Lecture 7: Sentence Production Flashcards
(23 cards)
How do we speak?
Decide what we want to say (Conceptualization)
Find words that fit our message (Lexicalization)
Figure out which order those words need to go in to properly express our message (???)
Figure out what the sentence structure needs to be given our message and the words we’ve selected (???)
Figure out how the words we’ve selected are going to sound (Form encoding)
Plan several muscle movements to create the appropriate series of sounds (articulation)
Producing Sentences
Message: WAKE (alarm clock, boy) (prelingistical representation, output of conceptualization process)
This is a notation for expressing semantic content. This is an idea we want to express.
We know the steps we need to go through, roughly. But how do those steps relate and how do we get started?
One possibility is we just retrieve all the words and then decide how to order them
-Serially is unlikely, since we speak so quickly…not very efficient. (2-4/sec)
-Simultaneously is difficult, given what we know about lexical competition.
Sentence production must be more incremental; we must retrieve words one at a time.
->build sentence based on what have as soon as have something, start building sentence, then build on as other information comes in, don’t wait to have enough stuff to start
Which takes us back to the question, where do we start?
Producing Sentences with Structural options
Message: WAKE (alarm clock, boy)
If we retrieve boy first, what sentence options do we have?
The boy was woken up by the alarm clock.
If we retrieve alarm clock first, what sentence options do we have?
The alarm clock woke the boy up.
The alarm clock woke up the boy.
If we retrieve wake first?
-not very many declarative ways when start with wake
- language provides us with structural options
- some cases can change order with minimal different meaning
- for any one message have structural options, structural options are important for how system works
- question of where start isn’t important, lots of ways to produce lexical sentence
- will have subtle meaning differences, shaped by meaning want to convey
-For every idea we want to express, our language gives us various options.
How do we choose between options?
Subtle meaning difference?
Psychological efficiency?
Chance?
(more in notes)
Stages of language production
What do I want to say?
Conceptualization
How do I find the right words to express my idea?
How do I order those words into well formed
sentences?
Formulation
How do I make my mouth or hands move to produce the words and sentences?
Execution/Articulation
Garrett’s Model of Production
Conceptualization Message Level
Functional Level Formulation Positional Level
Sound Level
Articulation Articulatory
Instructions
Bock & Levelt, 1994 Model
Message
Functional Processing: Lexical selection & functional assignment
-Lemmas retrieved and tagged for functions
Positional Processing: Constituent assembly & inflection
-Structural relations between words established, slots for
function words and inflections built
Phonological Encoding
(look notes)
Functional Processing
-functional level encapsulates lemma selection, lemma selection one part and function assignment next (where specify grammatical role will play in sentence) then passed down to positional where actually ordered
The grammatical functions are assigned to nouns: The boy (subject) kissed the girl (dir_object) The girl (subject) kissed the boy (dir_object) ->sentences different, why know this is because position in sentences have meaning The boy (subject) gave the girl (ind. object) the hat (direct object).
subject-> 1st noun, doing action
direct object-> 2nd noun
indirect object
-know this because assigned a function, have to be encoded grammatically, if don’t do function assignment risk of grammatical not making sense
If function assignment goes wrong, words get the wrong
grammatical role:
She offends his sense of how the world should be.
(Speaker intended to say: He offends her sense of how the world should be.)
Positional Processing
-ordering decisions, positions in sentence for all function words
Decisions about linear order and constituent structure are made here
If Girl is retrieved first and tagged as the subject of the sentence when she wasn’t the person who did the kissing, we might get a structure like:
The girl was kissed by the boy.
-lot of flexibility, because one stage for function and the other for structure, allows to move forward to form sentence
-incremental
The positional level creates the constituent structure of the sentence.
Other ordering decisions made at the positional level:
Order of adjectives: Tall, dark and handsome or dark, handsome and tall?
Conjunctions: Bread and butter or butter and bread?
Inflections and function words also inserted
Not part of frame
Evidence for 2-stages of grammatical encoding
Speech errors: just as speech errors supported the distinction between Lemmas and Lexemes, they also support the distinction between functional and positional processing.
Word exchange errors (I sent a mother to my letter) suggest that the right lemmas were selected but inserted into the wrong positional slots.
Stranding errors (I trucked my park) show that words move independent of their inflections. Again, right lemmas retrieved but put into wrong slot.
Shows determining order and lemma retrieval are separate.
Experimental evidence: syntactic priming
These studies evaluate the influence of prime stimuli on how a subsequent picture is described.
We’ll focus on how they related to the 2-stage model in our tutorial
Experimental evidence 2: agreement errors
When do we make them and why.
Testing these ideas with Syntactic Priming
-method that capitalizes on this pre alternation we have in english language
One alternation:
• Active: The alarm wakes the man.
