Summary | Introduction | Guiding Philosophy

Guiding philosophy


The dictionary is intended as a help to someone who knows roughly enough Latin for the document under study. It gives the accidence and meanings possible for an input Latin word. It is for someone reading Latin text.

This is a translation dictionary. Mostly it provides individual words in English that correspond to, and might be used in a translation of, words in Latin text. The program assumes a fair command of English. This is in contrast to a conventional same-language desktop dictionary which would explain the meanings of words in the same language. The distinction may be obvious but it is important. A Latin dictionary in medieval times would have explanations in Latin of Latin words.

There are various approaches to the preparation of a dictionary. The most scholarly might be to select only proper and correct entries, only correct derivations, grammar, and spelling. This would be a dictionary for one who wished to write ‘correct’ Latin. (Correct being defined as the way Cicero, or your favorite writer or grammarian, used it.) The current project has a different goal. This program is successful if a word found in text is given an appropriate meaning, whether or not that word is spelled in the generally approved way, or is ‘good Latin’. Thus the program includes various words and forms that may have been rejected by recent scholars, but still appear in some texts. Philosophically, thus program deals with Latin as it was, not as it should have been. I make no corrections to Cicero, which some might have been tempted to do if producing an academic dictionary instead of a program. Moreover I make no corrections of St Jerome. If your copy of the Vulgate has a particular spelling, that may be recognized by the program, either through a TRICK or as a dictionary entry that I have generated.

A philosophical difference from many dictionary projects is that this one has no firm model of the user or application. It is not limited to classical Latin, or to ‘good practice’, or to common words, or to words appearing in certain texts. As a result there will be a lot of chaff in the output. Some of this may be trimmed out automatically if desired, but it is there and available.

However inadequately, I hope to document decisions that went into the arrangement of the program and dictionary. I am surprised that there is little or no such information to the user of published dictionaries. If others generate similar products, or use the data from this one, they can do so in knowledge of how and why processes and forms were constructed.

I make few value judgments and those are mechanical, not scholarly, and are documented herein. Nevertheless some may be inappropriate, in spite of good intentions.


The program subtracts possible endings from an input words and searches a list of stems, trying to make a match. If no exact match is possible, it tries various modifications, beginning with prefixes and suffixes, and eventually involving various regular spelling variations (or ‘tricks’) common in classical and medieval Latin.

A choice was made that the base was classical Latin as defined by the Oxford Latin Dictionary (OLD). Their primary time period is arbitrary/roughly 100 BC to 100 AD.

The classical form of words is taken as the base. Modifications are in such a way to correct to this base. Further additions to local dictionaries should keep this in mind. Modifications are made to the input words, not to the dictionary stems. It could be done the other way, but the present situation was initially much easier. There are some consequences of this approach. For instance, it is easy to remove an ‘h’ from an input word to match with a stem. It is much more difficult (but not impossible) to add ‘h’ in all possible positions to check against stems.

It would be possible to match most words with a relatively smaller list of stems (or roots) and generous application of word construction. This approach is not followed. One difficulty is that while words may be constructed correctly, and the underlying meaning to be found from this construction, the common usage may be obscured by a formal interpretation of the parts. In practice this occurs in 20-40% of the cases. This method is still very useful in approaching a word for which there has been no dictionary interpretation, but it puts a considerable burden on the normal user. Further, in about 10% of constructions, the result is just wrong.

In normal usage, if the program finds a simple match, it does not go further and consider what constructed words might also be valid. (One can override and force prefix/suffix construction with a switch, but one might not want to force all possible tricks.)

For instance, if there is an adjective that matches, a corresponding identically spelled, logically valid noun will not be reported unless it is explicitly found in the dictionary, even though it could be constructed or inferred from the adjective or constructed with a suffix from a verb in the dictionary.

An exception to this is that enclitics (eg., -que) are always considered. Coloque can be a verb or collo-que. The latter is in Virgil and should not be omitted. Verb syncope is also favored. In the vast majority of cases, if there is a possible syncope it is the correct parse. This is given preference over word construction with suffix. Audii is syncope of audivi, but it could also be aud-i-i. The latter is considered very unlikely.

