What is an image schema? It is a way of structuring a scene, a way of structuring spatial relations. You see image schemas in the prepositional system in English. Another language from North-west Mexico, Cora, has 137 deictic locatives. In this system, one basic image schema is encoded in each phoneme in deictic locative terms. For instance, /m/ indicates medial distance, /u/ indicates the object is in fthe filed of vision and /x/ indicates the object is on the slope. So /mux/ means "in the medial distance, which is visible and is on the slope. Len Talmy also gives the example of a Northern California language, Atsugewi. This language has a number of suffixes which indicate ground geometries and paths, such as "into a liquid," "into a fire," and "into a corral or field." A full list of these suffixes can be found on p. 15 in the Talmy article from the reader. The important point about these examples is that the image schematic structures around the world are the same, but they are organized in different ways. For instance, the container schema and source/path/goal schemas are found in language after language, but are combined with other schemas differently.

Talmy claims that closed class morphemes tend to have image schematic structure and they are automatic. Speakers learn to use them automatically and unconsciously as part of the semantic system of the language, but the primitives that make them up appear to be universal. In this course we ask how can these primitive concepts be computed neurally? What part of the brain is used to do it and what kind of neural structures are needed to do it?

Image schemas are structures which mediate between what you see, your perceptions and your conceptual structure. Image schemas have their own logics. For instance, if you travel from point A to point C, you have reached all the points between A and C as well. This is an inference based on the source/path/goal schema. If A is contained in B and B is contained in C, then A is contained in C. This is the formal logic inference, modus ponens, in which categories are metaphorically understood as containers. Also, if A is not contained in B and C is contained in B, then A is not contained in C. This is modus tolens. These are spatial inferences based on the container schema. This kind of reasoning about categories done in formal logic is built into the spatial reasoning system.

Magnitude is important for certain kinds of reasoning about space, but spatial relations terms, such as prepositions, are independent of magnitude. They are topological. Knowledge of the world sometimes adds to our understanding of terms, such as 'to'. For instance, "going to Thailand" and "going to a neighbor's house" indicate different distances and forms of travel, but that kind of reasoning is based on real world knowledge, not on the spatial relations schema indicated by 'to'. Different part of language have different inferential structures. For instance, different things may be coded in verbs. Talmy's examples of spatial relations morphemes in Atsugewi are part of the verbs. Also, English codes information about the manner of an action in the verb, but Spanish doesn't.

Image schemas are conceptual primitves, but they have internal structure. For instance a container schema has a boundary, an interior and the exterior. Also, you have to know what part is activated and you need a figure/ground relation. This is called gestalt structure, rather than building block structure. You can't have a boundary without an interior, exterior, etc.

One problem involves how the spatial relations terms combine with the meaning of the verb? For instance, how do you understand the combination "pick up." It's not clear how people do this. How can you model it? Is the model able to explain the phenomenon cross-linguistically?

We also don't know how to model reasoning with spatial relations. Regier's model show how we can learn to recognize and name spatial relations, but it doesn't do reasoning. We do reasoning with the same structure which allows us to perceive spatial relations.

Spatial relations are not in the world. For instance, what counts as 'in' in "the bees are swarming in the garden." For learning these relations we need some innate neural structures, but some of these universals are learned by everyone through the environment. There's no simple-minded distinction between what's innate and what's learned. Universal spatial relations may not show up in the closed-class spatial relation systems of all languages, but they would probably show up elsewhere in the language. For instance, in the Guguu Ymithirr, the spatial coordinate system uses cardinal directions, but the language hasn't been studies to see if image schema universals are used in other parts of the language.

When linguists look for features which are "universal," they can't look at every language in the world, so they look at several language from different language families. If they find the feature in these unrelated languages, it is likely to be universal.

Is there any evidence for image shcemas in lower primates? This is hard to tell because the main evidence for them is linguistic. Animals have spatial abilities, so the question is whether or not the animal uses spatial schemas to comceptualize other things. A new book,The Evolution of Communication, by Hauser addresses this issue.