• Passive: The man is woken by the alarm.
Other Alternation Dative??
• Prepositional object (PO) structure: The cowboy gives a banana to the robber.
• Double object (DO) structure: The cowboy gives the robber a banana.
(almost same meaning but structure different)
-syntactic priming takes advantage of this different ways for expressing ourselves that don’t have much meaning difference to try and understand sentence production
-read or hear prime sentence that has one or other of alternative structures
Present prime sentence: The dog bit the girl.
OR
The girl was bitten by the dog.
-then target picture and produce sentence to describe
Present a target picture to be described.
The nun is following the sailor.
OR
The sailor is being followed by the nun.
-independent variable is how often people produce same time of sentence structure just read (priming)
-tendency to reuse structure just processed
After processing a sentence containing a certain syntactic
structure (e.g., a passive), a related syntactic structure (e.g., another passive) becomes more activated and is produced more often.
Research on syntactic priming in production has shown:
Passives prime passives, actives prime actives.
POs prime POs, DOs prime DOs.
This suggests that sentences with the same structures are mentally represented in the same way, and therefore, that syntax is involved in sentence production.
Structural vs functional words
Experiment
Bock 1989
Recipient: The secretary took a cake [to her boss].
Benifactive The secretary baked a cake [for her boss]
(different meanings)
-picture of man with fish and seal
Both prime prepositional dative structures (the man feeds fish to the seals) more than double object structures (the man feeds the seals fish).
Shows that priming is obtained even if prime and target have different argument roles (semantic) and different function elements (lexical similarity)
-priming is insensitive to semantics and to specific function words
-function words are inserted during positional processing and this is happening before this occurs
-position words have not be inserted yet during priming
Structural vs Semantics
-experiment
Bock and Loebell 1989
-is it all about structure or is it about meaning too
Active: The construction worker drove the bulldozer.
Passive: The construction worker was hit by the bulldozer.
Locative: The construction worker was digging by the bulldozer. (same structure as passive but very different meaning)
-lightening stuck the church or the church was struck by lightening (passive)
Both passives and locatives prime passive structures more than active structures despite different meaning.
Priming is syntactic, not semantic.
Constituent structure vs. surface similarities
Bock and Loebell 1989
-is driven by surface similarity or is it really syntax
DO prime: Susan brought Stella a book.
PO prime: Susan brought a book to Stella.
To-clause: Susan brought a book to study. (looks on the surface similar as PO prime but very different syntactically, one bringing it to someone, other bring it to do something)
Only the PO prime biases towards prepositional dative structure (The man feeds a fish to the seal). To-clause does not prime
If prime and target don’t have same constituent structures, no priming, even if similar on surface
Lexical Interactions
Pickering & Branigan, 1998
-what if the sentences have closer meaning or same verb meaning (before sentences didn’t relate, these ones did)
PO prime, rep: The mother feeds a biscuit to her child
PO prime, form: The mother is feeding a biscuit to her child. (lemma same as above, lexeme different, if phonological has effect will matter but if just functional level won’t matter)
PO prime, no rep: The mother gives a biscuit to her child. (verb different)
-still get syntactic priming effect
-if repeat the verb (verb same in prime sentence and target sentence) get even bigger priming effect that if different verb (lexical boost)
No change in priming magnitude when form of verb changes
Boost in priming when LEMMA repeated.
Suggests abstract lexical items (Lemmas) underlie lexical interaction, not form (Lexeme) similarity.
Suggests phrasal options linked to lemmas (like other grammatical features)
Implicates Functional processing
-happening before phonological encoding
(look notes, wee confused)
The importance of Structure
In addition to supporting the 2 stage theory, syntactic priming also gives us evidence about how we make choices between structural alternatives.
Not just about best expressing meaning; structures can be primed.
The syntactic priming effect is interpreted as a tendency for speakers to reuse STRUCTURE they recently encountered.
Syntactic frames independent of meaning, specific words & function words.
What is the evidence that the effect is STRUCTURAL??
Not Lexical repetition
Not Semantic similarity
Not Function word repetition
Shared structure needed
More evidence for syntax in sentence production
Syntactic priming studies provide insight into how we produce sentences. They also provide evidence for the use of structure during production.
Another source of evidence for the use of structure in production comes from Agreement errors. (every verb in english agrees with the subject in number)
Agreement errors are cases when the main VERB of the sentence does not share the same number feature as the SUBJECT of the sentence (when get wrong)
I am / go / have
You are / go / have
He is / goes / has
We are / go / have
Agreement errors
What determines whether you make an error?
In many languages, the main verb must agree in number with the subject.
The keys ARE in my bag.
The key IS in my bag.
