Hi Ada,
Of course what counts as a morpheme or as a lexical expression, and what inventory of compositional rules one assumes, is subject to one's theory, to language change, and also to ad-hoc playful reinterpretation (that's what I would see your "s or p?" example as). But these are notions that need not be 100% precise in order to delineate a research area such as MWE. There can be a gray area as to what counts as an MWE and what doesn't. For example, many MWE researchers would probably not count a prefix verb like *beguile* as an MWE, simply because it fulfills all the criteria of wordhood in traditional Western NLP. But if we assume a wide definition such as "lexical expression consisting of more than morpheme", it too would fall under the MWE label. In fact, it exhibits the same competition between a lexical/idiomatic reading and a compositional reading that is typical of more complex MWEs: "kick the bucket" could mean to kick the bucket or to die, "beguile" could mean to affect with guile or to deceive.
I would support a name change from MWE to CLE or similar, because I agree that "word" is not a very useful notion cross-linguistically. (Then again, the notion of MWE might still work okay if we assume Martin Haspelmath's retro-definition https://dlc.hypotheses.org/2621 of "word".)
Cheers, Kilian
Am Sa., 11. Feb. 2023 um 18:29 Uhr schrieb Ada Wan adawan919@gmail.com:
Hi Archna
Thanks for your reply.
Your justification of the continual usage of "MWEs"/"words" is based on history and shared understanding (from 09Feb2023: "since the term has been used for a long while, there is a bit of a shared understanding of this term, including about these stipulations"), both of these criteria are achievable with alternate formulations.
Re "the category of items, of which idioms is a subset, has been referred to as multiwords for a long time": "MWE" does not have that long of a history --- what is the earliest use of "MWEs" that you have in your records? And even if terms have been used for a long while, it doesn't mean that we cannot change them for the better, esp. when they have been inappropriately adopted or found outdated. What objections do you have with "lexical expressions", for example?
The issue/problem with "word" is that, aside from it not being necessary or sufficient in the study of language or in computing, there is also an implicit, shared understanding that it is arbitrary, redundant, and indeterminate. (This applies also to the notion of wordhood within one language.) The indeterminacy part is evident in your not having provided me with a definition of "words" thus far as well. Furthermore, as you confirmed earlier: "the notion of wordhood may not be applicable to every single language and in the same way", then how should "words" be robust enough for computational processing?
Re emojis: here are some examples of emoji combinations that show a sense of idiosyncrasy when they (co-)occur: 🤩 for "star-struck" (from https://unicode.org/emoji/charts/full-emoji-list.html) Or from from https://www.elitedaily.com/lifestyle/funny-emoji-combinations-tiktok: 👉 👈 (feeling shy/simping) 🚪🏃♀️💨 (time to leave) 🍿🤏😯 (when drama is happening/when something is going down) 👁👄👁 (blank stare) 🕳👨🦯 (I didn't see anything) 👩🤏👩🦲 (wig snatched) 🐂💩 (bullsh*t)
My concern is on "wordhood" in the "language space" (science/engineering/technology) in general, not just on lexical expressions. I do think, however, that SIGLEX could help play an important role in effecting some positive changes in this regard.
Hi Kilian
Let's suppose that what we have thus far known as "grammar" (the one that has been based on or related to "words" or "sentences", i.e. morphology/syntax (and some phonology)) can be decomposed into (sequential) ordering and linguistic attitudes/normativity [1]. I do think judgments/attitudes play a role in language as it exists in the social world and can affect, or even determine, how registers/styles etc. are defined, but I also think that there is more rigorous science of (the remaining aspects of) language possible if we were to separate such attitudes/prescriptivism from a more descriptive stance (e.g. in the direction of information sciences).
Once we remove the attitudes/normativity part from the science of language, lexical and contextual information as well as function/use remain.
The reason why I hesitated in referring to MWEs as "complex" is because (lexical) "complexity" can be broken down into vocabulary and length, with use/frequency accounting for pragmatic/functional one. Hence every expression (or any character string) is lexical. The element of idiosyncrasy/idiomaticity is really a pragmatic one (e.g. in the rarity/archaic-ness/uniqueness of the use of the expressions/segment/span or character n-grams). So "sing" can be seen as a lexical expression, just like "bing" or "ping". Let's not forget that (even according to traditional grammatical analyses) various linguistic effects can happen to expressions when they undergo frequent use over an extended period of time. E.g. "ping me" may be seen thus far as relatively more idiomatic than "sing me a song", but that's due to the former expression being more specialized, less general, or rarer in use. Also, e.g. in a conversation, if one said "sing me" and the other didn't quite catch the first bit of the phrase, they might ask "[s] or [p]?" or "'s' or 'p'?". And one can well imagine that if this becomes in use more frequently, "s" and "p" can be regarded as what we'd now interpret as "idiomatic". Hence "sing" does not have to be seen as a "single morpheme".
