Lexical categories for beginners?

[sunandshadow]

I have a book about lexical categories from a local university library, but it's too high-level for me, I don't have the educational background to be able to read it. Anyone want to recommend a webpage or book about lexical categories that would be good for someone deciding what lexical categories to put in their conlang and how to mark them with affixes or position within the sentence?

[roninbodhisattva]

What's the book?

[Terra]

Read "Describing Morphosyntax": http://www.amazon.com/Describing-Morpho ... 0521588057

[sunandshadow]

What's the book?

Cambridge Studies In Linguistics #102: Lexical Categories: Verbs, Nouns, and Adjectives, by Mark C. Baker
http://www.amazon.com/Lexical-Categorie ... +cambridge

[sunandshadow]

Read "Describing Morphosyntax": http://www.amazon.com/Describing-Morpho ... 0521588057

This is unfortunately not available at any libraries I have access to.


Also, wikipedia says, "Since the Greek grammar of 2nd century BC, parts of speech have been defined by morphological, syntactic and semantic criteria. However, there is currently no generally agreed-upon classification scheme that can apply to all languages, or even a set of criteria upon which such a scheme should be based." Argh! A grenerally-agreed upon classification of parts of speech that can apply to all languages is exactly what a conlanger would want.

[Sakir]

[tongue in cheek]

1) Nouny Things
2) Verby Things
3) Particle Things
4) Adverbs

[/tongue in cheek]

[CaesarVincens]

Sunandshadow, if you haven't already, take a look at what Zompist has written for the language construction kit (online edition if you don't have the book edition).

From my perusal of Baker's Lexical Categories, nouns are words which refer to entities or concepts, verbs create predicates, and adjectives are just semantic stuff that needs support from either a noun or verb.

Anyway, if you are specifically talking about lexical categories (words with real semantics) and not functional ones, then you are sure to have some set of each nouns, verbs, and adjectives if you are going for a level of naturalism. Even languages which are claimed to lack one category or another have at least a few words that behave unlike the other two categories (that is, something that doesn't behave like a noun or a verb, but is lexical must be an adjective, particularly if it acts like one.)
Nouns, verbs, and adjectives are the only three classes of lexical categories generally agreed upon, and other classes are often subclasses of those or thrown in to fill out a paradigm (as adpositions are sometimes).

If you are talking about parts of speech, then that's a larger and much more complex subject, including functional categories along with the lexical ones.

Edit: I completely forgot about adverbs, but these tend to either be some form of adjective (with or without more morphology) or functional in nature.

[sunandshadow]

@CaesarVincens What I'm mainly interested in is learning about things that are hidden or absent in English, so I have a more universal pool of options to draw on when inventing how my conlang will work grammatically. I'm interested in things like noun cases, especially ones that aren't found in Latinate languages either. I'm interested in languages that don't form questions by rearranging word order, but by some other method.

[CaesarVincens]

Sure, well like I said, if you haven't read Zompist's stuff yet, do that. (I do recommend both the print Language Construction Kit and Advanced Language Construction (alias LCK2: Electric Boogaloo)). His books are really a good introduction into linguistic basics (and more advanced stuff also).

As for things like case marking, other than the morphosyntactic ones (nominative/accusative, ergative/absolutive), case names are usually pretty worthless for descriptive purposes. It's best to think about what you want to mark morphologically on nouns (and/or with the help of adpositions) and bundle those things into some set of cases with semi-descriptive/traditional names. Noun case marking can become very specific, just look at Finnish.

For questions, some methods include a separate question word (or a clitic) for yes/no questions, verb morphology, or simply leaving it to context/intonation.

If you have the opportunity to take classes in linguistics and haven't already I (as a Linguistics graduate) definitely recommend that as well.
You might also check out others' conlangs to see what they did for these sorts of things.

For example, a language I'm working on right now has the following cases; because of the number of them they have mostly well delineated roles.

