Locating collocation

The Wikipedia entry on collocation says:
“..a collocation is a series of words or terms that co-occur more often than would be expected by chance”. 1
This is the description of collocation that is linked to by leaders in corpus tools SketchEngine in their syllabus for a new online course.2

Note the statistical aspect in the definition “more often than would be expected by chance”.

The wiki entry then reads “There are about six main types of collocations: adjective + noun, noun + noun (such as collective nouns), verb + noun, adverb + adjective, verbs + prepositional phrase (phrasal verbs), and verb + adverb. “

Note the emphasis on the grammar aspect of collocation.

Bill Louw would place this wiki definition (alongside Goran Kjellmer’s definition of collocation – ‘sequence of words that occurs more than once in identical form…and which is grammatically well structured’) at the bottom of the diagram below:

(Louw & Milojkovic 2016: 53)

The diagram shows two dimensions, the vertical dimension is how restrictive a view of collocation is with the most restrictive at the bottom and the least at the top. The horizontal dimension shows how much of the language a view of collocation covers, the top bulb of the diagram is larger than the bottom bulb.

Louw & Milojkovic (2016) argue that the link of collocation to context of situation is of great importance in applications of corpora in literature studies i.e. corpus stylistics.

Context of situation was illustrated by Firth in the following way:

“In his article ‘Personality and language in context’ Firth offers us what he calls a typical Cockney event in ‘one brief sentence’.
‘Ahng gunna gi’ wun fer Ber’. (I’m going to get one for Bert)
What is the minimum number of participants? Three? Four? Where might it happen? In a pub? Where is Bert? Outside? Or playing darts? What are the relevant objects? What is the effect of the sentence? ‘Obvious!’ you say. So is the convenience of the schematic construct called ‘context of situation’. It makes sure of the sociological component.” (Firth 1957: 182 as quoted in Louw & Milojkovic, 2016:61, emphasis added)

Awareness of the importance of context of situation is reflected in the following small Twitter poll where a majority of the 24 respondents opted for “meanings have words” over “words have meanings”:

Twitter poll

Although Louw concedes a view of collocation such as ngrams can reveal contexts of situation, opportunities to do so will be much rarer than if collocation is located near the top of the diagram – “abstracted at the level of syntax” as Firth put it.

Context of situation is also of great importance in language teaching and learning. For example task based teaching can be said to lay great weight on context of situation.

As Louw & Milojkovic (2016:26) put it :

“The closer collocation’s classifications are to context of situation, the more successful and enduring will be the approach of the scholars who placed them there. The more the term is constrained by the notion of language ‘levels’ and the linearity and other constraints of syntax, the less such classifications and the theories perched upon them are likely to endure. The reason for this is, as we shall see, that collocation takes us directly to situational meaning and acts as what Sinclair refers to as the ‘control mechanism’ for meaning”

Thanks for reading.

Notes

  1. Wikipedia Collocation https://en.wikipedia.org/wiki/Collocation
  2. Boot Camp online https://www.sketchengine.eu/bootcamp/boot-camp-online/#toggle-id-2

References

Louw, B., & Milojkovic, M. (2016). Corpus stylistics as contextual prosodic theory and subtext (Vol. 23). John Benjamins Publishing Company.

Thin word lists and fat concordances

One of the aspects of the proposed changes in the GCSE modern foreign language, MFL, syllabus in the UK is the use of corpus derived word lists 1. Distribution of words when counted follow a power law. A common power law is Pareto in economics – “Pareto showed that approximately 80% of the land in Italy was owned by 20% of the population” 2 . Similarly in any piece of text a large percentage of it comes from a relatively small amount of words – the top 100 words in English accounts for 50% of any text. The MFL review wants to use wordlists of the most frequent 2000 words – which would cover about 80% of any text.

Currently the MFL syllabus is topic based, so one issue here is that most words one can use for any particular topic will be limited to that topic. Or another way to say it is that although the word may be frequent within a topic it won’t have range and appear in other topics. The NCELP, National Centre for Excellence for Language Pedagogy in Vocabulary lists: Rationales and Uses writes “For example, many of the words for pets or hobbies will be low frequency words which are not useful beyond those particular topics. ” 3

There have been many critics of this wordlist driven proposal who have pointed out various weaknesses, see – AQA Exam board 4, ASCL, Association of School and College Leaders 5, Transform MFL 6 , Linguistics in MFL Project 7.

I want to take a different tack and argue that the wordlist driven approach is a half-hearted version of what could be a full blooded corpus approach to vocabulary content.

Corpus stylist Bill Louw writes that he “has become suspicious of decontextualised frequency lists” (Louw & Milojkovi, 2016:32). He calls such lists thin lists because they tend to cover things rather than events (Louw 2010). Events are states of affairs, what one of the originaters of the notion of meaning by collocation JR Firth has called context of situations. Looking at collocates of things in concordance lines allows us to “chunk the context of situation and culture into facts” (Louw 2010).