[1] I have tweeted this before on 28Jan2023: https://twitter.com/adawan919/status/1619401653962297344?cxt=HHwWgMDS0a3Oovk... In a way, I am reinterpreting "(non)-compositionality" as normalization/frequency effects via the decomposed view of "grammar" above.
*Hence, my proposal (not just for MWE workshop folks but perhaps for all who might be interested) would be: https://docs.google.com/document/d/1n4QRn0CxbVMj6kbLWo-byT3S26ODJOjicU-ZYw7j... https://docs.google.com/document/d/1n4QRn0CxbVMj6kbLWo-byT3S26ODJOjicU-ZYw7jRG0/edit?usp=sharing*
*Comments welcome. * Thanks and best Ada
On Sat, Feb 11, 2023 at 2:01 PM Kilian Evang kilian.evang@gmail.com wrote:
Hi Ada,
The problem I have with the term "expression" without further qualification is that to my mind it includes any kind of linguistic sign, including ones like "to pay a visit to my dear aunt Ruth" which can clearly be interpreted compositionally. So I think we do have to specify "lexical" to delineate what we are studying in the MWE community. "Lexical item" or, sure, "lexical expression". Either would also include signs, of course. I do also feel we have to add "complex" or similar, because otherwise it includes single-morpheme lexical expressions like "sing".
Cheers, Kilian
Am Fr., 10. Feb. 2023 um 23:32 Uhr schrieb Ada Wan adawan919@gmail.com:
Hi Archna
"Idioms"/"Idiomatic expressions" are established terms in the study of language [1], with a longer history than MWE [2]. "Fixed", e.g. in "fixed phrases", is mentioned in, inter alia, [3], which was the earliest cite from the earliest work on MWEs in the ACL Anthology [4]. If I understand correctly, "MWEs" was a term so coined in order to establish a practice based on "words" (if anyone should view this differently, please do correct me here).
You're right, the task I suggested can be seen as orthogonal to distinguishing between lexical expressions or non-lexical expressions. I think it's important to document also the contexts surrounding expressions, instead of just picking expressions out and studying them in an isolated manner. It was just a suggestion for those who might be interested in building a multilingual parallel lexical database as well as those who might want to get a more holistic understanding of language while weaning oneself of "words" --- now that it's become even more obvious how superfluous the term/concept is.
[1] See e.g. https://en.wikipedia.org/wiki/Phraseme [2] "Idiomatic expression" is just another formulation of "idiom" (see https://www.thefreedictionary.com/idiomatic+expression). According to Collins English Dictionary (accessed via https://www.thefreedictionary.com/idiom), "idiom" stems from the 16th century Latin idiōma, denoting "pecularity of language". [3] Nunberg, Geoffrey, Ivan A. Sag, and Thomas Wasow. 1994. Idioms. Language, 70:491–538. https://doi.org/10.2307/416483 (Many older references on "idioms" by linguists can be found therein.) [4] Ann Copestake, Fabre Lambeau, Aline Villavicencio, Francis Bond, Timothy Baldwin, Ivan A. Sag, and Dan Flickinger. 2002. Multiword expressions: linguistic precision and reusability. In Proceedings of the Third International Conference on Language Resources and Evaluation (LREC’02), Las Palmas, Canary Islands - Spain. European Language Resources Association (ELRA).
Hi Kilian
Sorry about my oversight on "item". I do think "item" could be better than "term" in this case, but it does carry a sense of "a single element", a more discrete "singleton". It's ok to combine it with "complex" to mitigate the sense of "singleton", but then "complex" as you suggested is dependent on morphology, which can be problematic.
Re "lexical": sure. (I think there have been so many different views/traditions/conventions among linguists and computational linguists in the past, we don't necessarily have to agree on how we or our definitions/methods might differ or might have differed, as long as we have the same goal now?)
One argument for "expressions" would be that they could include a sign (e.g. hand sign in motion).