There are a large number of paths and possibilities. Choices have been made in the code that result in the exclusion of some. It is hoped that they were the best choices. The method was constructed by taking a number of primary procedures and combining/assembling them in such a way as to give reasonable parses for a number of test cases. Basically, this is hacking, but it might be considered an empirical starting point from which one could construct a logical rationale.

Therefore, the philosophy is to populate the stem list as densely as possible. Even easily resolved differences are included redundantly (adligo as well as alligo - ad- is most of duplicates). The advantage is that while regular single-letter modifications are fairly easy, and two letter differences are possible (but more expensive), further deviations are problematical. The better populated the stem list, the better the chance of a result.

Even in easy cases the overpopulation is helpful. Antebasis is easily parsed as ante-basis (‘pedestal before’, which is reasonable), but inclusion as a separate word allows the additional information that it is the hindmost pillar of the pedestal of a ballista.

The stem list is also populated with variants suggested by different sources. The problem is that the remains of classical Latin have gone through many monks along the way. These copyists may have made simple mistakes (typos!), or have made what they thought were proper corrections (spell checkers!). And twenty centuries later scholars work hard to reassemble the best Latin to present in the dictionary. But a particular document in the form presented to the reader may have have a variety of spellings for exactly the same word in the same referenced passage (Pliny’s Natural History is often subject to this problem). (It may even be that modern texts and dictionaries have misprints!) All forms found in various dictionaries can be included, with the exception of those explicitly labeled ‘misread’ (and the argument probably could mandate their inclusion also). However, a single example of a variant in one case will not be included as a dictionary entry. If such a word is sufficiently important, if it is used frequently or by several authors, it will be entered as a UNIQUE.

Lewis and Short seem to be more willing than the more recent Oxford Latin Dictionary to raise a few examples of variation to an entry (at least an alternate). Generally, I make an entry if some dictionary does so. But within an entry I generate additional possible stems not noted elsewhere, e.g., I expand first declension verbs with ‘-av’ perfect stems, even though no example exists in classical Latin. This is often the practice in other dictionaries also.

Verb parts omitted from source dictionaries are mechanically added where it is clear, (ex. where the base verb is documented, but parts are omitted in compounds). Whether Cicero used them or not, some later text might.

In some cases I also have expanded adjectives and adverbs to include comparative and superlative stems where they seem reasonable or have corresponding English instances, even when there is no specific dictionary citation. This effort was motivated primarily by finding examples of such comparisons in processing of large amounts of text beyond the classical works upon which authoritative dictionaries are based, but even classical works yielded examples. The point is that, while these forms would usually be caught by the word formation (prefix/suffix) process in the program, the process is limited to how many operations can be done serially. Having more/expended stems allows another level of word modification to be implemented.

Adjectives are extrapolated to COMP and SUPER where it makes sense (when those meanings are reasonable, and in many cases they are not) even if the source dictionary only lists POS. They are expanded fully especially even when the source lists a COMP but no SUPER.

Perhaps a bit out of context, consider the common question of SECLORUM in the Great Seal of the USA. This pure word in not in any dictionary I know of, not the OLS or L+S. A simple trick gives seculorum (seculum = world), but the favored translation is from the twice modified saeculorum (saeculum = age), which would not be found by a minimalistic system.

It is often the practice in paper dictionaries to double up on an entry that may be either adjective or noun, usually by leading with the adjective and mentioning its use as a noun. A much larger set of adjective/noun pairs is favored with separate entries. It is the philosophy of this program to make separate entries whenever there is an example in any reference dictionary. This might facilitate the task of a larger translation program which would handle phrases or sentences. However there has been no effort to explicitly generate such pair expansion if there is no precedent, and the user must still recognize the possibility of unexpanded multiple possibilities for substantives.

An argument against a large stem list is that it increases the storage required (but this is extremely modest by current standards) and increases processing time for search of the stems (this is far offset by the processing which would be required to construct or analyze words working from a smaller stem list).

A significant objection is that artificially generated stems may conflict with real/common ones and produce false output confusing to the user. A certain amount of this is eliminated by trimming the output to emphasize the most probable results, but it is still a problem.

Perhaps a counterexample would be an inferred fourth stem to no/nare (swim). Natus conflicts with the fourth stem of nascor (be produced/born) and the nouns and adjectives stemming from it. The nare natus does not appear in dictionaries, nor does it occur in compounds of nare, so it has been omitted from the WORDS dictionary.