A concordance line brings together and displays instances of use of a particular word from the widely disparate contexts in which it occurs. To cover events one would need to examine collocates in concordances hence the term fat concordances.

The most frequent words are often bleached out of their literal meanings. Compare the word “take” on its own, most people would think of the meanings of “the act of receiving, picking up or even stealing” (Louw & Milojkovi, 2016:5), to a collocation such as “take place”, we see that the meaning here is distant from the literal meaning of “take” 8. When the NCELP say “Very high frequency words often have multiple meanings.” they are describing the notion of delexicalisation.

To demonstrate context of situation and context of culture, reproduced below is corpus linguist John Sinclair’s PhraseBite pamphlet which is reproduced in Louw (2008):

When she was- – – – – Phrasebite© John Sinclair, 2006.

  1. The first grammatical collocate of when is she
  2. The first grammatical collocate of when she is was
  3. The vocabulary collocates of when she was are hair-raising. On the first page:
    diagnosed, pregnant, divorced, raped, assaulted, attacked
    The diagnoses are not good, the pregnancies are all problematic.
  4. Select one that looks neutral: approached
  5. Look at the concordance, first page.
  6. Nos 1, 4, 5, 8,10 are of unpleasant physical attacks
  7. Nos 2, 3, 6, 7, 9 are of excellent opportunities
  8. How can you tell the difference?
  9. the nasties are all of people out and about, while the nice ones are of people working somewhere.
  10. Get wider cotext and look at verb tenses in front of citation.
  11. In all the nasties the verb is past progressive, setting a foreground for the approach.
  12. In the nice ones, the verb is non-progressive, either simple past or past-in-past.

Data for para 4 above.
(1) walking in Burnfield Road , Mansewood , when she was approached by a man who grabbed her bag
(2) teamed up with her mother in business when she was approached by Neiman Marcus , the department store
(3) resolved itself after a few months , when she was approached by Breege Keenan , a nun who
(4) Bridge Road close to the Causeway Hospital when she was approached by three men who attacked her
(5) Drive , off Saughton Mains Street , when she was approached by a man . He began talking the original
(6) film of The Stepford Wives when she was approached by producer Scott Rudin to star as
(7) bony. ‘ ‘ Kidd was just 15 when she was approached to be a model . Posing on
(8) near her home with an 11-year-old friend when she was approached by the fiend . The man
(9) finished a storming set of jazz standards when she was approached by SIR SEAN CONNERY . And she
(10) on Douglas Street in Cork city centre when she was approached by the pervert . The man persuaded

As Louw (2008) puts it:

“The power of this publication, coming as it did so close to Sinclair’s death, is to be found in the detail of his method. By beginning with a single word, she, from the whole of the Bank of English, Sinclair simply requests the most frequent collocate from the Bank of English (approximately 500 million words of running text). The computer provides it: when. The results are then merged: when+she. A new search is initiated for the most frequent collocate of this two-word phrase. The computer provides it: was. The concordances are scrutinized and cultural insights are gathered.”

The ASCL quotes applied linguist Vivian Cook:

“While word frequency has some relevance to teaching, other factors are also important, such as the ease with which the meaning of an item can be demonstrated (’blue’ is easier to explain than ‘local’) and its appropriateness for what pupils want to say (‘plane’ is more useful than ‘system’ if you want to travel)”

Blue is easier to explain than local because most collocates of blue are its literal colour meaning e.g. “blue eyes”. Yet consider this from a children’s corpus:

“There, I feel better. I’ve been needing a good cry for some time, and
now I shall be all right. Never mind it, Polly, I’m nervous and tired;
I’ve danced too much lately, and dyspepsia makes me blue;” and Fanny
wiped her eyes and laughed.” (An Old-fashioned Girl, by Louisa May Alcott)

So while it is true that blue is often associated with color, it also associates with mental states where the colour meaning is delexicalised, or washed out.

To conclude, the MFL proposals on using corpus derived word lists to drive content is not taking full advantage of corpora. They are promoting thin wordlists when they could also be promoting fat concordances.

Thanks for reading.

Notes

  1. MFL consultation – https://consult.education.gov.uk/ebacc-and-arts-and-humanities-team/gcse-mfl-subject-content-review/supporting_documents/GCSE%20MFL%20subject%20content%20consultation.pdf
  2. Pareto – https://en.wikipedia.org/wiki/Pareto_principle
  3. NCELP – https://resources.ncelp.org/concern/resources/t722h880z?locale=en)
  4. AQA – https://filestore.aqa.org.uk/content/our-standards/AQA-GCSE-MFL-POLICY-BRIEFING-APRIL-2021.PDF
  5. ASCL – https://www.ascl.org.uk/ASCL/media/ASCL/Our%20view/Consultation%20responses/2021/Draft-response-Consultation-on-the-Revised-Subject-Content-for-GCSE-Modern-Foreign-Languages.pdf
  6. Transform MFl – https://transformmfl.wordpress.com/2021/02/15/should-we-learn-words-in-frequency-order/
  7. Linguistics in MFL Project – http://www.meits.org/opinion-articles/article/the-dfe-ofqual-consultation-on-revised-gcse-qualifications-in-modern-foreign-languages-a-view-from-linguistics
  8. Take place – https://eflnotes.wordpress.com/2013/05/06/what-to-teach-from-corpora-output-frequency-and-transparency/

References

Louw, B. (2008). Consolidating empirical method in data-assisted stylistics. Directions in Empirical Literary Studies: In Honor of Willie Van Peer, 5, 243.

