why is language difficult? john barnden school of computer science university of birmingham natural...

30
Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natur Natur al al Language Process Language Process ing ing 1 1 2014/15 Semester 2

Upload: william-todd

Post on 29-Jan-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Why is Language Difficult?

John Barnden

School of Computer ScienceUniversity of Birmingham

NaturNatural al Language ProcessLanguage Processinging 1 1

2014/15 Semester 2

Page 2: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Some Questions to You

• How many of you like Mathematics?

• How many like learning another language?

• How many students are there at UB?

• Which of my aunts went to Sweden?

Page 3: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

• How many students are there at UB?

What did you do to answer me?

• How many of you like Mathematics?

How many like learning another language?

What did you do in this case?

And did I say “of you” in the second one?

The Profundity of the Obvious:Why is NLP Difficult?

Those Questions again …

Page 4: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Those Questions again …

• Which of my aunts went to Sweden?

What DIDN’T you do?

Did you …

Try to remember hard whether you’ve heard about my aunts?

Try to work it out from general knowledge of old English aunts?

Ask the person next to you?

Pretend to know the answer?

Etc.

Page 5: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Natural Language Processing

WHY DIFFICULT?

• CONTEXT-SENSITIVITY: to surrounding discourse, prevailing circumstances and culture, what bits of the world are like, …

• WE FILL IN AT LOT without (usually) realizing:

– An utterance is really just a CLUE to some intended message, not a faithful representation of the whole message.

Page 6: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Filling Things In

• “He stirred his coffee and handed the spoon to his wife.”

• “When she got to her front door she found she’d lost her key. She couldn’t find an open window either. So, she booked in at a hotel.”

Page 7: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

(More) Ambiguity & Context-Sensitivity

• At Reading railway station :

• What’s the euro / dollar exchange rate?

Page 8: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

More … CAUTION HORSES seen on the back of a big van on the M5.

What might this mean in isolation from that specific situation? …

– There are horses in this van, so be careful not to hurt them.

– There are horses in this van, they might attack you.

– Horses, be careful!

– Warn some horses [about something].

– Here are horses belonging to the Caution family.• There are people with surname Caution on the web!

– OTHERS??

Page 9: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

NLP is Not an Easy Probleme.g. going by Google Translate and back

• We do not have an army that fades away at the first sign of trouble.

No tenemos un ejército que se desvanece a la primera señal de problemas.

We have an army that vanishes at the first sign of trouble.

• We do not have an army that fades away at the first sign of trouble.

[Welsh:] Nid oes gennym byddin y pylu i ffwrdd ar yr arwydd cyntaf o drafferth.

We do not have the army faded away at the first sign of trouble.

Page 10: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Going by Google Translate and back, contd

• He drove a coach and horses through the problem.

Conducía un coche de caballos por el problema. (??)

He drove a cab for the problem.

Conducía un taxi para el problema.

TRIED: Conduce un coche de caballos por el problema.

Drive a horse carriage by the problem.

Page 11: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Various Standard Difficulties• Lexical ambiguity (i.e., ambiguity of individual words):

He wanted some money, so he went to the bank.

• Syntactic ambiguity [first two examples are of ambiguity of prepositional phrase attachment (PP attachment)]

He went to the bank with the pizza parlour next door.

He went to the bank with the lawyer next door.

She saw that gasoline can explode.

You shut the door. [declarative or imperative?]

• Anaphora (mainly = pronoun reference):

John shot Bill, but he didn’t kill him.

The terrorists killed the customers, so the police shot them.

The boss slashed the budget and abolished coffee time. I don’t like it.

Page 12: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Various Standard Difficulties, contd• Coreference more generally:

SnakeByte Technologies has taken over RabbitWare. … ST has been quite aggressive recently. … The company has doubled its profits.

• Indexicality:

Words such as: I, you, now, yesterday, here, that orange.

• Vagueness:

Many students are poor. / Everyone wants to see less crime.

About ten thousand students drive.

Ten thousand students drive.

The meal was great.

Teachers are good people.

There aren’t many Eskimos here. / That countryside is pretty.

Page 13: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Various Standard Difficulties, contd• “we” / “you” / “they”:

Who does it include, and how do you know?

Impersonal “you” versus specifically- referential “you”.

“they” meaning (roughly) people in power

• Noun-noun (and some other) compounds:

beer mug / clay mug

city shops / food shops

hotel room / shower room

pencil case / pencil mark / pencil shavings

pencil case zip repair specialist / West Midlands zip repair specialist

Page 14: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Various Standard Difficulties, contd

• Implications of various sorts (the examples below would often be called “implicatures”)

Some of the apples are rotten. (Implication: but not all of them.)

