logo ELT Concourse teacher training
Concourse 2



This is quite a long guide with several sections so you may like to search it (use Ctrl + F to do that).
The guide defines two directions of inferencing and then considers top-down and bottom-up strategies with some classroom implications of both.  On the way, it covers extra-, inter- and intra-linguistic clues to meaning.
You can click on these links if you are only interested in top down or bottom-up processing.


What does this mean?

If you glance at this image, it seems reasonable to assume that will understand that it tells you the direction to an emergency exit.  However, nothing in the image explicitly states anything about an emergency, an exit or what you are supposed to do with this information.  You inferred all the data you need from simply glancing at the image.
How did you do that?  Think for a moment and then click here.

This is not particularly mysterious, although it is quite clever.  People do it all the time.  In fact, it would be virtually impossible to live in any sort of society without drawing reasonable conclusions based on your experience and knowledge of other people from what you see, hear and read.


What is inferencing?

Inferencing is often described as a form of guessing.  It is not.  Guessing usually implies a stab-in-the-dark approach to knowing something.  Inferencing, on the other hand, may be defined as

deriving logical conclusions from premises known or assumed to be true

In other words, it is a method of calculating what is likely to be true based on prior expectations and knowledge.  In the field of statistics, one approach is to do what is called Bayesian processing to arrive at reasonable inferences based on expectations and previous knowledge of the world.

In the process of learning or using a language, inferencing is a key skill but it is not one, as we saw above, that learners need to be taught how to do so much as one they need deliberately to employ.


Here's an example:

If you are presented with a die and asked to infer what your chances of throwing a six (or any other number between 1 and 6) were, you would probably say, quite rightly, 1 in 6, i.e., a roughly 16.67% chance that any throw of the die will result in a six (or any other possibility).  You would, therefore, guess that your chances of throwing a particular number are 1 in 6.
Now, however, you are told that out of the last 100 throws of this particular die, 50 have resulted in a six.  What is your 'guess' now of the chances of a six coming up?
Right: inferencing is more than guessing.

Now take the situation in which you are faced with an unknown word in a sentence.  The word is, for argument's sake, chenlob and you have no idea what it means at all.  Can you infer its meaning?
No, you can't but you can make some guesses based on what you know about the language.

You have, roughly a 20% chance that the word is a noun and much less than a 1% chance that it's a hitherto unknown preposition.  It could easily be a verb (a chance of something like 15%) but it is less likely that it is an adjective or an adverb (around 7% each).  It is almost impossible that the word is a new (to you) determiner or a pronoun.

Now you may be presented with a little co-text and see:
    the chenlob
because you can infer from your knowledge of syntax that the word is a noun, preceded by a determiner, you may adjust your inference to decide that it is a noun of some kind.

Unfortunately, you now see more co-text and that is:
    the chenlob wall
so now you may discard your original inference and decide, on the basis of this new evidence, that you are presented with an adjective which is modifying the word wall.  Your knowledge of syntax allows this to be the most likely possibility because the sequence determiner + adjective + noun is a commonplace of English.
Moreover, you also know what a wall is and that can lead you understanding that the adjective does not mean something like speculative, electronic, papery, or happy because these are not epithets which can be applied sensibly to walls.  It could, naturally, describe a range of other possible ways a wall can be including high, brick, beautiful, surrounding, impenetrable and so on.

If, however, you were presented with
    The chenlob spoke at the meeting
you would be able to reinforce your suspicion that the word is a noun and, moreover, that it represents some kind of person because your knowledge of the world includes the fact that only humans speak at meetings.

If you are now given written context (with punctuation as well as co-text) such as:

The Chenlob spoke at the meeting and all her fellow Chenlobs applauded her even though the chairman had determined that Chenlobs do not have voting rights because they come from outside the county.

you may be able to make a pretty good guess that a Chenlob is a member of a definable set of people identified with a place.

What you have done here is apply probabilistic reasoning based on both your knowledge of the syntax of the language and your understanding of the world of meaning.  In other words, you have applied extra linguistic knowledge as well as intra-linguistic knowledge to reach your inference.

Dennett, 2017:269, puts it this way:

... the brain's strategy is continuously to create "forward models," or probabilistic anticipations, and use the incoming signals to prune them for accuracy – if needed.  When the organism is on a roll, in deeply familiar territory, the inbound corrections diminish to a trickle and the brain's guesses, unchallenged, give it a head start on what to do next.

What we do next, of course, is understand and, if necessary, act on the linguistic data we are receiving.


Why is inferencing important?

There are two main reasons:

  1. No user of a language can possibly know all the words, structures and expressions in the language so we have to infer meaning from what we do know.  This is, naturally, especially important for learners of a language.
  2. Very few texts, whether spoken or written, actually contain all the information we need fully to understand them.  Speakers and writers will always assume some prior knowledge in their hearers and readers.  If we didn't do this, texts would be interminably long.  Try explaining how to make an omelette to someone who doesn't know anything about food, what a kitchen is or what it's likely to contain and you will see what's meant.


