On ambiguity

We perceive something as ambiguous when it allows for two or more plausible interpretations ¹.

In progress

This is an ongoing exploration of a topic that I am still updating periodically.

What does this button do? How did the audience respond to my talk? What did the ending of the movie mean? Why did they wear that particular outfit? What is causing that smoke in the distance?

This page includes some thoughts about ambiguity in different domains.

Ambiguity in language

The causes of ambiguity are not restricted to language, but types of ambiguity are well-defined in that domain, so it’s a good place to start.

Lexical ambiguity occurs when a word has multiple meanings. In the case of homonyms, a word can have multiple, unrelated meanings (river bank, money bank). In polysemy, a word can have multiple related meanings (tree branch, git branch).

Syntactic ambiguity arises from an arrangement of words that generates multiple possible meanings: the program loads files from the server during startup (startup of the program, the server, or both?).

Semantic ambiguity comes from a sentence with multiple meanings, which can be due to many factors: update the app after the window closes (does the user close the window or does it happen automatically?).

Endophoric ambiguity is due to a reference that could refer to multiple independent parts: from the example above… (which of the preceding examples?).

Some of these cases are related; for example, lexical and syntactic ambiguity can contribute to semantic ambiguity.

Ambiguity in discovery

Ambiguity can be understood as a necessary stage in the activity of discovery.

Scientific discovery is often initiated by the recognition of an ambiguity — a phenomenon that admits of multiple possible interpretations. This prompts the scientist to seek the interpretation that has the highest probability of being correct.

Great scientists seek an ambiguity with just as much vigor as they do the resolution of that ambiguity.

Ambiguity can arise in different phases of learning and discovery:

• Identifying a new ambiguity to investigate (increased ambiguity)
• Accepting a subset of valid interpretation out of several possible interpretations (reduced ambiguity)
• Introducing ideas that challenge a prevailing view (restored ambiguity)

Ambiguity in technology

Technology can be ambiguous in different ways. There can be ambiguity of:

• Function: multiple interpretations of what it does
• Procedure: multiple interpretations of how to use it
• Mechanism: multiple interpretations of how it works

A technology may have multiple valid functions and different valid procedures for its use. This can result in that technology being experienced as ambiguous by the user, even if the design is intentional and useful. As one becomes skilled, the sense of ambiguity may be replaced with a sense of possibility.

A technology is discoverable when it enables the resolution of its own ambiguity. This may coincide with the user recognising the intention of the designer.

[The] machine’s complexity was tied less to its esoteric technical characteristics than to mundane difficulties of interpretation characteristic of any unfamiliar artifact. ²

Ambiguity in biology

Enzymes were classically considered to be specific to one substrate, a kind of lock-and-key relationship. In other words: an enzyme had a one-to-one binding relationship with its substrate. It is now known that some enzymes exhibit what scientists call “substrate ambiguity”, in which an enzyme can potentially interact with multiple different substrates.

When an enzyme has been demonstrated to bind with multiple enzymes then that is its real behaviour. If a new enzyme exhibits substrate ambiguity then it is ambiguous for the observer, who may have expected more narrow binding behaviour because of their model of reality. Forcing the enzyme’s behaviour into the classical lock-and-key model to reduce ambiguity could lead to a flawed account of reality.

Ambiguity and generalisation

Both ambiguity and generalisation involve one-to-many relationships:

• Ambiguity: one perception with many possible interpretations
• Generalisation: one class with many possible instances

The distinction is that the interpretations in a case of ambiguity are mutually independent, whereas the instances in a generalisation are members of some class.

A generalisation assumes some model that explains the class and its membership. Without an understanding of the model, a generalisation may have the appearance of an ambiguity.

Ambiguity and indeterminacy

An indeterminate expression refers to a thing without specifying all of its properties ³.

You need a Windows 11 machine to follow this tutorial.

— does not specify whether the Windows 11 machine is physical or virtual

Click “Save” to persist the data from your session.

— does not specify where the session data is persisted

In both cases, there is no obvious ambiguity with respect to the Windows 11 machine or the session data, they are just not exhaustively described.

Ambiguity in documentation

The general view is that ambiguity in documentation should be eliminated, so that the user can reliably and safely follow a given document. This is almost certainly good advice in most cases.

When operating industrial machinery, an ambiguous instruction can be fatal. Yet, not all ambiguous instructions are so consequential. When a recipe prompts us to add seasoning “to taste”, the instruction is ambiguous with respect to the quantity, but it is unlikely to harm us. Moreover, it grants us the agency to experiment and determine our preferred quantity.

Interplay with design

It may not be possible to eliminate ambiguity (or indeterminacy) from a user interface or its documentation.

Documentation can, however, be written in a way that is less ambiguous relative to the user interface.

This frees designers to leverage the virtues of ambiguity, such as efficiency of representation, and expands the possibility space for design.

In that sense, documentation balances against, and allows for, ambiguities in the interface.

Dynamics of ambiguity

Another way of thinking of a technical document, is that its reading should result in a progressive reduction in ambiguity; thus, for example, a document may introduce an action or goal that creates ambiguity for the user, before guiding the user towards a level of understanding that resolves that ambiguity.

There may not always be a linear reduction in ambiguity as a reader navigates a document. For example, a tutorial may introduce an ambiguous learning situation, which is resolved as the user develops competence. Then, the tutorial may prompt the user to apply their new skills in unfamiliar scenarios, reintroducing ambiguity, and thereby empowering the user to explore new possibilities.

Ambiguity in programming aphorisms

An aphorism is a short phrase that expresses some general truth.

While they are often syntactically clear, they can be interpreted many ways, and are therefore ambiguous.

Aphorisms are mostly associated with philosophers like Schopenhauer ⁴, but they do appear in programming culture. One of the most famous examples is the Zen of Python, which includes aphorisms like:

• Simple is better than complex.
• There should be one — and preferably only one — obvious way to do it.

What is meant by “simple” and “complex”? In what sense is one “better”?

As a language, Python often has multiple ways to achieve the same thing — in such cases, which one is “obvious”?

Nonetheless, these aphorisms strike many Python developers as reasonable and they are best understood as general principles that warrant reflection.

Aphorisms provoke debate; they don’t promote dogma. Though they’re short, aphorisms spur considered reflection, not Pavlovian partisanship ⁵.

Ambiguity in poetics

TODO

See Seven Types of Ambiguity

Tell all the truth but tell it slant

— Emily Dickinson

References