Lexical semantics can only take an NLP system so far, and the crucial ingredient in any metaphor-processing system is its array of conceptual representations. The field of cognitive linguistics, as exemplified by the work of Mark Johnson and George Lakoff [Johnson, 1987, Lakoff and Johnson, 1980], goes further still in emphasizing the role of world knowledge. A metaphor-processing system—whether human or machine—must possess an embodied understanding of the world: it must understand the mind’s relation to the body and the body’s relation to the world. These relationships are not literal, but are themselves governed by foundational metaphors that are grounded in our physical understanding of space, orientation, movement, and containment. These schematic structures–embodied concepts such as Path and Container and the metaphors built on top of them—are independent of language. Indeed, Lakoff and Johnson argue that metaphor is not primarily a linguistic phenomenon at all. Language is merely the stage on which metaphor does its most noticeable work.
The conceptual perspective sees metaphors as more than the finished products of creative language but as the very building blocks of thought itself. Complex world-views can be constructed from these conceptual metaphors, just as complex metaphors can themselves be constructed from simpler, lower-level metaphors. Grady [1997] argues that physical embodiment is the key to acquiring the latter kind of metaphor, which he dubs “primary.” For instance, from a young age we come to associate—or as Johnson [1999] puts it, conflate—affection with physical closeness and shared bodily warmth, and continue throughout our lives to express affection through physical touch. It is unsurprising then that we develop the primary metaphor (EMOTIONAL) AFFECTION IS (PHYSICAL) WARMTH, which allows us to talk of close relationships as warm relationships, of distant relationships as cold relationships, of giving someone “the cold shoulder” or “a warm welcome.” Primary metaphors are not arbitrary: rather, just as they are born of past physical experience they also offer a useful prediction of future physical experience. Thus, we can expect a warm welcome to involve a physical embrace, while we can also expect those who give us the cold shoulder to keep their distance from us.
So the physical world serves as a sand-box in which a cognitive agent learns to associate cause and effect and to fit cognitive structures to concrete reality. For instance, we learn that objects are usefully grouped into piles; that adding to a pile makes it bigger and higher; that taking from a pile makes it smaller and lower; and thus, more generally, that more of anything implies upward accumulation, while less of anything implies downward reduction. In other words, we learn that MORE IS UP and LESS IS DOWN. Likewise, our interactions with physical containers informs us of their affordances and limitations, and allows us to generalize a more abstract notion of containment that is conducive to cross-domain metaphorical reasoning. To appreciate the power of conceptual metaphors in a physical, non-linguistic setting, consider this tale of the Parsis and how they first came to settle in India. Time Out Guide to Mumbai offers a concise version of the legend:
They [The Parsis] arrived in Gujarat in the eighth or ninth century and sought asylum from the local king. He is said to have sent them away with a glass of milk full to the brim—his way of saying that his kingdom was full. The Parsi elders conferred, added some sugar to the milk and sent it back—to suggest that they would mix thoroughly and sweeten the life of the community.
So, with the right objects to hand, one does not need words to communicate with metaphors. The glass proves to be a remarkably versatile vehicle of meaning, for both the king’s message and that of the Parsi elders, because it instantiates a highly productive cognitive structure called the CONTAINER schema (see Johnson [1987]). The king uses the glass to represent a country: a country, like a glass, is a container, not just of liquids but of people and places and resources. Like a typical container, a country has physical boundaries that mark its extent; like a typical container, one can put things into a country or one can take them out again; and, like a typical container, a country can be full (of people) or it can be empty. Thus, the king uses a full glass of milk to convey the conceptual metaphor A COUNTRY IS A CONTAINER, but with the kicker “this one is full.” As the glass shuttles back and forth, it is used to carry subtly different meanings, each pivoting on the figurative affordances of the CONTAINER schema. The Parsis fashion their riposte from a different CONTAINER metaphor, LIFE IS A CONTAINER, one that can be figuratively filled with diverse events, relationships, and feelings. In contrast to the physical affordances of the COUNTRY IS A CONTAINER metaphor, which implies that countries are physical containers with physical limits on the amount of physical contents they can hold, LIFE IS A CONTAINER is a more abstract metaphor that places no such limits on its abstract contents. If LIFE IS A CONTAINER, it is a container with an unbounded capacity for emotional and cultural possibilities. The king and the Parsis are effectively playing a game of tropes, where the Parsis trump the king’s use of the CONTAINER schema with a more creative metaphor of their own.
The CONTAINER schema is ubiquitous in language and thought. We talk of minds as containers of beliefs (perhaps “bursting with ideas”), of lives as containers of events (allowing some to live “a full life”), of texts as containers of words, and words as containers of meanings. Our relationships can be figuratively conceived as containers, too (as in “there were three of us in that marriage”), although we most often anchor our metaphors for personal relationships in the SOURCE-PATH-GOAL (or SPG) schema. Thus, we talk of relationships that “are going nowhere,” of relationships that have “gone through a rough patch” and are still “on the rocks,” or those that have “hit the buffers” or even “come off the rails.” Purposeful activities of all kinds are conceptually anchored in the SPG schema [Johnson, 1987], allowing us to talk of “career paths,” of being “held back,” and of “getting ahead.” Schemas such as SPG and CONTAINER allow us to talk of the big intangible ideas, such as LIFE and LOVE, in terms of physical objects that we can see and grasp and follow. When said out loud, SPG-based conceptual metaphors such as LIFE IS A JOURNEY sound like the most hackneyed of clichés, yet, as Lakoff and Johnson [1980] demonstrate, they exude a powerful influence on the way we see and talk about the world. They are evident in the way our thoughts and our texts often cohere around a single unifying metaphor. Although conceptual metaphors are most apparent in our clichés, they underpin even our most creative efforts. Whenever we riff on a creative metaphor, as the Parsis did with the king’s figurative use of a physical container, we are instantiating and extending a conceptual metaphor.
Unsurprisingly, Conceptual Metaphor Theory (CMT) has significant implications for the computational modeling of metaphor in language [Lakoff, 1994, Lakoff and Johnson, 1980]. CMT-inspired computational models of metaphor range from symbolic approaches that focus on high-level data structures and knowledge-representations (e.g., Martin [1990]), to cognitive and neurological approaches that focus on lower-level arrangements of neuron-like activation elements and their interconnections (e.g., Feldman [2006]). But, regardless of which view of CMT is implemented, it is not enough that an NLP system goes beyond mere lexical semantics to embrace rich conceptual representations of target and source domains. It must embrace the right kinds of conceptual representations. Most theories of metaphor that lend themselves to a computational realization are ultimately agnostic about conceptual structure. Although they stipulate that such structures are necessary, and provide the general form of these structures—whether the set-theoretic genus/species representations of Aristotle, the graph-theoretic representations of Gentner’s analogical structure-mapping approach, or the textured category representations of Glucksberg’s category inclusion model—they do not make ontological commitments to the representations of particular concepts, such as CONTAINER, PATH, ORIENTATION, FORCE, etc. CMT argues that, for any agent to make and understand metaphors like a human, it must have the same conceptual biases as a human, and anchor its model of the world in the same foundational schemas. However, while these schematic concepts have an embodied basis in human cognition, this is not to suggest that an NLP system must also be embodied in the same way. Developmental roboticists may find some value in embodying a computational realization of CMT in an anthropomorphic robot, but this is neither necessary nor useful for a metaphor-capable NLP system. Such schemas are the conceptual products of physical embodiment,