Extra-linguistic knowledge

In order to understand the sign at the top of this page, you used no linguistic knowledge at all simply because no language was present.  What you used was extra-linguistic knowledge.  That is your knowledge of the world, people's motivations for informing you and your understanding of shape and movement.

Extra-linguistic knowledge can be further refined:

knowledge of the world
this was what you used to understand the three simple shapes on the sign, the implied direction of movement and the purposes of the sign in general
knowledge of people's motivations
the assumption you used here was that a sign is intended by someone else to tell you something helpful
knowledge of roles in society
you assumed, probably rightly, that the people who designed it and had the sign put up were a) in some kind of authority and b) in possession of knowledge that you need, i.e., better informed

As an example of how these can be applied to text, try figuring out what this means:

It's OK, ladies and gentlemen.  You can go back to your offices now.

  1. What is the meaning?
  2. Who is speaking to whom?
  3. Where is it happening?
  4. What has just happened?
  5. What happens next?
  6. How did you know all this?

Click here when you have answered those questions.

Most of the meaning and understanding of circumstances that you are able to extract from the language are not to do with language clues at all.  You have deployed two types of knowledge to do all this.

bottom-up processing

You used this to figure out the import of OK, go back and now.  If you were actually listening to the speaker, you would need to use your knowledge of the phonemes of English, of connected speech, of word meaning and word class and syntax.

top-down processing

You used this for everything else.  Your sociolinguistic knowledge, your knowledge of the world, your understanding of social relationships and your ability to draw logical conclusions from linguistic clues are all facets of top-down knowledge that you bring to complete your understanding.

in combination

It is very important to understand that, although we have separated out the types of processing you did here, you did not use one type of processing and then the other; you combined the types to make meaning clear.  That is how people do things.


Extra-linguistic knowledge and written texts

As far as strategies are concerned, the same considerations apply to written text.  We have more time and leisure to extract the meaning but fewer clues about the situation and role relationships to work on.  As a result, we use less extra-linguistic knowledge to figure out meanings.
That does not, however, mean that our knowledge of the world, our ability to draw logical conclusions and our knowledge of the social fabric around us are abandoned.
An email from a friend and a letter from the tax man will set up completely different expectations in us and will be dealt with in different ways.

reading between the lines

A lot of reading is actually reading between the lines, not simply extracting meaning from the written word.  Take, for example, these snippets from British newspapers and see what they mean and what they imply.

Some types of text will require more knowledge of the world or a particular culture than others if they are going to be correctly interpreted:

You can probably think of many more occasion when what you read needs to be interpreted by using extra-linguistic data.  Advertising language is, of course, a rich source of such things because the writers of it are trying to make a case for something.  Only the very optimistic will interpret Up to 50% off to mean that what we might actually want will be half price.


Inter-lingual clues: inferencing across languages

One form of knowledge which also lies just outside the text (spoken or written) is the knowledge learners have of their own languages.  Many teachers have a tendency to see learners' first languages as some kind of impediment to learning English and there are occasions when cross-language interference is a problem but using the forms, lexis and structures of one's first language to discover meaning in a second is often very helpful.
Many European languages for example, will have a range of cognate words and structures which aid comprehension and, although such things are often unreliable, they are, more often not, helpful.  For example:


Inferencing is a two-way street

We can, as you saw above, use inferencing skills to work from the language to the situation but, probably more importantly for learners of a language, we can also work from the situation and our knowledge and cognitive abilities to understand the language.  Graphically this two-way process is like this:

two-way inferencing

top down stairs

Using top-down processing to access texts

Knowledge external to the text itself is a powerful tool in helping us to understand.  Using such knowledge is called top-down processing (a term stolen from cognitive psychology, by the way).
We have seen the sorts of knowledge one might employ to figure out a context for an utterance (the office evacuation scene).  Now, we'll work the other way and see how the knowledge our learners already have or can be given helps them to understand a written or spoken text.

Knowledge of the world

Given a clue such as this, most of us can figure out what the following text will be about.


Protesters take to the streets

It's not at all difficult to work out the meaning of protesters and take to the streets from this.
In our first languages, we unconsciously use images and titles to predict what sort of text we are about to read.  We also use images, the situation and introductions to predict what we are going to hear (on TV, in an airport or on the radio, for example).

This is a process often called schema activation.  A schema can be defined as a mental framework that we can use to organize and understand information.  (The plural is, incidentally, schemata.)
The picture above and its headline caption serve to activate our schema concerning demonstrations and riots so we are prepared for what follows.
That can be very helpful but schemata cut both ways: activating a schema may lead us to exclude or ignore information which does not fit in to our world view or expectations.

