're just. I mean, I don't know what the difference between number of languages and number of descendants is.

The thing is that number of descendants includes also the language families that are under the particular.

Okay, that's confusing because it says descendant languages of which the number of languages is. I mean, this is probably should just have a. Languages.

Yeah, fair enough.

And this is. I mean, something. This is more like tree size or something like that or. I don't know what this is because that's not. What's. Children. Is that the number of children who speak these languages? I don't think so.

What.

What is the difference between. Is children. Children is.

Children is basically.

That's like the. The list of entities that corresponds to number of descendants.

No, children is number of. Number of descendants immediately below. Like as if you were a parent and you have children. Yes, but your descendants are the children of the children. Some.

It doesn't include grandchildren.

No, it just includes at the first level.

Look, we have conventions for this for trees. We should use them. I don't know what they are. Okay. We've got names of functions for this and so on. Let's base this nomenclature on that. Make sense?

Okay, sure.

I think we're not far off from that. No, the children and parent terminology there too.

What is this all about?

These were. This is another big set of properties that are in dev mode that are. That are about language structure. There's all the stuff. The World Almanac of Language Structures. This. This before was just what the output was. Christian has a PR up that's fixing some of this. We. We didn't like the property names. I think he's changed this. Right, Christian? So that you're using more directly the. The way that these things are. Are referred to in walls.

Yes.

And also the. The something plus. The plus as a separator is not really conventional. That's usually more a dash to indicate those sequences.

I don't think it should be plus plus. Seems like it could be in any order. I think it's like it. It's a dash more to indicate. It goes blah to blah to blah.

Right, that's. That's what I'm saying. This is. This is. This is the old version. This is what we didn't like last time. Scroll down. There's a. There's a. Oh, I think a screenshot from the pr. That's right. I think those dashes should be bigger and probably we should condense those. We don't need all the white space.

Yeah, That's nice. Now, things like subject, object, subject, verb, object in the language, presumably that thing, that property will return a list. Subject, verb, object. And then.

That was a question I had is how we want to represent those in the language. Like for most of these things, there's a finite set of orders that exist in all languages. So like is subject, verb, object is a specific thing that applies to a bunch of different languages, should that itself be represented as an entity? I mean, we do have a list of strings.

Right, Sorry, we do. Sorry, I'm sorry. We do have EP tables for this. And in those cases, the value is. These are entities. Right. So the entity in this case would be, I believe, sbo. That's the standard name. And the value would be. At least as you said.

What do you mean by the value as opposed to the entity? I mean the entity.

Give me one second for that.

Well, what I'm. One thing is the thing called subject here better be an entity of some type of.

Yeah. So the entity would be SOV for.

SBO in this case, but Steven's suggesting breaking that, decomposing that more. Right. You're suggesting that we would return a.

List which is grammatical unit, adjective is grammatical, listed as grammatical unit.

What is subject?

So I'm saying. So in that order of adjective and noun, there's two possibilities. Either in word order, it's either SVO or it's the list. Subject entity, verb entity, object entity.

Okay. Yeah, sure. I believe it would be better for it to be the list.

I think it's more useful and more flexible. Yes, is.

Now, would we consider subjects a. A grammatical unit?

I'm not sure. I was wondering the same thing.

I got another question we don't have. I'm just looking at the list of grammatical units we have now, and. We don't. But we have. We have weird things in there. It wouldn't be out of place.

We have.

We have even more specific things like passive, nominal, subject, noun, phrase as an adverbial modifier. So straightforward things like subject, object would not be out of place.

Okay. Noun, parent, plural.

No, I mean, I think that. Do you remember Ben? I want to say that this list was. This was all the things that grammatical structure could return. And so we identified all the things that could show up in.

You mean text structure or.

Text structure. That's it. Yeah.

Yeah, that sounds familiar, but I couldn't say for sure. Yeah. So now that I'm checking the code, when I mentioned the value is a list, so the standard name is SVO in this case, but probably not the best name. But we have a property called value, which gives a list of strings. Yeah. Yeah.

Look, one thing I'm concerned about is there are languages that have undefined word order, and I don't think we want to have a list of all the permutations of subject, verb, object.

I mean, that could be a missing undefined order or something. Could be a. If it's. If it's really not relevant, we can have a custom missing or some other.

Yeah. For these cases in the EP tables, the value would be a string that says there's no dominant. No dominant order.

Well, we can't do that. It's got to be something more computable.

It could be a missing. Not applicable. Not.

Not applicable. It would be a specific missing. It would be a missing, you know, no defined order or something.

No dominant order. Maybe.

