What is referred to as the rules for determining meaning of words in a particular sentence or phrase?

Psycholinguistics

• Overview
• Design Features of Language
• Grammatical Theory and Language Structure
• Finite State Grammers
• Phrase (Constituent) Structure Grammers
• Chomsky's Transformational Grammar
• Evidence for Innate Origins
• The Perception and Comprehension of Speech

OVERVIEW

• Competence refers to the fact that by age four, every normal child learns at least one of the worlds many different languages -- and does so in the presence of weak or nonexistent instruction and in the process masters some quite subtle rules of grammar.
• Productivity refers to the fact that unlike animal communication which is stereotypic, human language permits an infinite number of rearrangements of its elements.

One place to begin might be to look at the characteristics or design features of language. Some will be similar to systems of animal communication, others will not. Three major topics in psycholinguistics are:

• semantics -- the study of meaning;
• syntax -- the study of the rules for combining the the smallest acoustic elements (phonemes) into the smallest elements of meaning (morphemes), and morphemes into utterances or sentences;
• pragmatics -- the study of the social nature of language.

Return to Top

Hockett's 13 Design Features of Language

1. VOCAL - AUDITORY CHANNEL

Can maintain other, more directional, activities while engaging in language acts.
2. BROADCAST TRANSMISSION - DIRECTIONAL RECEPTION.

Goes out to everyone but received from a particular direction.
3. RAPID FADING

Exists only briefly and thus has a social consequence.
4. INTERCHANGEABILITY

Competent speakers can reproduce any message they can understand and the content is left undisturbed.
5. TOTAL FEEDBACK

We hear everything we produce as we are producing it. Indicates a highly controlled process where feedback is critical.
6. SPECIALIZATION

Specialized for communication -- not to some other activity or purpose.
7. SEMANTICITY

Linguistic sounds mean something.
8. ARBITRARINESS

No inherent relationship exists between linguistic sounds and the objects they represent.
9. DISCRETENESS

Language consists of small, separable units of sound (phonemes) each of which has an identity and can be combined to form larger units (morphemes).
Phonemes are the smallest differentiable units of sounds in any language. They are relatively small in number. English has about 40 - 42. Phonetics is the study of the rules that determine how the phonemes can be combined.
Using these rules, phonemes are combined to form morphemes which are the basic units of meaning (mostly words but also units like prefixes and suffixes -- called bound morphemes).
10. DUALITY OF PATTERNING

An infinite number of meaningful words can be created from a small number of phonemic building blocks. The words 'team' and 'meat' are built from the same units.
11. TRADITIONAL TRANSMISSION

Language is passed along culturally from one generation to the next.
12. DISPLACEMENT

Like thought, language permits the consideration of concepts unrestrained by time or place, e.g. "Long ago, in a galaxy far far away."
13. PRODUCTIVITY

An infinite number of novel arrangements of words can be uttered and comprehended because of an underlying lawfulness based on a finite set of rules.

Grammatical Theory and Language Structure

OVERVIEW

FINITE STATE GRAMMARS

(click icon for demo) 

(copyright F. G. Fidura, 1995)

Suggests that the rules for syntax are governed by finite state rules -- the next word choice at any point or state in a sentence is determined by consulting a finite numbers of candidates and picking one of them.Overview ! fntovr.bul

ASSUMPTIONS OF FINITE STATE GRAMMARS

• Humans produce sentences on a word-by-word basis. Having chosen one word, the person consults some mental list of acceptable choices and produces one of them.
• Human grammar consists of constraint rules that limit the word choices that can be made at any point in the production of a sentence
• The choice made at the left-most word of a sentence constrains the choice made at the next left-most work, and so on, until the sentence is completed.

(click icon for graphic) 

• The likelihood that one word will follow another, i.e., moving from one state to the next, is called a transitional probability.
• "Hand me the salt," has a high transitional probability. "Hand me the dog," has a lower probability. "Hand me the mountain," has a probability of 0.