Once primed by an active schema our minds are prepared to encounter the sorts of language we are likely to find in the text that follows (whether we are listening to / watching a TV programme or reading an article).  This has three main characteristics.  Think for a moment about what you expect the text to be like and then click here.


Teaching implications

Learners of a foreign language will draw on world knowledge and their various schemata unconsciously in their first languages.  However, in a second language, the argument is that they need some specific nudging to do so.  For this reason, pre-listening / reading tasks are often devised along the lines of, for example:

Lexis awareness and prediction:

Cross out the words you do not think will be in the text:

verbs nouns adjectives
Now discuss your lists with two others and say why you think the words will / will not be in the text.

Information checklist

Write down four questions you want to have answered by the text.  For example, "Where did this happen?"

Information staging

Where do you think this information will come?  Put B (beginning), M (middle) or E (end) next to the item.

Information B, M or E?
what happened?  
why was there a riot?  
was anyone hurt?  
what did the police do?  
how many people?  
what happens next?  
Now discuss your lists with two others and see if you agree.

Simple tasks like these, done before hearing or reading a text can encourage learners to make useful inferences and these, in turn, help them understand.

A word of caution: however well designed and helpful such tasks are, they will not provide learners with all the information they need.  If, for example, the text contains expressions such as carry out a baton charge, no amount of schema activation will help the learner understand if the lexis is unknown.  To do that, learners need to activate other kinds of knowledge.

bottom up

Using bottom-up processing to access texts

Bottom-up processing refers to using in-text information to understand meaning.  This means using not the context but the co-text.  In other words, the learner has to use intra-lingual clues.
If, for example, a text contains the following, what information can learners use to understand the underlined items?

The police carried out numerous powerful baton charges during the day and there were many casualties requiring hospital treatment.

Click here when you have thought of something.


Logical connections

An aid to understanding is also to look for logical connections in texts.  This means deliberately looking for things like hyponymy, synonymy and antonymy, cause, consequence and purpose.  Using logical connections is a prime case of applying both top-down and bottom-up knowledge simultaneously to get to meaning.
For example:


Teaching implications

Bottom-up processing is complicated and you cannot expect your learners to apply the whole range of strategies at the end of a couple of lessons.
As with much else, the teacher's job is to make eating the elephant a more digestible undertaking.  We need to take the skills individually, design teaching procedures and materials that target them and alert our learners to the usefulness of the strategies and when it is appropriate to use them.

Inferencing is undoubtedly a key skill but wild guessing is not to be encouraged if we are trying to access the instructions for operating a chain saw.

Morphology, word class and sense relations

If you are lucky enough to find a text at the right level which contains numerous opportunities to identify word class and analyse morphology then use it.  Otherwise, you may find it helpful to invent a text which contains items such as:

The trainee soldiers took careful aim at the target and let off a disorganised, ragged volley of shots. They all missed and some went very wide indeed.

Here, we can encourage learners to develop the skills of:


What do the words in red mean?
Are they things or people?


Purpose and consequence

Can you draw a picture of the words in red?

What kind of thing is the word in red in:

The trick, as always, is not to overload the learners but to present language and skills logically in manageable chunks.



Related guides
Krashen and the Natural Approach for the guide to this set of hypotheses
how learning happens for a general and simple overview
second-language acquisition for a guide to some current theories
semantics for a bit more on meaning and schema activation and how it works to refine meaning
input for an obviously related guide
noticing for more on a related and important learning skill
word formation for some more on affixation and word class / meaning
morphology for some analytical background to word formation
colligation for a guide to this form of grammatical collocation
unlocking learning this is a guide in the Delta section concerned partly with how inferencing should be done
the initial-plus skills index for some essential guides
the in-service skills index for those guides and a few more

References for this kind of analysis:
Dennett, D, 2017, From Bacteria to Bach and Back, UK: Penguin Random House
Hedge, T, 2000, Teaching and Learning in the Language Classroom. Oxford: Oxford University Press
Field, J, 2009, Listening in the Language Classroom, Cambridge: Cambridge University Press
Grellet, F, 1981, Developing Reading Skills, Cambridge: Cambridge University Press
Hudson, T, 2007, Teaching Second Language Reading, Oxford: Oxford University Press
Lynch, T, 2009, Teaching Second Language Listening, Oxford: Oxford University Press
Rost, M, 2002, Teaching and Researching Listening, Pearson
Underwood, M, 1989, Teaching Listening, Harlow: Longman
Ur, P, 1984, Teaching Listening Comprehension, Cambridge: Cambridge University Press
Wenden, A & Rubin, J, 1987, Learner Strategies in Language Learning, Hemel Hempstead: Prentice-Hall International
White, G, 1998, Listening, Oxford: Oxford University Press