I have built a database of Latin words but am having a few problems. I have two databases, one is from Python's CLTK, the other is from online-latin-dictionary.com (OLD). The Python database is derived from Collatinus which is more reliable but there are still problems with it. For example, 'malus' is only conjugated once as an adjective; it does not have the conjugation for 'malus' as a noun meaning apple tree. If it has that mistake you can imagine there are more mistakes like that. Plus the Python CLTK does not have macrons or vowel quantity. The OLD has macrons but there are a ton of mistakes in them. For example, they have 'mālus' marked in the head word with a macron over the 'a' but not in the conjugation. I'm going to go through my database and try to fix this. Sometimes the stem changes quite considerably during conjugation. Here I define 'stem' as that part of the word which is not part of a conjugation ending. I realize I've simply moved the problem of definition over to what is an ending but I hope you get what I mean. I need to know if the vowel quantity ever changes in the stem during conjugation. For example, suppose the quantity of 'a' were long in nominative in 'mālus' but short in the accusative 'malum'. That's just a hypothetical example to show you what I mean. I realize that the accusative of 'mālus' is not 'malum' but 'mālum'.

################## UPDATE ################

According to Asteroids there are 4 nouns that do this: pēs-pĕdis, mās-măris, sāl-sălis, pār-păris.

According to the Greenough Grammar in section 211 he lists the following verbs that have this property.

agō, ēgī , āct-, drive. lavō , lāvī , lōt- (laut-), wash (also regular of first conjugation). capiō , cēpī , capt-, take.

edō , ēdī , ēsum, eat (see § 201). legō ,143 lēgī , lēct-, gather. emō , ēmī , ēmpt-, buy. linō [LI], lēvī ( līvī ), lit-, smear. faciō , fēcī , fact-, make (see § 204). linquō [LIC], -līquī, -lict-, leave. fodiō , fōdī , foss-, dig. nōscō [GNO], nōvī , nōt- ((cō-gnit-, ā-gnit-, ad-gnit-), know. frangō [FRAG], frēgī , frāct-, break.

fugiō , fūgī , ( fugitūrus ), flee. rumpō [RUP], rūpī , rupt-, burst. fundō [FUD], fūdī , fūs-, pour. scabō , scābī , ----, scratch. faciō , iēcī , iact-, throw (-iciō, -iect-). vincō [VIC], vīcī , vict-, conquer.

Is this all of them? I checked with the developer of Collatinus and he said that Collatinus only checks for syllable quantity not vowel quantity. So once I understand how to use that database (I have to learn C++ to do this which will take some time) it will not be possible to get the vowel quantities from that database.

########### SECOND UPDATE ##########

The Jeanneau dictionary (my favorite Latin dictionary even though it's in French) does have irregular roots listed in its entries. So for the entry of pēs, they do have written pedis, with the hat over the e. (sorry I haven't yet figured out how to write the hat yet). And for the entry of vinco, they do have vīncī, written down. So I guess I can get the vowel quantity changes that way.

######## Third Update

Here's another relevant passage from the Greenough Grammar

a. Forms from the same stem have the same quantity: as, ămō , ămāvistī; gĕnus, gĕneris. Exceptions.—

1.  bōs , lār,mās, pār , pēs , sāl,—also arbōs,—have a long vowel in the nominative, though the stem-vowel is short (cf. genitive bŏvis etc.).
2.  Nouns in -or, genitive -ōris, have the vowel shortened before the final r: as,honŏr. (But this shortening is comparatively late, so that in early Latin these nominatives are often found long.)
3.  Verb-forms with vowel originally long regularly shorten it before final m, r, ort: as, amĕm , amĕr , dīcerĕr, amĕt (compare amēmus ), dīcerĕt , audĭt,fĭt. Note.--The final syllable in t of the perfect was long in old Latin, but is short in the classic period.