Louw, B. (2010). Collocation as instrumentation for meaning: a scientific fact. In Literary education and digital learning: methods and technologies for humanities studies (pp. 79-101). IGI Global.

Louw, B., & Milojkovic, M. (2016). Corpus stylistics as contextual prosodic theory and subtext (Vol. 23). John Benjamins Publishing Company.

Why the pineapple?

This post can be considered a follow on from the post Collocations need not be arbitrary.

One response that proponents of the lexical approach in language teaching could make to the issue of looking at meanings and collocations is simply to define collocation as one level of meaning. John Firth, as cited by Joseph (2003), put it thus:

“The statement of meaning by collocation and various collocabilities does not involve the definition of word meaning by means of further sentences in shifted terms. Meaning by collocation is an abstraction at the syntagmatic level and is not directly concerned with the conceptual or idea approach to the meaning of words. One of the meanings of night is its collocability with dark, and of dark, of course, collocation with night.”

Joseph, 2003: 130

Defining collocations as one level of meaning is reasonable but it does not provide an explanation that may be pedagogically useful. Cognitive linguistics claims to provide such a use.

Let’s take the question of the difference between choosing highest mountain and tallest mountain that arose in a class recently. One explanation is based on the distribution of what collocates with tall – that is living things (tall man, tall tree) and man made objects (tall building, tall pole). Tall tends not to collocate with natural objects such as mountains.

That is where a Firthian (and by consequence a lexical) approach stops. A cognitive analysis by Dirven and Taylor (1988) showed that general cognition (in the form of concepts) can explain further.

Highest mountain is preferred as the concept HIGH includes both a meaning of vertical position (positional meaning) as well as vertical length (extensional meaning) whereas the concept TALL only includes the meaning of vertical length. So although you can find tallest mountain people often think of being at the top of a mountain hence the vertical position is emphasised rather than vertical length (see figure below):

Figure 1. after Dirven & Taylor, 1988: 386

Thanks for reading. And do have a read of a less favourable view of cognitive linguistics at a recent Geoff Jordan blog Anybody seen a pineapple?

Update

Marc Jones writes about cueing as a way to learn chunks Pinneapples?

References

Dirven, R., & Taylor, J. R. (1988). The conceptualisation of vertical space in English: The case of tall. In Topics in cognitive linguistics, B. Rudzka-Ostyn (ed), 379. John Benjamins.

Joseph, J. (2003). Rethinking linguistic creativity. In Rethinking Linguistics, H. Davis & T.J. Taylor (eds), 121–150. London: Routledge.

Collocations need not be arbitrary

“On the whole, delexicalized verbs are a good way of introducing the concept of collocation to learners of any L1 background. I usually start with make/do and show how one goes with homework while the other goes with mistake (I did my homework; I made a lot of mistakes). Why is it this way and not the other way around? Because words have collocations – they prefer the company of certain other words.(Selivan, 2018:28, emphasis added)

The quote above, from a book published in 2018, reflects a pervasive view in the literature that collocations are arbitrary, that is, there is no particular reason why words “prefer the company of certain other words”, they just do.

Liu (2010) identifies this view of collocation-as-arbitrary as wide-spread amongst scholars, he also demonstrated that it is a common assumption in published teaching materials. Of the books, studies and websites on teaching collocations he observed collocation exercises as mainly noticing and memorising fixed units or in other words form focused exercises.

Example of such exercises are:

“identifying or marking collocations in a passage or in collocation dictionaries; reading passages with collocations highlighted or marked; filling in the blanks with the right word in a collocation; choosing or matching correct collocates; translating collocations from L2 back into L1 or vice versa; and memorization-type activities like repetition and rehearsal” (Liu, 2010:21)

There were fewer exercises on linking collocation forms to their meanings.

In addition to overlooking the motivated aspects of collocations, learners also miss the chance to generalise what they learn (Wray, 2000). That is, collocations also need to be analysed if students are to make the most of them in new situations of use.