Additional parts of verbs are included (first conjugation is easily filled out, even eccentric verbs if they are compounds of known parts), although they may not have been found in any well known texts. Cases can be logically constructed that are ‘missing’ in classical Latin. Verbs with prefix can be expanded when the base is known. That a form has not been found in surviving copies of classical texts does not mean that it was not on the lips of every centurion and his girl friend, or that it might not find its way into medieval texts.

It may be argued in some cases that forms are missing because their pronunciation would be awkward. This may well be true when Cicero is the arbiter, but others may not be so elegant. Moreover, much of the texts are represented by medieval documents, Latin the was written but may not have been spoken, so the problem did not arise. However, I might be willing to accept this argument for considering carefully some perfect stems of first conjugation verbs which otherwise would end in -avav. In the end, the only one I found that I could not support was lavo (wash), and its compounds, for which the perfect is lavi.

In some cases there are good reasons not to do the mathematical expansion, and these are pointedly avoided. There is no mechanical generation of, for instance, conl- words for every coll- word, unless there is some citation or reasonable rationale. They may be paired in almost every case, but, for instance, collis and collyra are not. However, forms that are mentioned in dictionaries explicitly, or implicitly by being derived from words having variant forms, are included in order to reduce the dependence on ‘tricks’. OLD has a conp- for almost every comp- (except derivatives from como). Rare exceptions seem to be rare words for which few examples (or only one) exist. Even in some of these cases, OLD (mechanically?) gives two forms. L+S follows the same pattern, except for words of late Latin (which would not be found in OLD). It is presumed that the general practice in later times was always to use comp-, and the program dictionary follows that. There are many acc-/adc- pairs, but OLD has a fair number of acc- words without mention of a corresponding adc-, and so the possible generation of these words has been resisted. If an example turns up in text, the appropriate trick procedure should suffice

One suspects that some amount of analytical expansion is present even in the best dictionaries. Otherwise how can one explain four alternate spellings for a word which apparently only appears in citation as a single inscription.

In a some few cases I have inferred a declension to certain very obscure Greek words which other dictionaries have treated as indeclinable (having only a single classical example of its use). My argument is that some later writer, using this word, might attempt to decline in it in a conventional manner, no matter what Vitruvius thought. I have indicated the indecl. option in the meaning.

Adjectives from participles are included if an entry is found in some reference dictionary. In some case the adjective has a special meaning not obvious from the verb. The program will return both the adjective and the participle with its verb meaning. The user should give some additional consideration to the adjective meaning in this case. If the adjective is marked rare while the verb is common, it is likely there is reference to a special meaning.

Tricks are expensive in processing time. Each possible modification is made, then the resulting word goes through the full recognition process. If it passed, that is reported as the answer. If it fails, another trick is tried. This is effective if very few words get this far. It is expected that application of single tricks will solve most of the resolvable difficulties. It would be impractical to mechanically apply several tricks in series to a word. A large stem population reduces the likelihood of multiple tricks being required. If the dictionary is heavily and redundantly populated, tricks are rarely necessary (and therefore not an overall processing burden) and largely successful (if the input word is a valid, but unusual, variant/construction).

Further, a conventional dictionary, especially one that wishes to set a standard for proper language, excludes words that may not meet criteria of propriety, slang, misspellings, etc. This may place the onus on the reader to convert words. A computer dictionary ought to relieve the reader as much as possible. The present program may be a far way from complete, but it’s goal is to strive for that.

Word Meanings

The meanings listed are generally those in the literature/dictionaries. In the case of common words, there is general agreement among authors. Some uncommon words display convoluted interpretations.

Generally, the meaning is given for the base word, as is usual for dictionaries. For the verb, it will be a present meaning, even when the tense input is perfect. For an adjective, the positive meaning is given, even if a comparative or superlative form is shown. This is also so when a word is constructed with a suffix, thus an adverb constructed from its adjective will show the base adjective meaning and an indication of how to make the adverb in English.

For the level of usage for this program, and for convenience in coding, the meaning field has been fixed at 80 characters. It is possible to have multiple 80 character lines for an entry, but this only necessary for the most common words. In order to conserve space, extraneous helpers like ‘a’, ‘the’, ‘to’, which sometimes appear in dictionary definitions, are generally omitted. The solidus (‘/’) is used both to separate equivalent English meanings and to conserve space.