Absolutive - subj. of intrans. verb/patient of trans. verb
Ergative - agent of trans. verb
Vocative - direct address to noun
Genitive - possession + partitive + possibly a few other things
Allative-Dative - end point of movement + recipient + possibly reference/interest + possibly 'patient' of anti-passive verb
Locative - location (in time or space)
Ablative - source of movement
Instrumental - instrument or manner of action + occasionally agent of 'passive' verb
Comitative - accompaniment or possibly manner of action + possibly reference/interest

Other possible situations or further refinement will use adpositions or verbal constructions (as for example contrasting being 'at somewhere' versus 'in somewhere').

Wikipedia has a list of case names; many of which are to do with location or movement.

[Salmoneus]

Read "Describing Morphosyntax": http://www.amazon.com/Describing-Morpho ... 0521588057

This is unfortunately not available at any libraries I have access to.


Also, wikipedia says, "Since the Greek grammar of 2nd century BC, parts of speech have been defined by morphological, syntactic and semantic criteria. However, there is currently no generally agreed-upon classification scheme that can apply to all languages, or even a set of criteria upon which such a scheme should be based." Argh! A grenerally-agreed upon classification of parts of speech that can apply to all languages is exactly what a conlanger would want.

A conlanger shouldn't want that, no - because it's impossible. What parts of speech a language has will depend on the language. For instance, english has a part of speech called 'adjectives'; but many languages do not have a part of speech equivalent to English adjectives, or, if they do, only as a subset of verbs. Similarly, English distinguishes between 'nouns' and 'prepositions', but some languages don't. On the other hand, English doesn't have a distinct part of speech for classifiers (if we use classifiers at all, they're just nouns), whereas some languages do. And so on.

[sunandshadow]

Read "Describing Morphosyntax": http://www.amazon.com/Describing-Morpho ... 0521588057

This is unfortunately not available at any libraries I have access to.


Also, wikipedia says, "Since the Greek grammar of 2nd century BC, parts of speech have been defined by morphological, syntactic and semantic criteria. However, there is currently no generally agreed-upon classification scheme that can apply to all languages, or even a set of criteria upon which such a scheme should be based." Argh! A grenerally-agreed upon classification of parts of speech that can apply to all languages is exactly what a conlanger would want.

A conlanger shouldn't want that, no - because it's impossible. What parts of speech a language has will depend on the language. For instance, english has a part of speech called 'adjectives'; but many languages do not have a part of speech equivalent to English adjectives, or, if they do, only as a subset of verbs. Similarly, English distinguishes between 'nouns' and 'prepositions', but some languages don't. On the other hand, English doesn't have a distinct part of speech for classifiers (if we use classifiers at all, they're just nouns), whereas some languages do. And so on.

I don't see why it's impossible to have a list of parts of speech found in all natural languages. There are a finite number of natural languages, and we can even exclude fragmentary ones. Of those, there are probably many cases where two closely related languages have exactly the same parts of speech. And there are a finite number of parts of speech in each language. The problem space isn't theoretically impossible, just impractically large.

[Terra]

However, there is currently no generally agreed-upon classification scheme that can apply to all languages, or even a set of criteria upon which such a scheme should be based." Argh! A grenerally-agreed upon classification of parts of speech that can apply to all languages is exactly what a conlanger would want.

How about this: How much do you know about computer programming? Read about BNF grammars: http://en.wikipedia.org/wiki/Backus%E2%80%93Naur_Form

Let's create a simple BNF grammar for English that can create only *simple* sentences:

Code: Select all

sentence = (noun_phrase transitive_verb_phrase noun_phrase)
noun_phrase = (pronoun) | (?article ?adjective noun)
transitive_verb_phrase = (adverb transitive_verb)

It can generate a sentence like "The man bravely broke the old mirror.", but not a sentence like "Because he was overweight, the man decided to stop eating bacon for breakfast.". This is because the grammar doesn't define prepositional phrases, adjectival predicates, reason clauses, and many other things. This grammar defines its parts of speech by only syntax.

This kind of approach creates problems because no two languages are going to have the exactly same grammar, so then how can we say that what each calls a "noun" is the same thing? Thus, a cross-language definition for a part-of-speech is defined semantically (by meaning). "Describing Morphosyntax" describes nouns as being things that are the most "time-stable". Things like a rock, tree, mountain, or horse are in this category. A fist or freedom are less so. Likewise, verbs are the least "time-stable". Events like dying, running, and breaking are in this category. Modifiers/adjectives are somewhere in-between these two categories. Some language simply do without them, and use either nouns or verbs in their place.