Yeah, but. But are there languages in which it's, you know, SBO or vso, but not osv. Or is it the case that there's either a defined word order and it's blah or there's no defined word order? Good question. I mean, that's an easy question to answer by looking at one of these atlases. Yeah.

Isn't Latin fairly flexible about word order?

Not when I learned that it wasn't. It was always verb at the end. But I think it, I think, I think.

Oh, I think it's changed a lot since then.

The. No, I mean, I think it probably. The poetry probably works differently, but. I mean, there certainly are languages where the word order is undefined and where there are decorators on the words. Some language I was reading about recently, some indigenous Australian language that has some even more exotic mechanism. But. Okay, so look, I think. Okay, provisionally, let's determine the answer to that question. Is it either determined or completely undetermined?

I think we might have to. I'm just. I'm looking at walls right now. I think we might have to. I think we have to do a little more manual work. I'm thinking that the, the classifications and walls may really be expressing that there's a dominant order. I believe there could be exceptions in many languages. But what.

Yeah, I mean, if you go Yoda speak, you're going out of the standard order.

Right.

I believe if you open the word order subtitle, maybe there could be there what unique values we have. Yeah. So those are the values that we, we have.

Right. That's. That's what I'm saying is that's what comes out of walls. But that doesn't mean that it's not possible to have a different. That a different order wouldn't be valid in some cases in that language. Just means this is the dominant order for a given language.

It's kind of interesting.

It is.

So all the orders used by somebody, OSV is used by Klingon and maybe a few actual languages, unless all of them turn out to be online.

Are you serious?

Osp? Yeah. OSB is, is in the Klingon conlang.

Okay.

It's an OSB language. I think there was a whole backstory to that. I don't know whether I think there are some natural. Well, maybe three natural languages. Huh. I mean, it is sort of useful to be able to just see these things as osv, sov, et cetera, et cetera, et cetera. Look, maybe we could say, we could have a property name that's like word order type, and then another one Called word order list. See what I'm saying? Word order type would be the entity svo, for example.

Yeah, yeah.

Word order list is the grammatical units. And we could follow that principle for some of these other ones where we don't necessarily want to have a sequence. I don't know that these are. See, I don't know what the hell those are. But this is looking a lot more like. Oh yeah. I mean, this is not well represented. Whatever obligatory double negation is. Really the only way to make a positive statement is to make a not, not statement. It's interesting.

Yeah.

I believe there. There must be some definitions that below. So maybe there's more explanation for that.

No, it's just.

No, no, no. In the other table, I believe it's a second table.

Well, that's different. I mean, double negation means something very different. I mean, in logic it means, you know, not, not P. And lot of excluded middle that means P. I was thinking, you were telling me there were languages where in order to say yes, you have to say no. No.

Interesting idea.

Indeed. Well, I. I'm thinking from this that we're going to have to have named features and then we can have. For this particular case, which is really common, we have another property which is the string, which is the listified version of this. See what I'm saying?

Yes. I mean, that could also. That. That could also be a property of the entity.

Yeah.

As I mentioned.

Go one more step. But that. That could be cleaner.

Yeah. You have the thing you call value right now.

Yeah. Probably a better name, a better property name. But it already exists.

Fair enough. Well, we can go with that for now.

Yeah. Right. And there. There are various other ones that could be represented the same way. The entities could return a list of other grammatical units.

Yeah, but I mean these names like adjective and noun. Yeah, that's not a good name. I mean, it should be something like adjectival order or something. I think these. These things.

I think. I think also Christian, I mean, you. You've been fixing the display names and the scanner. I think we might also want to update some of the standard names of some of the properties too, because they are also a little.

Sure, yeah.

More terse than they should be. They'll be confusing in language.

Yeah. It would be good to have something that's either more readable.

I see on our live stream that Sander mentioning that in Twents, which I'm guessing might be a language spoken in 20th in the Netherlands. I have no idea that Joa. Joa, which I'M pronouncing, no doubt, with totally the wrong phonemes is a yes. Yes, which means no. It's a famous story. Well, at least of some. I think it was some philosopher in Oxford who was giving some talk about how, you know, double negation is an affirmative, but that, you know, that any number of affirmatives always means affirmative, so to speak, and somebody immediately says, yeah, yeah, which of course, is a counterexample like this one, because yaya means no. I don't believe you. So language is complicated.

Yeah, yeah. That have to do more with the intonation that the. Or the cadence.

Yes, that's true. All right, so we've got lots of properties here and there. Probably a bunch of property names, which I have. You guys gone through these property names in detail or not?

I think we need to go through them again. It's been a while since we looked closely at them.

Yeah.

This is weird. Eight to nine. It's. It's.