To take the examples of “make” and “do”, the core meaning of “make” is create, which is a process that is purposeful and/or more effortful than the core meaning of “do” of completion/the finishing of something, which focuses on the end result of an activity rather than on any effort in the process of that activity. Understanding these core meanings can throw light on the following use of “did a mistake”:

“But I did a mistake in talking about it, you know, the last time and recently”

The larger context of this is from a spoken news report:

weren’t there. Let me handle it. I said, ” Yes, ma’am. ” ROSEN: The rebuke of Mr. Clinton by his wife came after the former president revived the dormant issue of Mrs. Clinton’s own misstatements about her 1996 trip to Bosnia. You’ll recall Mrs. Clinton, in recent months, spoke of sniper fire jeopardizing her landing. But contemporaneous video and eyewitness account revealed there was no such threat, and the senator effectively if belatedly defused the story with an omission of error in late March. SEN-HILLARY-CLINTO: But I did a mistake in talking about it, you know, the last time and recently. ROSEN: But in Jasper, Indiana, Thursday, Mr. Clinton blamed the controversy on the biased news media. B-CLINTON: She took a terrible beating in the press for a few days because she was exhausted at 11:00 at night when she started talking about Bosnia. ROSEN: In fact, Mrs. Clinton related the false Bosnia story numerous times including in a prepared speech delivered freshly at mid morning. B-CLINTON: And then the president (COCA SPOK: FOX SPECIAL REPORT WITH BRIT HUME 6:00 PM EST, 2008, emphasis added)

We could speculate that in using “did a mistake” Hilary Clinton was implying that in her “exhausted” state the “misstatement” was the opposite of a purposeful lie. It was just one of many activities she did that day which happened to be an error.

This can also be seen in another example from COCA – “If I do a mistake, I’m cooked”.

The context is from a written publication this time, although the language in question is in reported form:

three minutes, sometimes the whole roll — eleven minutes. It has an advantage: It takes you to the real tempo of life. Most movies are shot rather quickly and in a way where you can manipulate your reality because of the amount of coverage ” — shooting a scene from many different angles so that the director can choose among them in the editing room. ” Here my manipulation is quite different. I have to build it in with the lighting, with the framing. It requires much more attention at this stage. If I do a mistake, I’m cooked, ” he says with a laugh. # Wings’ visual style may be old-fashioned at heart, but its sound is high-tech all the way. Besides the six channels of top-notch stereo sound broadcast through the theater speakers, Wings audiences will hear two channels of three-dimensional sound through a special headset called the Personal Sound Environment (PSE) distributed to each moviegoer. Developed by Imax affiliate Sonics Associates of Birmingham, Alabama, the PSE incorporates both IMAX 3-D glasses and tiny speakers mounted between (COCA MAG: Omni, 1994, emphasis added)

The person is talking about a number of steps in their work routine in shooting a movie. The use of “do” here is to signal that any disastrous mistake is not to be blamed on the person considering all the other things he has to juggle.

Note that I could only find 3 uses of “do a mistake”, of which 2 are shown here (the third one I can’t offer any speculation on as I suspect more context needs to be chased up than that provided by COCA).

This blog was inspired by a question from a student about why a text had “in many respects” rather than “in many aspects”. I went onto COCA to have a look but could not discern any useful explanation. I just told the student that “aspects” does not seem to prefer “in many” compared to “respects”! Only later when I thought about the root word in common “spect” (meaning see) did a arguably useful explanation present itself – “in many respects” implies that the [re-seeings] have already been understood in some way. While “in many aspects” the reader may not yet know what these [partial-seeings] may be. These meanings could match up with the observation that “in many respects” often comes at the end of a clause or sentence while “in many aspects” may tend to come at the beginning of a clause or sentence.

Thanks for reading.

References:

Davies, M. (2008). Corpus of contemporary American English online. Retrieved from https://www.english-corpora.org/coca/.

Liu, D. (2010). Going beyond patterns: Involving cognitive analysis in the learning of collocations. TESOL Quarterly, 44(1), 4-30.

Selivan, L. (2018). Lexical Grammar: Activities for Teaching Chunks and Exploring Patterns. Cambridge University Press.

Wray, A. (2000). Formulaic sequences in second language teaching: Principle and practice. Applied linguistics, 21(4), 463-489.

Awareage Languness

This is a post to recount my recent attempts or (more pretentiously) my ongoing journey to understand how linguistics can help language teaching. The post was initiated by a video teaser (on linguistics and teacher education) to a talk by Bridget R. Schvarcz in the upcoming TESOL France 2018 colloquium.

I will include various prompts which led me to my current language (or communicative) awareness location as a way to illustrate my route.

My first prompt would have to be my CELTA training of 4 weeks of which a few days I vaguely remember being devoted to orthodox segregational linguistics that language can be compartmentalised into things such as parts of speech, subject verb object sentences and various other related grammar. I did not re-evaluate this basis for some time after my initial TEFL training.

Fast forward a few years to a point after many classroom incidents that kept showing me the inadequacies of my knowledge about the English language and its usefulness in helping my students. I had a vague idea of communicative language teaching or CLT (ignoring for now its basis in the segregational linguistics of speech act theory) – but marrying CLT up with my grammar knowledge was tortuous to say the least (hands up who has not tried to and/or is mandated to shoehorn grammar points into a lesson?!).