In sum, there are several things to remember about image schemas. They often show up in closed class morphemes. There are primitives which combine and fit together into morphemes in different ways in different languages. There seems to be a universal set from which they are chose, but we don't know if they are all used in every language. There are used for reasoning. They have a gestalt structure, as opposed to a building block structure.

There are different types of image schemas. Some are topological, for instance the container schema, contact, center/periphery, path. There are orientational schemas, such as front/back, up/down, left/right. There are also force schemas.

Deixis is the structuring of space relative to a reference point. If the speaker is the deictic reference point, the a term such as "I" refers to that person, and "you" refers to a hearer. If a new person becomes the speaker, "I" refers to that person, and the decitic reference point has shifted. Other words which have to do with deixis include "this," "that," "here," "there," etc. These terms are defined relative to the deictic reference point. Similarly verbs such as "go" and "come" are understood relative to the location of the speaker. Also a phrase, such as "I am sitting behind the desk" is interpreted according to a particular viewpoint, the hearer's viewpoint. The desk is interpreted as having a "face" which is facing toward the hearer. In English we assign front to things which don't have inherent fronts by projecting a body onto the object as if it were facing us. In other languages, the front is assigned to the part of the object facing away from the deictic center, as if there were a person facing away from them. We could assign a "front" to an object based on other frames; for instance, if we normally interact with a particular side of an object, then we are likely to name that the front. So we interact with the "front" of a computer. Also, we often assign the "front" to the side of an object which is toward the movement of the object. For instance the front of a car is the part which is in the direction of movement. When the movement is non-prototypical, such as backing up, there are conflicting frames of reference which we could take. For instance, in such as situation we might say, "Don't back into the front of that tree." How can this phenomenon of deixis be represented neurally?

Talmy had also observed a phenomenon he calls "fictive motion." In a phrase such as "the roof slopes downward," the roof doesn't move, but it is conceptualized a movement tracing along the line of the roof. Another example: "the road ran through the woods." Similarly, we can conceptualize something moving as a static line, for instance "he walked in a circle around the building." In this example the person's movement is conceptualized as a reified path. This is likely a universal feature of language. How does the brain do something like this? In English, we can characterize the shape of paths in fictive motion as well, for instance, "the road zig- zags through the woods."

This will give you some sense of the complexity of image schemas and their relationships. There is a systematic relationship between a path which goes to an endpoint and being at the endpoint. For instance, "he walked across the street" and "he lives across the street." So we can understand location as being the endpoint of a path. This is an example of the kinds of ways we structure space. Not all spatial concepts are structured the same.

Frame Semantics

 In Europe in the 40s and 50s, there was a theory of semantics having to do with words and their relationships which is called lexical field theory. These theorists studied certain related sets of words, such as day of the weeks. They noticed that there were certain relationships between these words in these sets. Other lexical fields include kinship terms, eating utensils, etc. In the mid-70s, Fillmore observed that to really understand the relationships between these words, you had to understand the structure underlying the fields. For instance, to understand the relationships between words such as "buy," "sell," etc., you have to understand the commercial event frame. In any frame, there are participants. In the commercial events frame, the participants are 'buyer' and 'seller'. Other entities are 'money' and 'goods'. In frames, there are also scenarios which have three states. In the commercial event frame, the initial state is:

Buyer has money and wants goods.
Seller has goods and wants money.
The middle state is an exchang:
The Buyer gives money to the Seller.
The Seller gives goods to the Buyer.
In the final state:
The Buyer has goods.
The Seller has money.

The predicates in the frame, such as 'have' and 'want', are simpler than 'buy' and 'sell', which are defined relative to the frame. Notice that there is a causal structure, a linear order structure and other types of structure in the frame. There are also inferences you can make given the structure. For instance, if you say "John bought a book from Harry," you can infer that John owns the book now and that Harry has more money now. The idea is that lexical items get their meanings from frames, from the overall structure of the frames. The frames relate all the lexical items in a lexical field. Frames also define semantic roles of the participants and entities, such as agent, patient, source, etc. In the phrase, "John bought the book from Henry", John is the agent and goal; Harry is the source; the book is the patient.
 

References:

 Reader:  5, 6, 7, 14