Anyway, that's how. That's how Walls divides them. Right.

They have.

They have some strange bins sometimes.

Boy, what a weird thing. Six or seven grammatical cases is the most common, apart from zero.

I mean, it's also a question of, you know, does. Is this a property for which we need to have an entity property table assigned, or should we. Should we just be returning, you know, zero interval, six, six, seven.

I think we should probably be returning.

You know, like these.

Blah, blah, blah. These.

These could be other Wolfram language structures and not.

I would think that. I don't know what grammatical cases, you know, zero or whatever.

Yeah. There's not much more you can say about those entities themselves because they do just represent some numerical bin.

Right. This is.

I was looking at this because you were asking again. This is, you know, their numbers, their intervals of things.

That's weird.

Walls doesn't give us VAL and constant inventories, but again, these are things that we can. We can get.

I think we should find these. I mean, I don't know why they're buying them this way. Seems.

I mean, it's. It's. It's what they do. I mean, they're. They're not. They're not. They're just trying to return their standard set of how for many dozens of properties in different categorical things. They do have a lot of references to other resources for all these languages, too, though, so dictionaries and papers and other things that have more detailed data for some of these things that we want to pick up.

Yeah. Moderately large constant inventory is not terribly useful to us. I Mean, look, even in English there's some fuzziness like with Y and things like this.

Yeah.

But I, Yeah, I mean, it also will be the case that if we can get what the actual consonants are in these languages, that would be super useful, right?

I believe it would be useful, but we also would have to consider that I, I believe the reason why it's like a range and not a set number is because it's, it kind of has variations depending of where language is spoken. For example.

Well, could also be that, you know, some, I don't know why in English, I don't think is often thought of as a consonant, but I don't know whether it's. I think we need something where. Well, we should see what people have done in terms of saying some things like, you know, m is uncontroversially a consonant in English and A is uncontroversially a vowel. And then there are more obscure cases like why, for example. And maybe we can have semi vowels. I don't know. I mean, maybe there's.

I mean it's a term that gets used sometimes.

Okay, well then that's a good thing. I mean, I'm thinking in, in. I'm not sure we've had that case very often of. There are three possibilities and things that definitively A, definitively B, or it's unclear. All right, let's just take a quick look at some of these other things.

Yeah, that's what I was going to ask. We could schedule another follow up and get into this because there's a lot here that's about. It ties, I mean, it ties in with the word list harmonization and how we want to deal with the fact that we have different word lists with, you know, different amounts of coverage of inflections and plurals and other things. And do we really want, do we want an entity for every possible inflected form of everything? If we don't, then how do we represent those things in all the different places that we're storing them? I wanted to ask, I mean, I see in the, the SDL meetings that you're coming up with a, some new form representing actions and, and some of these other things. They're not words, they're other things. And that's, I'm, I'm wondering if there's some, some other in between thing that we want to be thinking about with this like the notion of a decorator function for a word that says this is the, the plural of X. This is the whatever present perfect something of something.

Look, what we're going to do is we're going to have words which are things said in languages, and then we're going to have concepts, concept like things which represent the meanings of those words. So. And there should be a mapping from the word in the language to the meaning of the word. Although it could be that there are some meanings that are a single meaning but cannot be represented in less than five words in a given language.

Drew.

What on earth are those really weird. Things at the top?

Oh, I don't know. I mean, it's the legacy stuff from. From wordnet. Right. We've. We've generally clean. Yeah, I mean we've cleaned those out of the. The alpha version and the. The entity based version. It's just the. That's the legacy of wordnet that they had all this goofy stuff.

Well, I'm sure that people.

I mean, if it wasn't one of.

The creators of wordnet, I'll be disappointed.

I don't know.

Let's find out. I'm just gonna. Well, actually not. Oh, I see. Okay.

She still doesn't belong in word data.

She does not belong in word data. She might have had a very interesting and f. But she.

Great respect for her. But yeah. Okay, so should we schedule a follow up? Let's and try to get this in soon. And let's talk through the. The word specific part because I would like to move forward with figuring out what we're doing here.

Yes.

We're kind of right on the cusp of, you know, do we. Do we go in and add a few hundred thousand more word entities so we have something that represents all these different inflected forms. Because there's a lot we can do programmatically. There's some LLM assisted stuff that can help with this. We've been filling in, you know, sense tags for words with some LLM manual curation tag teaming. That's going well, but there's a bunch of little things that we want to just pin down so we can move forward.

Right. Well, I think we need.

Okay.

Yes.

Okay, let me. Let's get that on the schedule soon so we can keep going here.

Okay, sounds good. All right, we should probably wrap up for now and talk to you all soon. See you later.