In Japanese, "adjectives" can conjugate like verbs:
- Inu-ga {aruku}. == The dog {walks}.
- Inu-ga {aruita}. == The dog {walked}.
- Inu-ga {arukanai}. == The dog {doesn't walk}.
- Inu-ga {arukanakatta}. == The dog {didn't walk}.

- Fuyu-ga {samui}. == The winter {is cold}.
- Fuyu-ga {samukatta}. == The winter {was cold}.
- Fuyu-ga {samukunai}. == The winter {isn't cold}.
- Fuyu-ga {samukunakatta}. == The winter {wasn't cold}.

Also, Japanese has no such thing as relative clauses. A clause can modify a noun just like an "adjective".
- {Akai} inu-ga samui. == The {red} dog is cold.
- {Gakkou-ni aruita} inu-ga samui. == The dog {that walked to the school} is cold.

English doesn't have a distinct part of speech for classifiers (if we use classifiers at all, they're just nouns), whereas some languages do. And so on.

Indeed, Japanese does have a part of speech for classifiers. Mapping this Japanese PoS to an English PoS doesn't really make sense, because English doesn't have distinct category (like in the BNF grammar) for this kind of this, while Japanese does. But, it would make sense to map the Japanese category to a Korean or Chinese one, because those languages also have a distinct category.

Does this help you understand at all? Any questions?

(May I ask how old you are? Some of this may be hard to learn, because it involves unlearning what you learned in English class (like that nouns are "persons, places, or things"). English teachers are (generally) not good linguists.)

[CaesarVincens]

I don't see why it's impossible to have a list of parts of speech found in all natural languages. There are a finite number of natural languages, and we can even exclude fragmentary ones. Of those, there are probably many cases where two closely related languages have exactly the same parts of speech. And there are a finite number of parts of speech in each language. The problem space isn't theoretically impossible, just impractically large.

A comprehensive list of parts of speech is possible, but it has little value in part because different people disagree on what exactly to call different words in different languages. For example, if an adjective can be conjugated like a verb, some would say it's actually a verb. And when one gets to functional categories like classifiers, quantifiers, conjunctions/complementizers, it becomes all the more contentious.

I think the lesson to draw from this is not to worry too much about what part of speech some word or another is; that is only a short hand way of referring to its syntactic properties. For myself, a useful distinction might be something like: nouns, adjectives (plus adverbs), verbs, adpositions, determiners (including quantifiers and classifiers for most purposes), particles. There are probably a few more distinctions I could make, particle is after all a catch-all category for everything else.

[linguoboy]

I don't see why it's impossible to have a list of parts of speech found in all natural languages.

This assumes that everyone can agree on what parts of speech are found in any given natural language. But they can't; just compare a few competing analyses of a well-studied language like English or Chinese. (Quick question: What part of speech is his? What about therefore?)

[zompist]

What you're asking for is basically an introduction to morphology. The online LCK is a good place to start, and the print version is even more so.

A mere list is of no use, because syntactic categories are not 'real things'; they're an artifact of analysis, and very language-dependent. The real universals live at the level of morphosyntax; either Payne's book or my ALC will deal with that.

Here's some of the categories my Mandarin grammar uses:

place word
measure word
attributive
movable adverb
specifier
stative verb
classificatory verb
coverb
particle
marker
localizer

You would have to read the grammar or know Chinese to know why these are considered syntactic categories and how they behave. Just getting a list like this is nearly useless, and trying to combine it with (say) a list of Latin categories would just lead to confusion.

[sunandshadow]

Wow, lots of posts to respond to! Thank you all for taking the time to respond helpfully! I especially like the lists and examples, even though I'm not going to respond to them in this post; have to think about them and mentally digest them first. I'll try to answer all questions asked of me but if I miss one please point out that I missed it.