• Behavioral research was quite supportive of the theory. It fits nicely with the general stimulus-response theories of learning then in vogue.
• The pronunciation of each word in a sentence is a response. Because of total feedback, the hearing of a word acts as a discriminative stimulus, which cues us to produce the next word in the sentence.
• At the end of the sentence we are reinforced by our listener's compliance.
• This approach reduces the learning of a sentence to that of learning a chain of stimulus-response associations.
• Since there was considerable evidence that animals could learn lengthy S-R chains, and because children must certainly be capable of learning anything a rat can learn, finite state theories were thought to be the answer to the learning of syntax.

OBJECTIONS TO FINITE STATE THEORY

1. The theory was clearly incomplete. It was unable to handle the fact that any competent speaker can imbed any syntactic structure in an already well-formed sentences:

The boy ran to the store
The boy ran quickly to the store.
The boy ran quickly but with purpose to the store.
The boy, who by now was sweating, ran quickly but with purpose to the store.
It means that the assumption that word choice is constrained by previous word choice cannot be true. Clearly our knowledge of sytax consists of more than just simple chaining rules.
2. Judgment of grammaticality should depend on transitional probabilities, the number of times the words have been paired in the past. Consider:

Colorless green ideas sleep furiously
Was he went to the newspaper is in deep end
The first is universally judged as more grammatical even though TPs = 0.
3. If comprehension depends on association, how would productivity be possible? How could we understand sentences we never heard before.
4. Language acquisition under this theory would probably be impossible -- certainly by age 4. A finite state grammar with only seven states would require over a google (10 to the 100th) possible candidates. There simply isn't time to learn that many.

Return to Top

Phrase (Constituent) Structure Grammers

OVERVIEW

• Phrase structure grammars were originally developed by the linguist Leonard Bloomfield near the turn of the century. Unlike finite state grammars which are arranged left-to- right, phrase structure grammars are organized hierarchically -- that is, from the top down.
• Grammaticality is to be found in rewrite rules which allow us to map words onto the various parts of speech and string these together into valid phrase structures and ultimately into sentences.

ASSUMPTIONS OF PHRASE STRUCTURE GRAMMARS

• Perhaps it might be best to begin with an example. Consider the sentence:
• "The boy will hit the ball."
• Notice that it has the general form of a simple english sentence, namely:
• Subject --> Action --> Object
• The above elements are generally represented in english by: Noun phrase Verb Phrase Noun Phrase
• The above phrases are made up of the smaller units:
• Det Noun Aux Verb Det Noun
• The boy will hit the ball.
• Notice that we could have started at the lowest level by first assigning the words to their parts of speech and then grouping these into their phrase structures and then the phrase structures into a sentence.
• The structures we have been describing can be converted to a treelike depiction called a phrase marker. The following figure shows the phrase marker for this sentence:

(click on icon for figure)

(from Best, Cognitive Psychology, 1995)

• So grammatical knowledge consists of knowing a relatively small number of phrase structures and the rules for mapping words onto the parts of speech and the rewrite rules for forming the parts of speech into phrase structures and then phrase structures into sentences.

• Language acquisition is a much simpler task. Children only need to learn a relatively small number of phrase structures and the rewrite rules governing their construction.
• Accounts for judgments of grammaticality. Even nonsense sounds grammatical as long as it follows the rewrite rules. Consider the following:
• "Gelax frimaged a leeble."
• The abstract nature of phrase structures. Grammar is not tied to particular words or transitional probabilities. We can accept as grammatical, sentences we have never heard before.
• Notice that the theory responds to both questions of competence and productivity.

OBJECTIONS TO PHRASE STRUCTURE GRAMMARS

• Objections to phrase structure grammars are raised in two areas: a) some kinds of linguistic knowledge are not accounted for by the theory; and b) ambiguities.
• As an example of the first objection, note that meaning should be changed by a change in structure and often it is not. Consider the following:
• "Sarah parked the car."
• "The car was parked by Sarah."
• On the other hand, similar structures should yield similar meanings and they don't. Consider the following:
• "John is easy to please."
• "John is eager to please."
• Finally there is the problem of ambiguities, identical structures which can be interpreted in different ways. Consider the following:
• "They are cooking apples."
• "The lamb is too hot to eat."
• "Visiting relatives can be tiresome."
• "Time flies like an arrow."
• Notice that the meaning of the first can be resolved by establishing the correct phrase marker.