Integrational linguistics can be summed in three words – language presupposes communication.  My route to it was from the notion of meaning invariance. As teachers of language one form one meaning is obviously appealing – e.g. adding an -s to cat means there is more than one cat.

My next prompt was being aware of Columbia School Linguistics and its focus on analysing invariant meanings. For example some/any. One of the scholars associated with Columbia School Linguistics is Ricardo Otheguy whose talk here led me to discover that there exists another line of thought (held by people such as Roy Harris, Sinfree Makoni, Alastair Pennycook) which questions the validity of invariance in language.

Integrational linguistics holds signs (meaning makers) are radically indeterminate – that is both form and meaning are not fixed codes which can be plucked and used – signs do not pre-exist acts of communication but are made in the act of communication. Hence they are not determined before the act of communication itself.

Now to a language teacher this seems most unhelpful. If form-meaning pairs do not pre-exist acts of communication then how do we teach them?

Fear not, we don’t need such codes to start thinking about planning lessons. We can focus on the acts of communication themselves. Here we see echoes of approaches such as task based language teaching, comprehensible input and total physical response.

This whiggish history of my language awareness, my awareage linguness, if you will, pardon my poor joke, hopes to have piqued you into maybe asking yourself some questions posed by Bridget R. Schvarcz in her teaser video:

What influenced who you are as a teacher today?

What comes to your mind when you hear the word linguistics?

Have you ever used any of the theoretical knowledge about language structure in your teaching?

Do you think you are a better teacher because you have studied linguistics?

If you are headed to TESOL France 2018 hope to bump into you, thanks for reading.

Explore some topics in applied linguistics

Thanks to a post by Jason Anderson I read a paper called – Research trends in applied linguistics from 2005 to 2016: A bibliometric analysis and its implications.

The authors kindly sent me a file of the abstracts that they had collected. I thought some topic modelling would be interesting to do on the data. Topic modelling is a way to discern what a set of documents is “about” by getting a program to find clusters of words. Lei & Liu (2018) used a different approach called n-grams to find their topics.

They found for example that between 2005-2016 there was a significant decrease in formal linguistic issues, such as phonology and syntax.

The topic modelling also shows this decrease (note that the corpus used in the topic modelling runs from 2000-2016):

By contrast they found significant increases in topics related to sociocultural issues. The topic modelling also indicates this:

The topic model correlational matrix is interesting to look at. The screenshot below shows that the topic “english chinese paper” (full topic cluster is “english chinese paper hong use varieties kong world local language”) is significantly related to the topic “language social identity” (full cluster is “language social identity how practices literacy languages policy linguistic multilingual”):

Though I am not sure how to interpret the red blobs! If you do let me know.

Finally the model indicates that topics related to child language development seem to be on the wane (full cluster is “children language age children’s development early study acquisition adults years”):

Have a play with the model and if you spot anything interesting do leave a comment. Note running iterations can be a tad slow.

Thanks for reading.

References:

Lei, L., & Liu, D. (2018). Research Trends in Applied Linguistics from 2005 to 2016: A Bibliometric Analysis and Its Implications. Applied Linguistics.

The paradox of re-usability in language materials

The title is adapted from a critique of learning objects (orginally defined as digital resources used to aid learning) in the field of instructional design by David Wiley. What follows is borrowed heavily/paraphrased from his writings.

Author Julie Moore raises intellectual rights and copyright issues with the idea of having editable materials from course books. However there is a deeper paradox in editable or re-usable material.

If we look at a typical unit in a coursebook it may have sections such as language focus and practice, input reading and/or listening and output speaking and/or writing all centered around the unit topic. We could describe this unit has having an internal context, that is the elements which make up the unit – instructions on how to use a language point, practice exercises on this language point, a picture that goes with the reading text, a role play that goes with a speaking activity etc. The more elements that are in the unit the larger the internal context of the unit.

External context would be the other units in the coursebook. A learning object is said to have no external context independent of its instructional use. That is external contexts exists for a learning object only for the purposes of some instructional procedure.

The number of external contexts in which a learning object will instructionally fit varies according to the internal context of the said object. An instructional fit is the effectiveness of the object.

A large object (i.e. one with many elements) has a greater internal context than a small object. Larger objects fit into fewer external contexts than smaller objects.

To restate this the fit of an object with other objects is a function of 1) its internal context and 2) its external context with other objects. The more internal context you have i.e. the more elements, the better will be the (pedagogical) effectiveness of the object. But the internal context is inversely related to the number of other objects in the external context. So the paradox is that the effectiveness of a learning object and its potential for reuse (i.e. to fit in with external context) are contradictory.

So you have a trade-off between effectiveness and re-usability. For editable materials the more you make it re-usable the less effective it will be pedagogically.