Note: The number of bags (the Direct Object) is irrelevant to the form of the verb (form of the verb is only sensitive to the number features of the subject)
The key IS in my BAGS.
Speakers sometimes make agreement errors, especially when a second noun is in the subject NP
The key to the cabinets ARE*/IS in my bag. (when gets more complicated when subject gets more complicated, subject and verb are far appart)
Speakers make the same number of errors for statements and questions (verb isn’t next to conflicting word, when further away still get errors, no effect of closeness conflicting info)
ARE/IS the key to the cabinets in my bag. (attracted to cabinets so want to say are not is)
This suggests that agreement is calculated BEFORE linear order is determined. (word order not enough to get grammatical sentences)
Further evidence for a separation between functional and positional processes
->agreement must be calculated before linear order
What determines whether you make an error?
Are we fooled by ‘fake’ plural nouns?
The player on the courts (more error)
The player on the course (fewer errors)
No, phonological cues play no role
Does the meaning of the noun phrase matter?
The label on the bottles
The baby on the blankets
Sometimes. It seems that if there are enough confusing cues to support a plural hypothesis then agreement errors are more likely.
Semantic cues can play a role. Phonological cues can’t. Positional cues don’t. Supports top-down model.
Incrementally in Sentence Production
We already said it is unlikely sentence production proceeds in discrete fashion.
Incrementality implies an assembly-line process. When something is ready at stage 1, pass it along to stage 2 and move on to the next bit.
We know we can start talking before whole sentence complete, which supports incrementality.
-have to be fast
-can come in different fashions
But what does incrementality look like for sentence production?
-two ideas about drive of sentence production
Lexically-driven grammatical encoding
-start with the concepts, retrieve on or other first then just kinda work with it, start with words and build words together and what ever words get dictates order
The order in which words are retrieved influences the eventual structure, not necessarily the other way around.
Grammatical encoder is incremental and the positional processor builds a structure that is compatible with order of retrieval and labeled functions.
eg The dog eats dry food
- retrieve dog first, because most important (lemma selection on dog) pass over to function assignment says thats my subject and then pass down to positional assignment which builds NP and will insert into subject position and insert the, next more lemma selection verb pass to function assignment, positional get VP and gets inserted ect.
- If dog had been tagged with a different function, or if it wasn’t first retrieved, another structure would be produced
-words come first, what ever words you retrieve will work with
Structurally-driven grammatical encoding
An alternative view is that the structure you want to produce drives further processing.
Structural slots influence order of retrieval of lemmas
->start with structure retrieve lexical information that fits structure
->syntactic priming supports this explanation more, structure seems to prime sentence production
Eg. Start with NP and retrieve word that fits with structure want to have (retrieve Dog)
Incrementality in Sentence Production
If you build your sentence around the availability of the words, does that mean you don’t know what you are going to say when you start speaking?
Evidence?
Sentences that begin with complex phrases take longer to initiate than sentences with complex phrases at the end
The dog and the kite move above the house
The dog moves above the kite and the house
Suggest whole sentence need not be planned before we start to speak.
-how complex NP is in relation to 2nd NP, 2nd sentence (simple 1st NP) is quicker to initiate than sentences with complex NP
->what happens at end (complex end) doesn’t effect speed, only beginning effects speed
-> incremental make up as go “on the fly”
Making choices on the fly
-experiment
Griffin & Garton, 2003
-picutre matching task
Can we speak fluently when making ordering decision on the fly.
They used a picture matching task.
A confederate describes a picture.
The participant then confirms whether s/he has the same picture (parodying it back) or corrects the description (making up themeselves).
Is the young boy feeding fish to the seal?
Is the young boy swimming?
When subjects generated a sentence themselves, fixation transitions between recipient (seal) and theme (fish) increased relative to repeating the sentence they heard. (shows decision making process while saying sentence)
Speakers looked back and forth between elements they needed to choose between, which was interpreted as indecision
Disfluencies also increased.
Signs of indecision support incremental planning.
Speakers start talking before they have the whole sentence planned.
Speaker starts with ‘The woman is feeding…’ and decides what to say next while speaking.
How incremental is it?
Summary
Clearly, we don’t have to wait until the whole sentence is planned
before we start to speak.
But, how soon can we start to speak and what drives the choices we
make?
Lexically-driven incrementality suggests that structure determined by
order of retrieved lemmas.
But, we also have to respect the grammar of the language and convey the
meaning we intended.
There must be interactions between fit-to-purpose structures and lemma
retrieval – structurally-driven incrementality
Structurally-driven incrementality is supported by syntactic priming, as
it shows that STRUCTURES do exist, can be primed and are
independent of individual lexical items.
Under this view, incrementality encourages the selection of lexical
items that could fill a particular syntactic slot in the sentence