(click icon for figure) 

(from Best, Cognitive Psychology, 1995)
This is not true of the other three and it was this fact that led Noam Chomsky to propose tranformational grammar.

Chomsky's Transormational Grammar

OVERVIEW

• In the first type, the meaning is made clear by determing between the two possible phrase markers which represent the sentence.
• In the other cases however, this is not possible -- there is only one possible phrase marker.

(click icon for figure) 

(from Best, Cognitive Psychology, 1995)

ASSUMPTIONS OF TRANSFORMATIONAL GRAMMAR

• The first of these is the SURFACE STRUCTURE -- a level of grammatical knowledge that is seen in the superficial appearance of the sentence (as actually spoken or written).
• Instantiated in rewrite (transformational) rules.
• The second of these is the DEEP STRUCTURE -- a level of grammitical knowledge that is closely related to the meaning of a sentence and is instantiated in the phrase structure rules.

As evidence, notice that our problems in extracting the meaning of the "visiting relatives" sentence is not a result of selecting the wrong phrase structure as was the case with the "cooking apples" sentence. Rather, the ambiguity is produced because certain elements of the sentence's deep structure have been left out of the surface form.

• So, phrase structure rules (which govern single elements) operate on KERNAL sentences in deep structure.
• These kernal sentences can be transformed in a number of ways using rewrite rules to produce a variety of different surface structures.

(click on icon for figure) 

(from Best, Cognitive Psychology, 1995)

THEORETICAL ADVANTAGES OF TRANSFORMATIONAL GRAMMAR

• As an extension of phrase structure grammar, it has the same advantages in terms of judgments of grammaticality.
• In addition, it accounts for ambiguities. These occur because phrase structure rules operate on the deep structure and may not be apparant at the surface level.
• Also accounts for why sentences with similar structures are interpreted differently and those with different structures are interpreted the same.
• Since the theory suggests that in comprehension, we reduce surface structure to its simplest kernal form in deep structure, it accounts for the literature on memory for prose, processing and remembering negatives, etc.

CHOMSKY'S VIEWS ON THE ORIGINS OF GRAMMATICAL KNOWLEDGE

• While experience plays a crucial role in language acquisition (remember traditional transmission), Chomsky believes that much of our grammatical knowledge is based on innate predispositions.
• These guide and channel our experience and make us sensitive to the regularities of phonology, syntax, and discourse.
• Children are innately predisposed to notice orderly patterns in the language acts going on around them.
• Chomsky believes that, much like little scientists, they form hypotheses about the rules or laws underlying the patterns they are experiencing.

(click icon for figure) 

(from Best, Cognitive Psychology, 1995)
• They test these hypotheses and continue to observe and thus refine them over time. (Consider the regularization of irregular verbs as an example).
• Ultimately they end up with an accurate hypothesis and develop a good working knowledge of language.

Evidence for Innate Origins

OVERVIEW

• There is a substantial body of literture suggesting that our linguistic knowledge is based on innate structures and predispositions.
• By innate linguistic predispositions we do not mean the language per se or even grammar. Rather, there is a blueprint or mental template which children bring into play to analyze, understand and learn language in the world around them. It's as if we were born with this mental language organ that goes out into the social world looking for lawfulness and expecting there to be meaning in the utterances around us.
• Learning occurs more readily than it would if children were unprepared for the gabbing sounds which emerge from humans.

ANATOMICAL AND BREATHING SPECIALIZATIONS

• Humans exhibit dentition less specialized for cutting and tearing of food (as in other primates) and more specialized for articulation. In conjunction with the tongue, permits the control of air flow necessary for the formation of consonants.
• The same can be said for our short stubby tongue which is more mobile than in other primates. This flexibility allows for the production of vowel sounds.