Now one could argue that learning objects is concerned with conceptual knowledge (e.g. teaching someone how to develop web pages) whereas language goes beyond the limits of such knowledge. Language avoids the re-usability paradox. It has both a lot of internal context (systems such as phonology, syntax) and a lot of external context (systems such as semantics, pragmatics).

However as the world of language course books currently exists it could be said to follow the path set by other books that deal with conceptual knowledge. If this is the case then the re-usability paradox applies.

Furthermore the paradox is due to the author rights issues covered by Julie Moore. In the copyright context “reuse” means more or less “use as exactly as is”. So is there a way out of the re-usability paradox?

As David Wiley puts it:

The way to escape from the Reusability Paradox is simply by using an open license. If I publish my educational materials using an open license, I can produce something deeply contextualized and highly effective for my local context AND give you permission to revise and remix it until it is equally effective to reuse in your own local context. Poof! The paradox disappears. I’ve produced something with a strong internal context which you have permission to make fit into other external contexts.

How likely are we to see open content from commercial parties judging by the state of current play in the ELT publishing world? Happily individual teachers and grassroots organizations are already thinking and working on this.

Thanks for reading.

ELTJam, machine learning, knowledge and skill

“Knowledge and skill are different. Vocabulary acquisition tools help learners improve their knowledge, which may in turn have a positive impact on skill, but it’s important to be cognisant of the differences.” [https://eltjam.com/machine-learning-summer-school-day-4/]

“We need to be careful though not to oversell the technology and be clear about what it can and can’t do. There is no silver bullet. This is especially the case when it comes to skills vs knowledge; a lot of the applications that could come from this sort of technology will help improve knowledge of English, and may contribute to accuracy” [https://eltjam.com/machine-learning-summer-school-day-5/]

The above two quotes are from a nice series of posts by ELTJam on a machine learning workshop. The first point from the first quote is indeed important to recognize. Bill VanPatten (2010) has argued that knowledge and skill are different. However what is meant by knowledge and what is meant by skill? For a nice video summary of the VanPatten paper see the video linked below.

Knowledge is mental representation which in turn is the abstract, implicit and underlying linguistic system in a speaker’s head. Abstract does not mean the rules in a pedagogical grammar rather it refers to a collection of abstract properties which can result in rule-like behaviors. Implicit means that the content of mental representation is not accessible to the learner consciously or with awareness. Underlying refers to the view that a linguistic system underlies all surface forms of language.

The actual content of mental representations include all formal features of syntax, phonology, lexicon-morphology, semantics. And a mental representation grows due to input being acted on by systems from the learners mind/brain.

Skill is the speed and accuracy with which people can do certain behaviours. For language skill this refers to reading, listening, writing, speaking, conversational interaction, turn taking. To be sure being skilled means that the person has a developed mental representation of the language. However having a developed mental representation does not entail being skilled. How skill develops depends on the tasks that people are doing. A person learns to swim by swimming. A person learns to write essays by writing essays.

It follows that the Write&Improve (W&I) tool (as the flagship example of machine learning based tool for language learning) can be seen as targeting how to be skillful in writing Cambridge English Exam texts. The claim that machine learning, and by implication the feedback by W&I, is changing the knowledge of the learner’s English does not accord with VanPatten’s description of knowledge as mental representation. His description implies that no explicit information, in the form of feedback in the case of the writing tool, can lead to changes in the mental representation of the language of writing. He states that research into writing is unclear as to whether feedback impacts writing development.

My point in this post is to briefly clarify the distinction between knowledge and skills (do read the VanPatten paper) and to suggest that the best machine learning based tools can offer are opportunities for students to practice certain skills.

Postnote

W&I has never claimed that its tool has impact on language knowledge. See Diane Nicholls comment below.

References

Van Patten, B. (2010). The two faces of SLA: Mental representation and skill. International Journal of English Studies, 10(1), 1-18. PDF available [https://www.researchgate.net/publication/267793221_The_Two_Faces_of_SLA_Mental_Representation_and_Skill]

BlackBox Videocast 2: Mental Representation and Skill

What Chomsky said about “native speakers” in 1985

This is taken from a rambling but fascinating project by lexicographer Thomas M. Paikeday titled The Native Speaker is Dead published in 1985. He sent a 10 point memo to some linguists on the question of what is a native speaker.  I thought it would be useful to put this up here, since notable ELT bods such as Scott Thornbury used a recent native speaker debate to critique Chomsky (see Geoff Jordan’s response). As to whether Chomsky answered the memo is up for grabs. Personally I think, like David Crystal who also responded to the Paikeday memo, that Chomsky deftly sidesteps the import of the initial memo. The Paikeday book is available on the net but takes some searching, let me know and I can email it to anyone interested.

I marked one passage in orange as it is not clear if this was a response to a specific and separate question asked by Paikeday (on what Chomsky meant by “grammaticalness” from his book Aspects of the Theory of Syntax) or whether it was excerpted from the response Chomsky gave to the Paikeday memo. In Paikeday’s book this passage is the first one but it seems to be oddly placed to me.