I have taken it upon myself to add some interpretations and synonyms, and propose common usage for otherwise complex descriptive definitions. The idea is to prompt the reader, expecting that the text may not be that from which some dictionary copied the meaning (from some 18th century translator!).

In the meanings I only use words of which I know the meaning. I find that in some cases the Oxford Latin Dictionary uses English that is not in the Oxford English Dictionary.

Where available, the Linnean or ‘scientific Latin’ name is given in parentheses, mostly for plants. This is not a classical Latin name, but a modern designation. Similarity of this designation to some Latin word may not be historically significant.

The spelling of the English meanings is US (plow not plough, color not colour, and English corn is rendered as grain or wheat), in spite of the fact that most of the Latin dictionaries that I have are British and use British spelling. The reason for this is (besides uniformity in the program) that there is much computer processing and checking of the dictionary data, including spell-checking of the English. (This is not to say that everything is correct, but it is much better than it would be without the computer checking.) All my programs speak US English, so I can count on it. Only some are available in UK English, and I do not have all of those versions.

Latin dictionaries seem to be locked into the 19th century. The English terms seem stilted, even by current British usage. This is probably because much work in translation was started then and later work tended to copy from the previous dictionaries. While this dictionary has done some modernization, some of the previous obscurities have been preserved. This was done in order that certain machine processes could compare the results of automatic translation with existing published work.

In addition, I have given US meanings to some terms that seem to be literally translated from the Latin (or German!) (a person who steals/drives off cattle is a rustler in the US).

Most dictionaries have an etymological approach, they are driven by the derivation of words to distinguish with separate entries words that may be identical in spelling but different derivations. But they can lump entirely different, even contradictory, meanings in a single entry if there is some common derivation. Philosophically, this dictionary is usually not sensitive to derivations, but sometimes supports multiple entries for vastly different meanings, application areas, or eras.

In a very small number of cases a source, such as OLD, will have an entry for which no English meaning is provided. Instead, a few words of Latin text containing the word is given. If they cannot figure it out, I certainly cannot. Such a source entry is usually omitted from WORDS.

Proper Names

Only a very few proper names are included, many just for test purposes, others that users have requested. The number of proper names is almost limitless but very few are applicable to a particular document, and if it is an obscure document it is unlikely that the names would be found in any dictionary.

Meaning for proper names may cite a likely example of a person with that name. This is just an example; there are lots of others with that name.

There is a switch (defaulted to Yes) that allows the program to assume that any capitalized unknown word is a proper name, and to ignore it. Also, one can make up a local dictionary of names for one’s particular application.

Letter Conventions (u/v, i/j, w)

U and/or V

Strictly speaking, Latin did not have a V, just a consonant U, or a U character that was easier in capitals (the way Latin was written by the Romans) to write or chisel in stone as V. However, many modern texts and dictionaries (with the important exception of the OLD) make the distinction with two characters (u and v). It appeared most appropriate in a computer context (never destroy information) to make the distinction and follow the common practice. So all dictionary entries maintain the V/v. However, an input word following the U convention will be found. At an earlier version, an algorithm was kludged to convert where necessary. While this worked in most cases, there were difficulties. The present system processes the dictionary and the input word as though U and V were the same letter, although the basic dictionary maintains the distinction and the output reflects this. There is no need for the user to set modes for this process.

I and/or J

A similar situation arises with I, and its consonant form, J. In this instance, the common practice is use only I, but there are many counter-examples, both text and dictionaries. (Lewis+Short uses J, but OLD does not.) Because of common practice, the program started out as pure-I dictionary with conversion of J-to-I on input. It remained that way through many versions, in spite of the logical inconsistency with U-V. The technique worked perfectly, but eventually the aesthetic of consistency won out and the U/V technique described above was extended to I/J.


While the letter W does not exist in classical Latin, there are examples of W in medieval Latin. I have not directly faced this, and have few words in the dictionary yet with W. The W problem is not analogous to U/V. While W sometimes could correspond to V or UU, in most cases it is a valid letter, reflecting a Germanic origin of the word. It will be treated as a real letter, and tricks employed as useful.