That wasn’t helpful. (Implication: it was the opposite of helpful.)

Can you pass the salt? (Implied request to actually pass the salt!)

It’s cold in here. (Could imply: close the window.)

It’s raining. (Implication: here, not just anywhere. Implicit indexicality.)

To get to the airport, call a taxi. (Implication: and ride there in it!)

And “filling in” sorts of example generally.

Page 15: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Various Standard Difficulties, contd(though less well studied in NLP)

• Ellipsis.

John wanted strawberry ice-cream, but Mary chocolate.

John loves his wife. Bill too. [but whose wife?!]

• Metonymy (indirect reference to something via something else):

Mary likes Bach. [could mean: … the music of Bach]

She drank the whole bottle. [she didn’t melt the glass and drink that!]

England scored three goals. [the team playing for England did so]

You’ll find Plato on the top shelf. [can’t he climb down?!]

Page 16: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Sign on door in Arts LR5: What Issues Raised?

Page 17: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

On to Metaphor• A “liquid” use of language (or of ...).

– Roughly: talking (or …) about something as though it were something else (to which it seems or comes to seem similar).

• NOT just of importance in “high” literature.

Important also in, for example,

– news articles, classroom discourse, advertisements, doctor/patient dialogue, blogs, internet chat, mundane fiction, …

– So, an important practical matter for AI.

Page 18: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Mundane Examples of MetaphorMeeting life head on.

He drove a coach and horses though the problem.

My work overflowed into the weekend.

When you come to our hotel, your car gets a holiday too.

When your car doesn’t want to accelerate, check the fuel line.

The idea was buried in the recesses of her mind.

The news knocked me off my feet.

They’re marching ahead and will soon overtake the other political parties.

The thought of her step-mother's arrival hung over her mind like a dark angry cloud.

This slide is stuffed full of examples!

Page 19: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Mundane Examples of Metaphor

We do not have a chocolate army that fades away at the first sign of trouble.

The weather is settling into a drier frame of mind.

But Ireland is also an island, divided, angry, full of old demons and old hate.

You’re a dog / cow / pig / devil / monster / baby / lunatic / angel / god / ….

Page 20: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Mundane Examples(!) of Metaphor

Sharon pulled herself out of her jeans, the words “How

could he? How could he?” jumping about her wearied brain.

Senseless, leaving her empty, cold, helpless. Another voice,

angry and vindictive, shouted in her ear, “Serves you right,

you silly fool: play with fire and watch your life go up in

flames. It was all so predictable.”

Page 21: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Mundane Examples of MetaphorFrom Foreword, Preface and Introduction of the book Lima by James Higgins (Oxford:

Signal Books, 2005):

– Brasilia comes up and hits you with its modernity, …

– Lima … sits there inscrutably in its sandy desert, keeping itself to itself.

– Higgins delights the reader with clues about how to pierce Lima’s mysteries and riddles.

– Higgins deciphers the politics that lurks behind that [decision].

– Higgins brings Lima closer to non-Peruvians …

– Yet if the city tends to be identified with Spanish colonialism, there are other Limas.

[Contrast this with examples like: “Is Birmingham the Lima of England?”]

– For all its size Lima is a small village and through [my friend] I got to know [other important people].

Page 22: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Describing the Abstract

• Metaphor is hugely important (prevalent, useful, crucial) for describing & thinking about relatively abstract matters.

Time:– We’re rapidly approaching Christmas.– Christmas is galloping towards us.

• Financial matters:– Money flows across borders so easily now.– Stock prices plummeted, soared, rocketed into space, …

• Many sorts of processes, including in lives and relationships:– They seem to be going on different paths, taking them further apart.

• Computing:– web, window, navigating, surfing, spiders, viruses, kill a process, dump

the memory, trees, loops, arrays, heaps, …

Page 23: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Not Just in Language!!

• Metaphor arises in all media of communication/expression:

gestures, pictures, diagrams, music, dance, …

• Many theorists (including me) regard metaphor as a fundamental aspect of thought.

Its appearance in language, pictures, etc. is just a reflection of that.

Page 24: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

The Goods, The Ads and the Bubbly:

Metaphor in Advertisements

Page 25: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

Advertising mobile apps

Page 26: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

(Partly) Pictorial Metaphor

Following slide is courtesy of Marianna Bolognesi, 2013.

Page 27: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2
Page 28: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2

(Partly) Pictorial Metaphor

Following two slides are courtesy of Elisabeth El Rafaie, 2013.

Page 29: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2
Page 30: Why is Language Difficult? John Barnden School of Computer Science University of Birmingham Natural Language Processing 1 2014/15 Semester 2