I did go and reread the relevant Language Construction Kit section - it's a nice sampler, I was just wishing for something more systematic. Like "survey course of parts of speech around the world" or something like that. The Lexical Categories book I have but am having difficulty understanding isn't a survey either, it's a generative grammar based theory about what differentiates lexical categories from each other. So it has mentioned some interesting things in passing (benefactive case? desiderative case? focus particle?) and that works because I can google those one at a time. But it will end up being incomplete, and meanwhile it's inefficient.

I am 33. I had traditional grammar in elementary and middle school with this kind of diagramming: http://www.grossmont.edu/karl.sherlock/ ... amming.htm and then I had an introduction to transformational generative grammar course in college. That was slightly before the Lexical Categories book I have was written. And according to the that book, transformational generative grammar theory before that point focused on how the parts of speech were the same, and had little to say about what differentiated them. I went on to do a project about transformational generation of story structure, but didn't do anything further with specifics of grammar. I had never encountered the term "clitic" until this past month.

If syntactic categories are artifacts of analysis, are there schools of analysis, such that each school tends to identify the same features across several languages? If so, maybe I could just find a school that felt right to me and go with their analysis.

His - third person singular masculine possessive pronoun? Therefore - conjunction?

[CaesarVincens]

Wow, lots of posts to respond to! Thank you all for taking the time to respond helpfully! I especially like the lists and examples, even though I'm not going to respond to them in this post; have to think about them and mentally digest them first. I'll try to answer all questions asked of me but if I miss one please point out that I missed it.

I did go and reread the relevant Language Construction Kit section - it's a nice sampler, I was just wishing for something more systematic. Like "survey course of parts of speech around the world" or something like that. The Lexical Categories book I have but am having difficulty understanding isn't a survey either, it's a generative grammar based theory about what differentiates lexical categories from each other. So it has mentioned some interesting things in passing (benefactive case? desiderative case? focus particle?) and that works because I can google those one at a time. But it will end up being incomplete, and meanwhile it's inefficient.

I am 33. I had traditional grammar in elementary and middle school with this kind of diagramming: http://www.grossmont.edu/karl.sherlock/ ... amming.htm and then I had an introduction to transformational generative grammar course in college. That was slightly before the Lexical Categories book I have was written. And according to the that book, transformational generative grammar theory before that point focused on how the parts of speech were the same, and had little to say about what differentiated them. I went on to do a project about transformational generation of story structure, but didn't do anything further with specifics of grammar. I had never encountered the term "clitic" until this past month.

If syntactic categories are artifacts of analysis, are there schools of analysis, such that each school tends to identify the same features across several languages? If so, maybe I could just find a school that felt right to me and go with their analysis.

His - third person singular masculine possessive pronoun? Therefore - conjunction?

I think you are coming to a good conclusion. The importance is not so much in the names, but in the actual patterns of usage. Thus we call words that act like verbs verbs (being marked for Tense, Aspect, Mood, Personal Agreement) and so forth, but when one comes to trickier categories (usually functional rather than lexical) terminology multiplies.

[Terra]

If syntactic categories are artifacts of analysis, are there schools of analysis, such that each school tends to identify the same features across several languages? If so, maybe I could just find a school that felt right to me and go with their analysis.

His - third person singular masculine possessive pronoun? Therefore - conjunction?

What's the PoS of "and" and "or" then?
Good: The man and the woman went to the store.
Bad: The man therefore the woman went to the store.

So it has mentioned some interesting things in passing (benefactive case? desiderative case? focus particle?

Those aren't parts of speech though. Also, be aware that cases are often specific to a certain language.
For example, IE languages often have their word/case for "with" do things that require multiple words/cases in other languages.

For example (English vs Japanese):
I walked to the shop with my wife. :: ("with" marks a companion) == Mise-ni tsuma-to aruita.
I cut the vegetables with the knife. :: ("with" marks a tool) == Aomono-o naifu-de kita.

English uses "with" in both, but Japanese uses "to" and "de" depending on the meaning. The term to refer to meanings free of language-specific case is "semantic role". Perhaps you'll enjoy reading this: http://www-01.sil.org/linguistics/Gloss ... icRole.htm

Bonus: Japanese uses "de" in places where English uses "by" or verbs the whole thing:
I biked to the shop. // I went to the shop by bike. == Mise-ni jitensha-de itta.