SPECIALIZATIONS IN THE BRAIN

• As far back as the 1860's Paul Broca discovered that if a certain area of cortex was damaged, an aphasia or language disorder developed.
• The areas of the brain which exhibit this characteristic are quite specific, as are the deficits which result.

CATEGORICAL PERCEPTION OF SPEECH SOUNDS

• There is evidence that babies clearly distinguish between human speech sounds and other sounds.
• Babies immitate the sounds of human speech but not other sounds present at the same time.
• There is good evidence that this ability to discriminate speech sounds from other sounds appears as early in life as two weeks.

OTHER GENETIC TEMPLATE MODELS

• There are a number of other examples in the animal world where nature provides a predisposition in the form of a blueprint or template and experience provides the specific information.
• Some classic examples are imprinting in waterfowl and gallinacious birds, attachement in primates, and species song in some songbirds.

The Perception and Comprehension of Speech

Having established that the basic discrete acoustic unit of language is the phoneme, speech perception should simply reduce to a matter of:
1. Mapping the acoustic signal onto known elements of phonetics
2. Determining the phoneme that has been uttered
3. Storing that phoneme in working memory
4. Add to the next phoneme when uttered until morpheme is established
But what we've described above would sound like the synthetic speech produced by computers. Speech doesn't occur in discrete units. This presents a number of problems:
1. Speech sounds are continuous and non-discrete -- how do we pick phonemes out of this stream?

Consider the sentence: "Ja eet?" or even worse "Jeet?"
2. Patterns with slightly different acoustic properties are interpreted the same.

Consider the m in "Tim" and "mink"
3. Patterns with identical acoustic properties are interpreted differently.

Consider "writer" and "rider"
4. Order is meaningful -- requires identifying other phonetic segments

Consider "pill" "lip"
The overall problem is sometimes referred to as the segmentation problem -- Because speech resembles a warbling siren, with hardly any pauses, how is the listener supposed to know where to put the boundaries around phonetic segments to identify them?

But it's clear that they are interpreted somehow and that their use is governed by rules. Could their interpretation be based on acoustic cues? You already know that phonetics consists of the categorization of basic speech sounds. But phonetics can be approached on the basis of the physical properties of the sound produced. This is called acoustic phonetics and what we've been describing so far.

Articulatory phonetics, on the other hand, is the branch of phonetics that studies the nature of the formation and production of these sounds. Sounds are produced through what are called "articulatory gestures" that are determined by such actions as the placement of the tongue, control of airflow, and changes in the vocal cavity.

There is one additional topic we need to articulate (please forgive the pun) in order to begin to understand speech comprehension -- phonology. Phonology is a branch of linguistics that attempts to determine the rules, or principles that characterize the production and comprehension of speech sounds. Phonologists don't deal with the sounds directly, rather, they are interested in the more abstract aspects of speech sound -- the general knowledge a person has that enables her/him to formulate specific utterances. Consider the following example:
• Notice the lawful change in the articulation of the letter "c" when we change from adjective to noun
• electric -- electricity
• egocentric -- egocentricity
• I know a man who quite bustric -- he exhibits lots of bustricity
• In point of fact, bustric is a word I just invented, but you would have no difficulty producing or precessing the above sentence because it conforms to the rules

The English language contains somewhere between 40 and 42 (depending on who you read) phonemes (basic speech sounds). They are shown, along with examples in the following table:

(click for table)

Articulatory gestures can be divided into two broad categories:

• Those that produce vowel sounds
• Those that produce consonants
Perhaps the difference in how thay are produced and used provides for the marking function in speech comprehension for which we are searching. Consider:
• Consonants -- usually involve some degree of constriction of the oral cavity generally accompanied by some movement of the tongue
• Vowel sounds -- open oral cavity generally static involving little movement of the tongue
• Speech proceeds by the production of syllables -- constructed by embedding vowels in strings of consonants. So it begins with constriction and movement followed by a short period when the vowel is sounded and completed with a consonantal constriction
So it is this syllabic structure that allows us to mark and process discrete units in the continuous stream of speech. While this answers the question about processing discrete units, it still begs the question: How do we identify the phonemes we are hearing in a way that allows us to build them into morphemes? It turns out that the articulatory gestures associated with any given phoneme generate a kind of bundle of distinctive features more or less different from all other phonemes. "b" and "x" would be an example of more while the "b" and "p" sounds would be an example of less different.