Chomsky:

I read your comments on the concept “native speaker” with interest. In my view, questions of this sort arise because they presuppose a somewhat misleading conception of the nature of language and of knowledge of language. Essentially, they begin with what seem to me incorrect metaphysical assumptions: in particular, the assumption that among the things in the world there are languages or dialects, and that individuals come to acquire them.

And then we ask, is an individual who has acquired the dialect D a native speaker of it or not, the question for which you request an “acid test” at the end of your letter.

In the real world, however, what we find is something rather different, though for the usual purposes of ordinary communication it is sufficient to work with a rather gross approximation to the facts, just as we refer freely to water, knowing, however, that the various things we call “water” have a wide range of variation including pollutants, etc.

To see what’s wrong with the question, let’s consider a similar one (which no one asks). Each human being has developed a visual system, and in fact visual systems differ from individual to individual depending on accidents of personal history and maybe even genetic differences. Suppose we go on (absurdly) to assume that among the things in the world, independently of people, there are visual systems, and particular individuals acquire one or the other of them (in analogy to the way we think of languages).

Then we could ask, who has a “native” visual system V, and what is the acid test for distinguishing such a person from someone who has in some more complex or roundabout way come to be “highly proficient” in the use of V (say, by surgery, or by training after having “natively” acquired a different visual system, etc.). Of course, all of this is nonsense.

But I think uncritical acceptance of the apparent ontological implications of ordinary talk about language leads to similar nonsense.

What we would say in the case of the visual system is this. There is a genetically determined human faculty V, with its specific properties, which we can refer to as “the organ of vision.” There may be differences among individuals in their genetic endowment, but for the sake of discussion, let’s put these aside and assume identity across the species, so we can now speak of the visual organ V with its fixed initial state V-0 common to humans, but different from monkeys, cats, insects, etc. In the course of early experience, V-0 undergoes changes and soon reaches a fairly steady state V-s which then remains essentially unchanged apart from minor modifications (putting aside pathology, injury, etc.). That’s the way biological systems behave, and to a very good first approximation, this description is adequate. The things in the real world are V-0 and the various states V-s attained by various individuals, or more broadly, the class of potential states V-s that could be attained in principle as experience varies.

We then see that the question about “native” acquisition is silly, as is the assumption that visual systems exist in some Platonic heaven and are acquired by humans.

Suppose now that we look at language in essentially the same way – as, I think, we should – extricating ourselves from much misleading historical and philosophical baggage. Each human has a faculty L, call it “the language faculty” or, if you like, “the language organ,” which is genetically-determined.

Again, we may assume to a very good first approximation that [the language faculty or language organ] is identical across the species (gross pathology aside), so that we can speak of the initial state L-0 of this organ, common to humans, and as far as is known, unique in the universe to the human species (in fact, with no known homologous systems in closely related or other species, in contrast now to V). In early childhood, the organ undergoes changes through experience and reaches a relatively stable steady state L-s, probably before puberty; afterwards, it normally undergoes only marginal changes, like adding vocabulary. There could be more radical modifications of a complex sort, as in late second language learning, but in fact the same is very likely true of the visual system and others.

Putting these complications aside, what is a “language” or “dialect”? Keeping to the real world, what we have is the various states L-s attained by various individuals, or more generally, the set of potential states L-s attained that could in principle be attained by various individuals as experience varies. Again, we see that the question of what are the “languages” or “dialects” attained, and what is the difference between “native” or “non-native” acquisition, is just pointless.

Languages and dialects don’t exist in a Platonic heaven any more than visual systems do. In both cases, there is a fixed genetic endowment that determines the initial state of some faculty or organ (putting aside possible genetic variation), and there are the various states attained by these systems in the course of maturation, triggered by external stimuli and to some rather limited extent shaped by them. In both cases, there is overwhelming reason to believe that the character of the steady state attained is largely determined by the genetic endowment, which provides a highly structured and organized system which does, however, have certain options that can be fixed by experience.

We could think of the initial state of the language faculty, for example, as being something like an intricately wired system with fixed and complex properties, but with some connections left open, to be fixed in one or another way on the basis of experience (e.g., do the heads of constructions precede their complements as in English, or follow them as in Japanese?). Experience completes the connections, yielding the steady state, though as in the case of vision, or the heart, or the liver, etc., various other complications can take place. So then what is a language and who is a native speaker? Answer, a language is a system L-s, it is the steady state attained by the language organ. And everyone is a native speaker of the particular L-s that that person has “grown” in his / her mind / brain. In the real world, that is all there is to say.

Now as in the case of water, etc., the scientific description is too precise to be useful for ordinary purposes, so we abstract from it and speak of “languages,” “dialects,” etc., when people are “close enough” in the steady states attained to be regarded as identical for practical purposes (in fact, our ordinary usage of the term “language” is much more abstract and complex, in fact hardly coherent, since it involves colors on maps, political systems, etc.). All of that is fine for ordinary usage. Troubles arise, however, when ordinary usage is uncritically understood as having ontological implications; the same problems would arise if we were to make the same moves in the case of visual systems, hearts, water, etc.