"de" can mark either a tool or a vehicle.

[sunandshadow]

His - third person singular masculine possessive pronoun? Therefore - conjunction?

What's the PoS of "and" and "or" then?
Good: The man and the woman went to the store.
Bad: The man therefore the woman went to the store.

There's no reason all conjunctions would have to work in that particular sentence. You can't say "The man but the woman went to the store." either. Even "The man or the woman went to the store" is weird because it's just something one would never say. But if you re-worded a bit, you could say, "The man went therefore the woman went too." And you could say, "The man went but the woman did not." And, "Did the man or the woman go?" A conjunction expresses a relationship between two things, but some things can't sensibly or clearly be related in some ways in some contexts, so we don't say it that way.

So it has mentioned some interesting things in passing (benefactive case? desiderative case? focus particle?

Those aren't parts of speech though. Also, be aware that cases are often specific to a certain language.
For example, IE languages often have their word/case for "with" do things that require multiple words/cases in other languages.[/quote]
Are you sure there isn't a language somewhere in the world where they aren't parts of speech, or at least subtypes of parts of speech? In English, we can say that pronouns in the subjective case and pronouns in the objective case are two different subtypes of a part of speech, right? Then logically the same could be said about benefactive nouns in some other language.

Would you consider affixes to be a part of speech? I think separable-prefix verbs in German are an interesting example of some affixes revealing their true nature as prepositions that are normally stuck to a verb, but not actually "verby" in themselves. On the other hand, a normal pronoun isn't an affix, so an "affixed pronoun" has to be a slightly different part of speech than a pronoun, I would think.

For example (English vs Japanese):
I walked to the shop with my wife. :: ("with" marks a companion) == Mise-ni tsuma-to aruita.
I cut the vegetables with the knife. :: ("with" marks a tool) == Aomono-o naifu-de kita.

English uses "with" in both, but Japanese uses "to" and "de" depending on the meaning. The term to refer to meanings free of language-specific case is "semantic role". Perhaps you'll enjoy reading this: http://www-01.sil.org/linguistics/Gloss ... icRole.htm

Bonus: Japanese uses "de" in places where English uses "by" or verbs the whole thing:
I biked to the shop. // I went to the shop by bike. == Mise-ni jitensha-de itta.

"de" can mark either a tool or a vehicle.

I do like that link about semantic roles. I'm very interested in how different semantic roles are represented differently in different languages. Something like your example of a noun taking a different ending depending on whether it is a companion or a tool is precisely the kind of thing I want to learn about and decide what kind of similar things to put into my conlang! ^_^

[sunandshadow]

I think you are coming to a good conclusion. The importance is not so much in the names, but in the actual patterns of usage. Thus we call words that act like verbs verbs (being marked for Tense, Aspect, Mood, Personal Agreement) and so forth, but when one comes to trickier categories (usually functional rather than lexical) terminology multiplies.

Yeah, I'm not hung up on the names so much. I just want to be able to label the functions I want to have in my language, so I have an English term to label each group of conlang words with and an English term to use as a guideline if I have to diagram a sentence I'm trying to translate from English to the conlang to make sure I translate it consistently with my own rules.

[zompist]

His - third person singular masculine possessive pronoun? Therefore - conjunction?

What's the PoS of "and" and "or" then?
Good: The man and the woman went to the store.
Bad: The man therefore the woman went to the store.

There's no reason all conjunctions would have to work in that particular sentence. You can't say "The man but the woman went to the store." either. Even "The man or the woman went to the store" is weird because it's just something one would never say. But if you re-worded a bit, you could say, "The man went therefore the woman went too." And you could say, "The man went but the woman did not." And, "Did the man or the woman go?" A conjunction expresses a relationship between two things, but some things can't sensibly or clearly be related in some ways in some contexts, so we don't say it that way.

Ehhhhh, not quite. You want to avoid, or at least minimize, defining categories semantically. Ideally you do it with syntax alone. E.g., a conjunction allows replacing a constituent with two or more of the same syntactic type. So there is some surprise that you can't say "The man therefore the woman".