So how about updating "MWEs" to: i. "lexical expressions", or ii. "lexical expressions (of one character or more when written)*", or iii. [i] or [ii] without "lexical", or iv. others?
- I'm trying to incorporate how expressions with emojis would/should be
treated too.
What do you all think?
Thanks and best Ada
On Fri, Feb 10, 2023 at 10:58 AM Kilian Evang via Corpora < corpora@list.elra.info> wrote:
Forwarded message from Archna below
---------- Forwarded message --------- Von: Archna Bhatia abhatia@ihmc.org Date: Do., 9. Feb. 2023 um 19:58 Uhr Subject: Re: [Corpora-List] Deadline extension: 19th Workshop on Multiword Expressions (MWE 2023) To: Ada Wan adawan919@gmail.com, kilian Evang <kilian.evang@gmail.com
Cc: Mike Scott mike@lexically.net, mweworkshop2023@googlegroups.com < mweworkshop2023@googlegroups.com>, corpora@list.elra.info < corpora@list.elra.info>
Thanks, Ada. I think using the terms “fixed” and “idiomatic” make the category appear more restrictive, and would need qualifications such as “fixed” is a relative term here, etc. With “multiwords/multiword expressions” also, there are stipulations (the notion of wordhood may not be applicable to every single language and in the same way) but since the term has been used for a long while, there is a bit of a shared understanding of this term, including about these stipulations. I am open to better terminology. Using just “expressions”, however, seems too vague and loses some generalizations about the idiosyncrasies that "multiword expressions” demonstrate. Every expression in not the same, “multiword expressions” show characteristics different from other expressions. I understand there is some fluidity also there when trying to distinguish between multiwords and non multiword expressions.
There are so many angles that one could look at language from. I don’t see anything wrong with the view that studies expressions covering all aspects as you suggest without distinguishing between expressions based on notions of wordhood. The task you suggest will help in developing understanding about language and how languages are similar or different and how they are used. I don’t think it disqualifies efforts that distinguish between “multiword expressions” and non-multiword expressions though, and the idiosyncrasies are not limited to morphology/syntax, idiosyncrasies are found in other linguistic aspects too when characterizing "multiword expressions”.
~ Archna
On Feb 9, 2023, at 11:17 AM, Ada Wan adawan919@gmail.com wrote:
Hi Archna, hi Kilian, hi all
Thanks for your replies.
TLDR on my part: I'd be fine going with "expressions" (instead of "fixed/idiomatic expressions"). Neither "word" nor "morphology/syntax" (apart from the ordering of elements and/or sequential patterns) is necessary in the analyses of such.
More specifically:
[@Archna] Re "fixed/idiomatic expressions": I don't think it matters much whether they are "fixed" or "idiomatic". A "fixed expression" is one that is usually more impervious to (lexical) change. One can measure this quality in a longitudinal study, e.g. in relation to other aspects of language change etc.. Re how "fixed" is "fixed": it's relative, much like many other aspects of language studies. By "idiomatic", one could mean that there is an element of idiosyncrasy (as "idiom"/"idioma").
The message that I am trying to get across is that "word" is a superflous category in the study of language. Would you mind please justifying why you need "words"?
The same goes for morphology, actually. In essence, morphological analyses involve selective decomposition, not decomposition of all decomposable units. Hence if one is only accounting for variations within an expression as a ((sub-)character) sequence involving "morphemes" (assuming definable rigorously) and discounting the changes in other parts of the sequence, that would be an incomplete analysis of the expression. Instead, one can just refer to expressions as "expressions", as e.g. sequences/strings of various lengths/vocabs in (sub-)characters --- such an account is also more flexible and accommodating to diverse languages/registers/modalities.
A study of "expressions" can cover all other aspects --- not just lexical but also functional ones. One doesn't need to incorporate/impose any ad hoc notions of "wordhood" in these studies.
Suggestion: I believe there are many more interesting tasks in this area, instead of trying to find/define "words" within expressions, or to "parse" them according to some structuralist assumptions (i.e. morphologically/syntactically). For example, the community could start (some multi-year project) building an international multilingual parallel (note: not everything would be parallelizable) database of all expressions and terminologies ever existed with contextual (historical/cultural/social) information and start verifying their sources and status of current use. (Just be aware, though, that one is not reinforcing values that shouldn't be further emphasized / transfered to posterity --- as an ethical consideration. So if something is in the grey area now, document clearly what the current attitudes towards a certain value are, so posterity can look back and evaluate with respect to their point of view.)
