According to Schmauder (1991) they had a narrow range of frequency of occurrence (our nonlemma count yields an average of 68; Francis & Kucera, 1982).
1995, Peter John Gentry, The Asterisked Materials in the Greek Job, Scholars Press, ISSN 1044‒6761, →ISBN (10), →ISBN (13), 81:
Since the wording of the lemma differs from the non-lemma materials they cannot both beθʹ. In 18:15a the only difference is that the lemma has [Greek] for [Hebrew] in the parent text while the non-lemma has [Greek].
1996, David L. Blank, Ammonius: On Aristotle On Interpretation 1–8, Bloomsbury (first paperback edition, 2014), →ISBN, “Index of Passages Cited”, 195/1:
1997, Michael Cahill, Scriptores Celtigenae: Pars II (Corpus Christianorum: Series Latina, volume LXXXII), Brepols, 80:
That this is to be rated as an error is confirmed by the fact that the same phrase from Mark 1,15 is quoted as “credite euangelio” at the very beginning of the Commentary in a non-lemma setting (1,5); “cum uenient” for “conueniunt” (Mark 7,1) – a type of error found more than once in A; “adueniens” for “et ueniens” (Mark 9,13); “exiuit” for “exi” (Mark 9,24); “adprofiscente” for “et profiscente” […]
Their knowledge scores were higher; this was also true in a later timed experiment, in which 12 Lemma programmers were compared with nine non-Lemma programmers.
Therefore, each call to a non-lemma procedure and each return from a non-lemma procedure either increase the size of the stack or increase the number of lemmas Let N be the sum of the stack size and twice the number of lemmas. Then N increases by 1 whenever a non-lemma procedure is called or returns successfully.
We identified five error sources: […]Non-lemmas: The word supplied by the lexical transfer module is an inflected word and not a lemma; for that reason, it is not possible to use it in generation.
Occasionally, non-lemma “capitals at the beginning of sentences … have a daub of red”.
2004, Walid A. Saleh, The Formation of the Classical Tafsīr Tradition: The Qurʾān Commentary of Al-Thaʿlabī (d. 427/1035), Koninklijke Brill, ISSN 1567‒2808, →ISBN, chapter vi, § F, 201:
Al-Ṭabarī did not interrupt the flow of the commentary proper to discuss any non-lemma related matter.
For example, phrase translation modelp(e|f)can be calculated as, [¶] p(e|f) = ɑ₁pₗ(e|f) + ɑ₂pₙₗ(e|f) [¶] wherepₗ(e|f)andpₙₗ(e|f)is the phrase translation models corresponding to the lemmatization system and non-lemma system.ɑ₁ + ɑ₂ = 1.
The lemma ID in a relation may also be a negative number, in which case it represents a mapping to a non-lemma slot value – it would correspond to a value like that, to, etc.
2009, Vincent Vandeghinste, “Scaling up a hybrid MT system: From low to full resources” in Evaluation of Translation Technology (Linguistica Antverpiensia New Series – Themes in Translation Studies VIII), eds. Walter Daelemans and Véronique Hoste, VUB Press, →ISBN, part I, chapter iii, § 3.1, 71:
Part-of-Speech Tag mapping rules which convert the source language tags (Van Eynde, 2005) into target language tags are used to translate the non-lemma features of the source language tags (singular vs. plural, present vs. past, etc) into features of the target language tag (for instance, the Dutch tag the Dutch tag WW(pv,tgw,ev) is converted into VVB).
2010, Rufus H. Gouws, “Fixed word combinations as second level treatment units in dictionaries” in Feste Wortverbindungen und Lexikographie: Kolloquium zur Lexikographie und Wörterbuchforschung, ed. Peter Ďurčo, Walter de Gruyter GmbH & Co. KG, ISSN 0175‒9264, →ISBN, e-→ISBN, chapter vi, § 3.1, 52:
Access to the treatment of the lemma does not initially proceed via the lemma sign but via a non-lemma reduced guiding element, i.e. a lemma part in the lemma extemal article entrance position.
Initialize lemmatizer by providing dictionary file. If unknown, leave as non lemmata not found.
2011, Bart Jacobs et al., “VeriFast: A Powerful, Sound, Predictable, Fast Verifier for C and Java” in NASA Formal Methods: Third International Symposium, NFM 2011, Pasadena, CA, USA, April 2011 – Proceedings, eds. Mihaela Bobaru et al., Springer-Verlag Berlin–Heidelberg, ISSN 0302‒9743, e-ISSN 1611‒3349, →ISBN, e-→ISBN, part II, chapter iv, § 4, pages 52–53:
A simple approach to ensure termination of lemma functions in the presence of lemma function pointers would be to allow lemma function pointer calls only in non-lemma functions.
Non-lemma functions may produce lemma function pointer chunks arbitrarily.
2011, Karin Harbusch and István Bátori, “Clausal Coordinate Ellipsis (CCE) in Hungarian compared to CCE in Dutch, German, and Estonian” in Approaches to Hungarian, volume XIII: Papers from the 2011 Lund Conference, eds. Johan Brandtler et al., John Benjamins Publishing Company (2013), →ISBN (hardback), e-→ISBN, chapter iii, § 3.1, 57:
Elision of the wordform-identical Objectdas Fahrrad‘the bike’ is “blocked” by the preceding non-lemma-identical Modifiers (‘expertly’ vs. ‘diligently’).
Morphological filtering limits the intrinsic lexical over-generation of the guessers and is always used. Very often non-words or non-lemmas are generated as potential derivative bases, especially for input lemmas that are not derivatives.
In step 3, non-words or non-lemmas, that are often generated by the guesser modules, are filtered out from the result. Morphological filtering limits the intrinsic lexical over-generation of the guessers. Very often non-words or non-lemmas are generated as potential derivative bases, especially for input lemmas that are not derivatives.