4.  A few long stem-syllables are shortened: as,ācer, ăcerbus . Sodē-iĕrō and pē-iĕrō, weakened from iūrō .

 b. Forms from the same root often show inherited variations of vowel quantity (see § 17): as, dīcō (cf. maledĭcus ); dūcō (dŭx, dŭcis); fīdō ( perfĭdus ) vōx , vōcis ( vŏcō ); lēx , lēgis ( lĕgō ).  c. Compounds retain the quantity of the words which compose them as, oc-cĭdō ( cădō ), oc-cīdō ( caedō ), in-īquus ( aequus ).  Note.--Greek words compounded with πρό have o short: as, prŏphēta , prŏlŏgus . Some Latin compounds of prō have o short: as, prŏficīscor , prŏfiteor . Compounds with ne vary: as, nĕfās , nĕgō , nĕqueō , nēquam .

  • 1
    This isn't at all an answer to your question, but I'm curious—what's the end goal of the database? And have you looked into automatic macronizers?
    – Draconis
    Commented Jun 23, 2021 at 4:00
  • 3
    Have you taken a look at this older question about vowel length variation in verb conjugation? It gives a pretty solid set of rules for all verb forms from the present stem.
    – Joonas Ilmavirta
    Commented Jun 23, 2021 at 10:52
  • 3
    just so you know, usually the abbreviation OLD is used to refer to the Oxford Latin Dictionary, the best English-Latin dictionary yet published. global.oup.com/academic/product/… I almost spilled my coffee when I was reading your sentence "The OLD has macrons but there are a ton of mistakes in them" :)
    – Alex B.
    Commented Jun 23, 2021 at 14:01
  • 1
    It needs to be digital. Print is worthless. They say they have a digital edition but I can't find it. I probably won't be able to use it however. If I can't get the words into a database then there is no point. In order to learn a language you need to learn a ton of rules. You can't quantify the rules but let's just say you could and that that number was 10,000. In order to remember 10,000 rules, you need to organize them into a database. You need to arrange them in order of importance and keep tabs on the rules you've mastered and the rules you're still struggling with.
    – bobsmith76
    Commented Jun 23, 2021 at 15:01
  • 2
    @bobsmith76: By the way, it's not that hard to find an ordered text-based version of the OED online. I use one in Golden Dictionary, so I rarely use the official OED programme from the CDs any more.
    – Cerberus
    Commented Jun 24, 2021 at 1:35

2 Answers 2


Yes; this is more common with conjugation (inflection of verbs) than with declension (inflection of nouns or adjectives), but there are examples with both.

To set aside the issue of "stems", I'll first give examples of cases where the vowel in the first syllable of a word changes (which would probably be considered part of the stem by most people).

In conjugation, a fairly large number of verbs have a short vowel in the present but a long vowel in the perfect. A few examples are lĕgo-lēgī, vĭdĕo-vīdī, ăgo-ēgī. The perfect passive participle stem can also have different vowel length from one or both of those stems.

In declension, there are a handful of words that unpredictably have a monosyllabic nominative singular form with a long vowel and other forms with a short vowel: pēs-pĕdis, mās-măris, sāl-sălis, pār-păris.

I think that the only vowels that change length between different forms of a declined noun or adjective are ones that fall in the last syllable in the nominative singular form of a word (the extremely irregular form "Iuppiter" is an exception due to the -piter stuck on the end).

As commenters have mentioned, vowels change length in many disyllabic or polysyllabic nouns and adjectives also, but I wasn't sure which of those you would consider to have "endings" since many were formed (at some point) with derivational suffixes. Examples are nouns ending in -or, -ōris or -al, -ālis. These both are usually categorized in one declension class (the "third declension") but the division of Latin declensional classes is a little arbitrary. Most vowel length changes in declension are based on regular rules about vowel length.

The existence of variation like this is part of the reason for the traditional concept of Latin words having multiple "principal parts" for a learner to memorize (even though Latin words mostly have forms that can all be derived from a single stem if you apply certain rules to it).