Counter questions to Archna: What are the motivations behind your suggestion to access/interpret language using "words"? How do you define "words" and justify the sufficiency/necessity of morphology/syntax in relation to the study of these expressions, esp. when the morphological decomposition of these expressions is arbitrary and helps little (or not at all) with explanation or prediction?
Re "complex lexical terms", @Kilian: I'm just wondering what kind of terms that would be considered "terms" that wouldn't be considered lexical (I was tempted to add "lexical" to "expressions" as well, but thought that might be a bit redundant)? It depends on how one defines "terms", of course. And how "complex" are expressions really? They are just more calcified units after all, aren't they? (Why do we/some always seem to want to add the term "complex" to everything? Things that aren't "complex" are also worthy of studying!)
Curious what you think...
Thanks and best Ada
Why I'm advocating #noWords: Fairness in Representation for Multilingual NLP: Insights from Controlled Experiments on Conditional Language Modeling https://openreview.net/forum?id=-llS6TiOew
https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fopenreview.net%2Fforum%3Fid%3D-llS6TiOew&data=05%7C01%7Cabhatia%40ihmc.org%7C3d437044e42f42c2c61408db0ab92ccb%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638115562691707319%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=jea7YNI7295cJ2CY0jwxrsjID7DcDqerqI3IQxj9hUc%3D&reserved=0 https://drive.google.com/file/d/1eKbhdZkPJ0HgU1RsGXGFBPGameWIVdt9/view
On Thu, Feb 9, 2023 at 3:27 PM Mike Scott via Corpora < corpora@list.elra.info> wrote:
I must say I'm perfectly happy with "multi-word expression", or "multi-word unit".
I feel sympathy with Archna's post (and incidentally wish Archna didn't have to go through a friend!) Cheers -- Mike
--
Mike Scottlexically.net https://nam02.safelinks.protection.outlook.com/?url=http%3A%2F%2Flexically.net%2F&data=05%7C01%7Cabhatia%40ihmc.org%7C3d437044e42f42c2c61408db0ab92ccb%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638115562691707319%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=lnpEPfv%2B4UmB1e0xVkC4hsIs%2B9GqwDnSzzMpwiFWZHw%3D&reserved=0 Lexical Analysis Software and Aston University
Corpora mailing list -- corpora@list.elra.info https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/ https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Flist.elra.info%2Fmailman3%2Fpostorius%2Flists%2Fcorpora.list.elra.info%2F&data=05%7C01%7Cabhatia%40ihmc.org%7C3d437044e42f42c2c61408db0ab92ccb%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638115562691707319%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=78A%2BL24tQ0GBhZ3lAGipq9tLPZU%2FmydmGBGX1yE4BSA%3D&reserved=0 To unsubscribe send an email to corpora-leave@list.elra.info
-- You received this message because you are subscribed to the Google Groups "MWE Workshop 2023 Organizers" group. To unsubscribe from this group and stop receiving emails from it, send an email to mweworkshop2023+unsubscribe@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/mweworkshop2023/CAB7Mis_GSyFjZOVw_XWp431VM... https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fmweworkshop2023%2FCAB7Mis_GSyFjZOVw_XWp431VMJJBo0BnPqjFsqqTP_sEE58Ezw%2540mail.gmail.com%3Futm_medium%3Demail%26utm_source%3Dfooter&data=05%7C01%7Cabhatia%40ihmc.org%7C3d437044e42f42c2c61408db0ab92ccb%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638115562691707319%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=FX%2BoXH1j9XL5X0tJuqc%2BfKdFkuugawZvrtzdXNUG2%2FA%3D&reserved=0 . For more options, visit https://groups.google.com/d/optout https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Foptout&data=05%7C01%7Cabhatia%40ihmc.org%7C3d437044e42f42c2c61408db0ab92ccb%7C2b38115bebad4aba9ea3b3779d8f4f43%7C1%7C0%7C638115562691707319%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=1SGOAvFNmKwsKKOx6Kc%2Fm1wHzDbm%2F4xiEge3RY5etrE%3D&reserved=0 .
-- Archna Bhatia, Ph.D. Research Scientist, Institute for Human & Machine Cognition 15 SE Osceola Ave, Ocala, FL 34471 (352) 387-3061
Corpora mailing list -- corpora@list.elra.info https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/ To unsubscribe send an email to corpora-leave@list.elra.info