Consonants vary in three ways

• Place of articulation (there are seven of them)
• Manner of articulation (there are six of them)
• Voicing -- accompanied by vibration of vocal cords -- the "m" sound is voiced, the "t" sound is not
If your eyes have not yet glazed over, some of these differences are summarized in the following tables:

(click on table)

A two-way classification has been done for vowels defined by the position of the tongue and the part of the tongue involved. This is summarized in the following table:


(click on table)

The articulatory gestures associated with each of the phonemes yield a bundle of more or less distinctive features that characterize each of the phonemes in a unique way. The "b" sound is more like the "p" sound and less like the "x" sound, but nevertheless unique. These distinctive features can be seen in the following table:


(click on table)

We have presented this lengthy discussion not to put you to sleep nor to display my erudition in the area of phonetics, but rather to provide the tools to begin to understand speech comprehension and now we're ready to do that (finally).

Pisoni (1978) suggest that speech perception seems to take place in a series of independent stages, each of which modifies and elaborates the code produced by the preceding stage. The basic components (as interpreted by Best, 1995) are:
1. Auditory stage.

At the auditory stage, the acoustic signal is converted into a neurological representation that preserves various features of the physical signal. For example, feature analysis at the auditory stage is presumed to encode the sound's fundamental frequency, as well as some details of its harmonic structure. In addition, in the auditory stage, a code representing the signal's overall intensity and duration is produced. This code is presumed to be stored in some form of sensory storage, and for this reason, the code at the auditory stage is sometimes called "raw." At this point, no phonetic or phonological information has been extracted from the signal.
2. Phonetic stage.

The main purpose of the phonetic stage is to name the speech sounds correctly, that is, to assign phonetic labels to the speech signal that are congruent with the speaker's intentions. Here, the listener faces a major difficulty known as the segmentation problem, which was alluded to in an earlier section of these notes. The segmentation problem can be stated this way: Because speech resembles a warbling siren, with hardly any pauses, how is the listener supposed to know where to put the boundaries around phonetic segments to identify them? Fortunately for us, speech is constructed by syllables. This means that the influences that vowels have on the production of consonants will probably be most often limited to a range of one syllable. The implication is that some perceptual mechanism must be set, or tuned, to look for patterns of alternating constriction and openness, which are then categorically boxed into syllables whose phonetic names are subsequently determined.
3. 3. Phonological stage.

At the phonological stage, the phonetic segments that have just been identified are mapped onto underlying (more abstract) phonological rules that extract the true essence from the phonetic segment. This true essence refers to information about the phoneme that permits its other features to be computed from knowledge of phonological rules. As we saw, for example, knowledge of the two-consonant rule enables a person to compute the identity of the first consonant in a two-consonant sequence. Thus, if /s/ happened to be misidentified as /§/ in the phonetic stage, then cognitive processes in the phonological stage would correct this error if the next segment identified was another true consonant.
It is important to recognize that top-down processing and pragmatics also play a role in speech comprehension.

Communication through spoken language frequently occurs in "noisy" circumstances where the phonemic elements may very likely be lost. In addition, the high rate of processing required to interpret continuous speech would make it a most difficult if not impossible task. Clearly, bottom up processes that act on the acoustic signal cannot do the whole job of speech perception. Fortunately, english, like most languages, is highly redundant. Synthesizing elements we have already stored provides a context to replace lost elements even as we analyze the incoming signal.

Consider the following sentence (suggested by Best, 1995):

"She did really well on the test (cough) in fact, (cough) got the highest grade in the (cough)."
Notice that context allowed you to establish the meaning of the sentence.