  • Would it be possible to write down as many example as you can. I think online-latin-dictionary might follow a pattern, in that if they forgot to write the macron in the 1st person singular indicative active for verbs or nominative masculine singular for adjectives or nominative singular for nouns, then it follows that they forgot on each declined form. But if they remembered to put it in those locations and the macron disappears in other declined forms, then this is not a mistake. I could then search the whole database and find each mistake.
    – bobsmith76
    Commented Jun 23, 2021 at 4:12
  • Then I could post a list of all words whose vowel quantity changes on at latindiscussion.com and you could take a look at the list I've generated. For example, online-latin-dictionary.come appears to have gotten the examples concerning 'video', 'lego', 'pedes' right.
    – bobsmith76
    Commented Jun 23, 2021 at 4:13
  • I found this good list of rules, so you don't need to link to this rharriso.sites.truman.edu/latin_vowel-quantity_macrons_macra
    – bobsmith76
    Commented Jun 23, 2021 at 4:17
  • 3rd person singular also lacks a long vowel. Cf. habēs v. habet.
    – cmw
    Commented Jun 23, 2021 at 4:52
  • There's also the honor-honōris type, with short vowel in the nom. sg. but long vowel elsewhere.
    – TKR
    Commented Jun 23, 2021 at 5:51

You mention in a comment:

The end goal is to learn Latin. In any system which has a gazillion rules, you need to organize the rules.

There are rules and there is vocabulary. I recommend that you redraw the line between them.

For the purposes of learning Latin, you should not think of a Latin word as a single stem, even if that is an etymologically accurate description. You should think of a word as a "set of stems" or "principal parts" or perhaps some other name. That is, "to touch" should not be seen as tag- but as the triplet tang-/tetig-/tact- with a bit of additional information describing the conjugation. Similarly, "hare" should not be seen as lepos- but as the doublet lepus/lepor- with a bit of additional information describing the declension.

In my experience this is far more efficient and reliable than trying to work out all the forms from a single stem. Both verbs and nouns often have too many possible rules you could follow, and I find little sense in trying to perfectly define which verbs form perfect forms through a lengthened vowel and which ones by reduplication. And some forms come out of left field: Iuppiter becomes Iovis and fero becomes tuli.

Latin morphology is very, very regular in comparison to any other language I know. Vowel quantity in the stem of a vowel is the same in all oblique cases of a noun — the only counterexample I can think of is bos. There are only a handful of words that need special treatment (pronouns, esse, posse, velle, ferre, bos, …) and for most words any irregularities are confined to one or two forms (dic, memento, …). You get the vast — really vast — majority of all forms correctly by taking the right principal and the right ending and sticking them together. The vowel quantities or any other properties between the two components don't seem to interact at all.

My point is this: If for every word you learn (or have a machine learn) the principal parts with all vowel lengths [this is vocabulary] and you learn all the endings of the relatively few classes of declension and conjugation [these are the rules], then you can easily and correctly reproduce almost all vowel lengths in Latin. You don't have to understand anything about the processes that relate the various principal parts together, although it does help remembering. The remaining set of exceptions is tiny.

Shift some of what you consider as rules to vocabulary and it all becomes far easier to manage. The rules become absolutely reliable but relatively simple, also regarding vowel quantity. From this point of view, the vowel quantity in the stem never changes in inflection.

  • So there are cases in which a word in nominative has one quantity but another quantity in oblique. But what about in the different oblique cases? Is it ever the case that a word has one vowel quantity in the stem in, say, genitive and another quantity in ablative? It seems like there is an exception to almost every rule in Latin. If there are only a handful of exceptions then I'm not worried about it, but if there are many, then I am.
    – bobsmith76
    Commented Jun 25, 2021 at 23:41
  • 3
    @bobsmith76 The different oblique cases always have the same quantity. That's what makes it all so easy! If there are exceptions, they're rare enough to be learnt one by one. All I can think of is various pronouns and bos. Latin morphology has very little exceptions compared to any other language I know.
    – Joonas Ilmavirta
    Commented Jun 26, 2021 at 0:42
  • Thanks, I appreciate your input.
    – bobsmith76
    Commented Jun 26, 2021 at 0:51
  • 1
    "Shift some of what you consider as rules to vocabulary and it all becomes far easier to manage" - hear! hear! couldn't agree more
    – Alex B.
    Commented Jun 26, 2021 at 14:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.