

The next essay is Theoretical Explanation. Sellars asks what we are doing when we explain by appeal to theories, and how the words for unobservable items get their meaning and earn their right to be believed. He keeps the rule we have been using. Tie ontology and meaning to the roles that terms play in integrated practices of measurement, inference, and intervention. In this essay he applies that rule to scientific laws, models, and theoretical terms.
Here's the idea. A fridge stops cooling. A technician explains the failure by saying the refrigerant is not expanding properly in the evaporator coil. Words like refrigerant and evaporator point to items you do not see directly when you open the door, yet the explanation guides tests and fixes. Pressure gauges are attached, a vacuum is pulled, a leak is found and sealed, the system is recharged, and the fridge returns to service. This is a model of theoretical explanation. We cite unobservables that figure in laws and mechanisms, connect them to observations through instruments and procedures, and use the package to predict and to intervene. On Sellars’ view, the meaning of the theoretical terms is given by the roles they play across these entry, intra, and exit links, exactly the pattern we saw with ordinary concepts.
Students learn that gas pressure in a sealed flask rises when the flask is heated. The theory says that molecules move faster at higher temperature and collide more often and harder with the walls. No one sees molecules with the naked eye. The term molecule is tied to a network of rules. It predicts pressure changes, diffusion rates, and spectral lines. It guides action. If you want to increase the rate of a reaction in the gas, you heat it or compress it. The theoretical term earns its meaning and its reality by this role, not by a private inspection behind appearances.
Sellars stresses the function of laws. A law is not a mere regularity. It is a constraint that supports counterfactuals. If the temperature had been higher, the pressure would have been higher, ceteris paribus, which means holding other relevant factors fixed. Laws also knit explanations together across domains. The ideal gas law links chemistry with thermodynamics. Ohm’s law links circuit design with materials. To see why this matters, compare two stories about why a steel beam bowed. A bare chronology says the beam was heated and then bowed. A theoretical explanation says that thermal expansion, constrained by the supports, created compressive stress that exceeded the elastic limit. That explanation can be used to predict bowing in other beams and to redesign the joint. That ability to be applied elsewhere is the mark of law governed roles.
He also explains how theoretical terms get introduced without circularity. One disciplined way is to state a theory in a form that treats the theoretical terms as whatever it takes for the claims about observables to come out true, a method often called Ramseyfication. In practice you replace the theoretical words with variables and say there exist things of some kind that, together with the observational claims and bridge rules, make the whole story true. Think of a recipe that says use ingredient X so that the cake rises and sets in twenty five minutes at one hundred and eighty degrees. If you later discover that X is baking powder, you have not changed the functional role the recipe specified. You have found a realiser of X. In science the role often gets a richer and richer realiser as theory improves, but the early role already ties the term to tests and uses in a public way.
Examples from medicine show how the pattern works across theory change. Before microbes were seen, physicians posited disease carriers to explain contagion. Later, specific bacteria and viruses were identified, and many carriers were given names and detailed structures. The terms for the earlier posits were not empty noises. They had roles in explaining spread and in guiding quarantine and sanitation. As microscopes, staining, and culturing improved, the theoretical story was refined and anchored to new instruments. The content of the terms deepened while the basic role remained. This is theory development without giving up the requirement that terms be fixed by practice.
Sellars treats models as tools rather than pictures. A ball and spring model of a crystal does not resemble the crystal in a photographic way. It encodes the roles that matter for prediction. If a spring constant is increased, the model says the vibrational frequency goes up. That guides inferences about heat capacity and thermal conductivity. A weather model on a computer works the same way. It does not depict air parcels as visible spheres. It encodes constraints and propensities. When forecasters adjust humidity and pressure fields and see a front sharpen, the model earns its keep. Its terms mean what their roles in the integrated pattern make them mean, and those roles are tested by the success of downstream predictions and actions, from issuing warnings to placing gritters.
He is careful about explanation by mechanisms and by unifying laws. A mechanism story tells how stages produce an effect, as when a catalyst lowers an activation barrier so that a reaction path opens. A unification story shows how diverse phenomena fall under one set of principles, as when Maxwell’s equations cover light, radio, and circuits. Theories can do both. In either case, the point is not verbal economy. It is integration that enlarges reliable inference and intervention. In a hospital, an oxygen saturation curve derived from basic binding laws lets clinicians infer how much oxygen a mask will deliver and what to expect when a patient’s temperature rises. In engineering, a single turbulence model lets designers move from wind tunnels to flight predictions. Theories are scorecards for what follows from what and for what to do.
Because theoretical explanation is role based, intertheoretic reduction becomes a controlled claim. Sometimes a higher level theory can be connected to a lower level through bridge principles that allow replacement of terms in limited contexts. Temperature can be reduced to mean kinetic energy under well specified conditions. Sometimes no neat reduction is available and we have layered descriptions that cooperate. The pressure of air in a football can be tied to molecular motion, while the bounce on a cold day is best described using the gas law and elasticity without reciting a molecular story. Sellars’ moral is simple. Keep your eye on explanatory roles, on what supports counterfactuals, on what connects to measurements and to doings. Do not insist on one level of story everywhere.
Two paired examples show how the method handles success and failure. First, plate tectonics unifies earthquakes, volcanoes, and mountain building. Theoretical terms such as subduction zone and spreading centre get their meaning from roles that tie seismic records, magnetised sea floor stripes, and GPS measurements into one package that predicts and guides hazard planning. Second, phlogiston once played a unifying role in combustion theory but failed to integrate with mass measurements and gas chemistry. Oxygen theory replaced it because it did more explanatory work with better fit across neighbouring practices. On Sellars’ rule, the commitment follows the explanation. Theoretical terms survive when their roles earn them a place in the web.
He also makes room for idealisations. Frictionless planes, point masses, perfect gases, and infinite populations are not lies. They are role setting devices that show which dependencies matter. A school experiment with a low friction object does not eliminate friction. It limits it so that the law of inertia can be seen. Theoretical explanation often proceeds by staging such controlled distortions, then adding corrections. Meaning and commitment are still fixed by the success of the organised practice. The point mass is not a tiny bead glued to the Earth. It is a well defined way to calculate trajectories that can be corrected for air resistance and size when needed.
Finally, Sellars links theoretical explanation back to intentionality. The content of a theoretical claim is the place it occupies in the inferential game of giving and asking for reasons, anchored to observations and actions by trained entry and exit moves. That is why lab protocols, calibration charts, unit standards, error models, and replication are not mere housekeeping. They are the scaffolding that fixes meaning and supports truth. When a spectrometer reports a line at a given frequency, the claim has content and can be right or wrong because it sits in that disciplined web. This is how unobservable electrons, fields, and genes become part of what we are entitled to say exists.
The upshot is a standard for good explanation. Use theories that enlarge reliable counterfactual inference, that knit diverse findings together, and that guide successful action. Let theoretical terms take their meaning from the roles they play in that success. Shift or abandon them when those roles cannot be sustained. Placed in the sequence of essays, this sets the stage for the next piece, which measures this account against instrumentalism and argues for scientific realism on the same role based grounds.