About the term “grammaticalness,” I purposely chose a neologism in the hope that it would be understood that the term was to be regarded as a technical term, with exactly the meaning that was given to it, and not assimilated to some term of ordinary discourse with a sense and connotations not to the point in this context.

Such questions as “how many languages are there” have no clear meaning; we could say that there is only one language, namely, L-0 with its various modifications, or that there are as many languages as there are states of mind/brain L-s, or potential states L-s. Or anything in between. These are questions of convenience for certain purposes, not factual questions, like the question of “how many (kinds of) human visual system are there?”

Apparent problems about the number of languages, native speakers, etc. arise when we make the kind of philosophical error that Wittgenstein and others warned against.

I think that looked at [my] way, the questions you raise no longer seem puzzling, and in fact dissolve.

References:

Paikeday, T. M. (1985). The native speaker is dead! An informal discussion of a linguistic myth with Noam Chomsky and other linguists, philosophers, psychologists, and lexicographers. Toronto and New York: Paikeday Publishing

How many words do scholars have for “How many words do Eskimos have for snow”?

The photos above were taken from a book on mountains (the title of which I had forgotten to note) whilst on a winter break. I remember feeling vaguely superior to the author of the mountain book as I vaguely recalled the “debunking” of the Eskimo-languages-have-so-many-words-for-snow myth. Then I promptly forgot about the issue till @EngliciousUCL tweeted a study by Regier et. al. (2016) called “Languages Support Efficient Communication about the Environment: Words for Snow Revisited”.

This study makes the point that in all the fuss about the status of the Eskimo words for snow an underlying principle has not been tested, that is “language is shaped by the need for efficient communication”. The authors go on to demonstrate the support for this principle.

This post is a limited attempt to list interesting articles written by scholars on the Eskimo snow words topic either for the public or more specialist audiences.

So back to the question – How many words do scholars have for “How many words do Eskimos have for snow”?

There are at least 40000 words using the following references (click here):

As mentioned the above is a limited list as this does not include texts with a passing mention of “Eskimo words for snow” and/or that use it as a prompt for other related examples (many of which you can find by doing a search in Language Log). For texts before Martin (1986), Cichocki and Marcin (2010) provide a comprehensive history.

The Language Log blog is also the stomping ground of the author of one of the most popular descriptions of the refutation of the “How many words do Eskimos have for snow” – Geoffrey Pullum.

To recap Pullum (1991) following Martin (1986) points out that the number of distinct words, defined as root forms, that Eskimo/Inuit languages have for snow is four. This number is taken from the text published by the anthropologist Franz Boas in 1911. The problems of identifying what are words is compounded in Inuit languages as they often add bases together to form whatever description they need. Hence the number of words for snow or for seals or for ice is unbounded. The number of sentences in English to describe say a wintry scene is similarly infinite.

Bearing the above in mind, specialists in Eskimo/Inuit languages such as Kaplan (2003) say the number of root forms amount to 3. Whilst other specialists like Woodbury (1991) gives 15 lexeme meanings, where a lexeme is similar to a root.

One study of several Eskimo dictionaries by Krupnik & Müller-Will (2010) argues that independent words that are derived from roots represent “a meaningful and clearly distinguishable phenomenon to indigenous speakers”. Hence the number of ways Eskimo languages describe snow is quite rich, even more so for words to describe ice. They add if you really want a language with a 100 words for snow look to the Norwegian Sámi.

This shallow trek into the Eskimo words for snow trope brings up a couple of points – 1) how various factoids one learns about language often hides more interesting principles as Regier et. el. (2016) show; 2) how using English as a comparison language as well as the metalanguage of comparison may result in erroneous native speaker intuitions projected onto a language with a very different classification system (Silverstein 1991).

I’ll leave you with a question asked to Michael Silverstein one of the players in the early drama:

I asked Silverstein if he had ever thought about popularizing the field of linguistics in the way academics in other disciplines have.

He recoiled. “That’s an ethical question,” he said. “There are people who are scientific evangelists, who are no different in kind than any other evangelist. I’m enough of a Menckenite to be a skeptic–that is to say, to realize that my claim to systematic knowledge of a social phenomenon is just one more thing that might go into the hopper of whatever the phenomenon happened to be. One might say that that’s just not the kind of phenomenon that responds to that sort of treatment. ‘You just think you’re studying it, you’re not really studying it; what the phenomenon is is exactly what John Simon says it is, and pooh-pooh on all of your stuff.’ Because, remember, the phenomena are us.”

(Watch Your Language! Anthropological Linguist Michael Silverstein on Australian Aborigines, Wine Nuts, Dear Abby, and the Language Police by Bill Wyman, February 14 1991)

Thanks for reading.