It can occur between sentences, as in your example, but notice that you can also say

The man went; the woman therefore went with him.
"He's gone. I will therefore go after him."

You can't do that with "and", "or", "but", etc. That makes two strikes against "therefore" as a conjunction. My dictionary in fact lists it as an adverb. (But don't always trust dictionaries on this. "Adverb" is all too often the trash bin of syntax, taking everything that doesn't fit elsewhere.)

I go into this to underline that it really can be complex, and to give just a hint of what syntax is like. It's one of my favorite fields, but I don't know of a really good simple introduction. James McCawley's "The Syntactic Phenomena of English" is good, but long.

[Ngohe]

It's with parts of speech as with philosophical concepts like "time", "existence" or "justice" - we can understand them in an everyday sense, but it quickly becomes more problematic when we seek to formally define them in an absolute, universal way, applicable anywhere and anytime.


I think the preliminary answer might be that which parts of speech there are depends on what is most useful for the particular purpose. Definitions of parts of speech can thus depend on which language one is studying. Even for the same language, different scholars might posit different parts of speech.

Especially adjectives seem to be an often contested part of speech. In many languages, they share characteristics either with nouns or with verb. But even if they can be grouped together with verbs (or nouns), they might have certain characteristics that sets them apart from other verbs (or nouns) - such as only taking certain aspects, or appearing in certain contexts where regular verbs (or nouns) cannot appear (or vice versa).

(As for Japanese, I think Baker's position in Lexical Categories was that it has adjective. I don't remember his argument, but apparently he must have thought the "adjectival" nouns and verbs have enough in common that motivates treating them as a distinct category.)

Not all verb (or noun, etc.) behave exactly the same. You can frequently find sub-classes of verb, with various characteristics that sets them apart from other verb. The question is, how distinct should a certain sub-class be, in order to be considered a "separate" part of speech, rather than as sub-class of some other part of speech?

My conlang Wakeu is a language in which adjectives behave like "stative verbs". When I describe Wakeu, I can sometimes treat adjectives as a separate category, and sometimes as a sub-category of verbs. For me, it's much a question of what's simplest or most pedagogical in the situation. At least in a regular "school-grammar" type presentation, I don't need to take side on the theoretical issue whether there "really" are adjectives in my language.

[Salmoneus]

The example of conjunctions shows one reason why you can't just draw up a list of all parts of speech in use in real languages. Because the more you look at languages the more complicated they get. And you're left with increasingly small sets of words that act in particular ways. To what extent can these sets be seen as subgroups of parts of speech, to what extent are they parts of speech themselves, to what extent are they all basically the same part of speech but with specific lexical constraints on certain specimens, to what extent are these irregularities in fact only semantic rather than grammatical.... or are irregularities due to certain words in fact underlyingly being two different words that sound the same?

For instance, "and" and "but". We can say:
1) You went to the shop and I went to the house
and
2) You went to the shop but I went to the house

So they look like the same part of speech. But then:
3) You and I went to the shop
*4) You but I went to the shop

Why are these different? Some possible reasons:
a) They are both conjunctions, 4 is impossible purely for semantic reasons because it makes no sense
b) They are both conjunctions, but 4 is impossible because 'but' has a specific rule that says it can only be used between two nouns if at least one is qualified or negated
c) They are both conjunctions, but 'and' is one subtype and 'but' is another. The first subtype can conjoin any strings of equal status, but the second cannot conjoin bare nominal phrases - however, adding a qualifier lifts a phrase above the status of bare nominal phrase, and hence 'but' can conjoin things with qualifiers
d) As above, but these are two different parts of speech
e) Both words are the same POS (which cannot conjoin bare nominal phrases) - but 'and' is homophonous with another word 'and' that can, and can only, conjoin bare nominal phrases (in some other languages, these 'two' words are not the same).

And then look more closely at 'but':
*5) X but Y
6) not X but Y
Hence what I said about qualifiers. And yet:
*7) some X but Y
So in e), we might want to say that 'but' is a clause-level conjunction, but that 'not... but' is a phrase-level conjunction. But then what about 'never... but sometimes' and so on? Are all these variants all lexical items, or are they one lexical item ('but') with complicated rules?