Evidence for the role of context is provided in a study by Pollack and Pickett (1964):

• They covertly recorded people in spontaneous conversations. The Ss gave every indication that they understood one another -- they replied to questions, laughed at jokes -- all appropriately.
• The researcher then cut the tapes to make recordings of isolated words.
• The tapes were played for subjects who were asked to identify single words.
• They were successful on only 47% of trials
• The researchers then added larger and larger segments of the original tape. As they did so, they found that identification accuracy slowly improved until a certain critical point was reached. At this critical point, accuracy dramatically improved.

A related study defines what is referred to as the Phonemic Restoration Effect.
• Warren (1970) presented 20 Ss with a tape on which the following sentence had been recorded

"The state governors met with their respective legi*latures convening in the capital city."
• The asterisk marks the point were .12 second was chopped out of the original utterance and the recording of a cough substituted in its place.
• Ss were asked if they detected any sounds missing from the recording. 19 of the Ss said no. The Ss did detect the presence of the cough but could not locate it correctly.
• Later studies found that a substantial part of a word could be removed without destroying the illusion.
• In addition, it was found that a tone or a buzz could be substitued without Ss noticing that any sound were missing. However, Ss were quick to detect a silince and they were also accurate in reporting its placement.

Best (1995) suggests that the stream of speech is an extremely ambiguous signal.
This stream offers some acoustic information that is quickly analyzed and categorized into its configuration of features. Probably at the same time these bottom-up cognitive processes are at work, top-down processes commence their operation, helping the individual infer, or fill in, missing or undetected speech information. (Best, 1995, p. 320)
Liberman summarizes these ideas by saying:
Some of the distinctive features that specify each phonetic segment probably can be determined from the available acoustic signal. Other distinctive features cannot be uniquely identified. The listener therefore forms a hypothesis concerning the probable phonetic content of the message that is consistent with the known features. However, he cannot test this hypothesis for its syntactic and semantic consistency until he gets a fairly long segment of speech into his temporary processsing space. The speech signal therefore remains unintelligible until the signal abruptly becomes intelligible. The acoustic signal is, of course, necessary to provide even a partial specification of the phonetic signal. However, these experiemts indicate that in many instances the phonetic signal that the listener, "hears" is internally computed. (Liberman, 1967, p.165)
The analysis by synthesis model was originally proposed by Halle and Stevens (1964).

Consider the following sentence:
"It's getting really hot in here, don't you think?"
• Is it simply a statement of fact that one might mention to a friend in a hot auditorium while waiting for the show to begin?
• Is it a subtle request to open the window or turn down the heat as might occur when visiting the home of a friend?
• Is it a statement to a colleague that describes the emotional atmosphere at a meeting where passionately-held political positions are being debated?
It is, of course, the social context that defines the meaning of this utterance. Pragmatics refers to the social rules underlying language use and the strategies used by speakers to establish coherence across several sentences (Best, 1995). Language is a social instrument involving a social contract. Not only does it occur within a social context that may give precise meaning to the utterance, it also has rules and conventions that one is expected to follow.

One convention that governs speakers in a conversation is the one meaning convention or the given-new strategy. It assumed that a speaker has only a single semantic intention -- that utterances generally contain given (understood) information. If the speaker has more than a single intention, he or she is expected to give some other clue to start looking for other meanings. The given meaning must be there to serve as a point of departure to integrate the new information or meaning that is given.

There are also social conventions about how conversations are to be staged. One very powerful rule is that participants in a conversation alternate in opportunities to speak. This rule is established very early on in language development. One of the most delightful elements about the egocentric conversations of young children is that they carefully alternate even though talking about totally different things. Their egocentricity is defined by not having yet learned the given-new strategy. They nevertheless, carefully alternate opportunities to speak.

Conversations are seen as being divided into three phases:

1. Initial Greeting

One or more stock expressions rarely taken seriously: "How are ya," "What's happenin'," etc
2. Content

Here the parties exchange meaningful information using the conventions given above
3. Closing

Conversation winds down, informational issues are resolved and closed, a stock closing expression is used. As in:

What do we call a set of rules for combining words into sentences?

What Is syntax and grammar?

What are the rules of syntax?

Is defined as the meaning of words and sentences in a particular language?