And then you've got 'therefore', 'while', 'yet' and so on, which are never phrase-level conjunctions, but do work as clause-level conjunctions. But these don't work all the same way either, because while:
I went to the shop while I was eating a pygmy
and
9) I went to the shop therefore I was eating a pygmy,
nonetheless
*10) I went to the shop therefore eating a pygmy
even though
11) I went to the shop while eating a pygmy
And yet probably
?12) I was going to the shop, therefore eating a pygmy
[OK, the semantics there make it iffy, but try "I was on the street after eight, therefore breaking curfew". Now, is this an issue of how 'therefore' can be used, or is it an issue of when repeated subjects and verbs can be elided?]

So which things do you call 'conjunctions'? All of them? Some of them? And how do these categories line up with the category you label 'conjunctions' in another language?

[CaesarVincens]

Conjunctions are a terrible category in general. Often there are "conjunctions" related to prepositions (after, before); plus the classical distinction of "subordinating conjunction" vs. "coordinating conjunction".

Myself, I would posit at least two categories: conjunctions (joining any two equal X and Y) and complementizers (classical subordinators).

For most natural speech, one cannot have a bare clause that uses a complementizer without a matrix(=main) clause, without recourse to an elided clause from context.

*After he ate the pygmy. (Unless in response to "when did he leave?")
He ran before the rain fell.

But is a special case in English (can probably most languages), I wanted to call it a disjunction before I remembered that term is reserved in logic for "or". I might call it a "conversative" and perhaps throw in words like "although".
It seems to me that "but" requires a semantic reversal somehow so that if I say one thing, but mean another, that's okay, whereas if I ran but jumped, it's strange to ungrammatical. Using a negation on either side of the "but" makes it much easier to imply or state that reversal it seems. ("I ran but didn't jump" is perfectly fine to me; further compare to "I ran and didn't jump".)

[Terra]

There's no reason all conjunctions would have to work in that particular sentence.

Yes, yes there is: Because that's the actual (syntactic) definition of a conjunction: A conjunction is anything that can take that place in that sentence. If something can't take that place, then it's not a conjunction.

There's no reason all conjunctions would have to work in that particular sentence. You can't say "The man but the woman went to the store." either. Even "The man or the woman went to the store" is weird because it's just something one would never say. But if you re-worded a bit, you could say, "The man went therefore the woman went too." And you could say, "The man went but the woman did not." And, "Did the man or the woman go?" A conjunction expresses a relationship between two things, but some things can't sensibly or clearly be related in some ways in some contexts, so we don't say it that way.

So? What makes you think that "but" is the same kind of "conjunction" as "and" and "or"?

"The man went therefore the woman went too."

This is different. You are no longer joining 2 nouns, but 2 clauses.

Examples:
(Joining 2 nouns)
Good: The man and the woman are hungry.
Good: (Either) the man or the woman is hungry.
Bad: The man but the woman are/is hungry.
Bad: The man therefore the woman are/is hungry.

(Joining 2 clauses)
Good: The man is hungry, and the woman is thirsty.
Good: (Either) the man is hungry, or the the woman is thirsty.
Good: The man is hungry, but the woman is thirsty.
Good: The man is hungry, therefore the woman is thirsty.

"but" looks like it's the same PoS as "therefore", not "and" or "or". Furthermore, it doesn't make sense to label "and" and "or" as a single PoS. Instead, I would say that they can be one of two PoS depending on the situation.

A conjunction expresses a relationship between two things, but some things can't sensibly or clearly be related in some ways in some contexts, so we don't say it that way.

As Zompist said, that's defining a category semantically. Within a single language, you should aim to (and can) do it with only syntax. Defining a semantic definition should be reserved cross-linguistic senses, because as I said before, no two languages have the exact same grammar.

I suppose you could define "conjunction" as a semantic category in English, which includes two parts-of-speech, one that joins nouns ("and, or"), and one that joins verbs ("and, or, but, therefore"), but you should be very clear about the difference: parts-of-speech are syntactic and ultimately the only thing that matters for grammaticalness.