Towards a Profanity-Inclusive Construction Grammar (Pt. 1)

This post was originally intended to describe some issues involved in building a grammar that allows for the usage and understanding of profanity. Instead, I ended up giving a brief introduction to construction grammar, as well as a particular version of construction called Embodied Construction Grammar. If you’re familiar with these things already, feel free to skip to pt. 2, which will have the more fun bits about profanity. (Or don’t, and tell me what I got wrong.)

Construction Grammar

One of my research interests is an idea called construction grammar. Historically, linguists treated different “levels” of language as entirely separable. There was phonology (the sounds of language), morphology (the words and meaningful parts of words in language), syntax (the way words are put together), semantics (the meanings of those words), and pragmatics (the study of how context and “usage” affects meaning). In the traditional view, form was entirely separable from meaning; for example, the syntax of a sentence was unrelated to its semantics. This notion of the “arbitrariness of signs”, which applies to all levels of form and meaning, is very important in linguistics, and dates back to Ferdinand de Saussure. That is, most words (other than onomatopoeia), parts of words (e.g. suffixes, prefixes), and patterns of words have no relation to their meaning; there’s nothing about the word CAT that specifically denotes the furry and frequently aloof feline.

This all seems quite obvious once it’s pointed out. However, recent work in linguistics suggests that certain meanings actually cluster in systematic ways with certain forms. This happens at many levels of language, from sounds (https://en.wikipedia.org/wiki/Phonestheme)[1] to particular patterns of words[2]. That’s not to say particular meanings are somehow inherent in particular forms; meanings are meaningful insofar as they’re meaningful to a person, so it seems somewhat absurd to posit that a given meaning is contained within a word[3]. But it is the case that these meanings tend to cluster around particular forms. More concretely, if a particular form is observed, there’s a significant chance of a particular meaning being associated with it.

A well-studied area of this systematicity is syntax, which was classically seen as orthogonal to semantics. But some rather observant linguists discovered that certain syntactic patterns were often paired with certain meanings[4]. For example, the “Ditransitive” construction, in which a verb (e.g. “give”) is followed by two Noun Phrases (e.g. “the ball”), almost always evokes a meaning of TRANSFER. Consider the following sentences:

  1. John gave Sally the ball.
  2. John kicked Sally the ball.
  3. John sent Sally a letter.
  4. John sent Sally his wishes.
  5. John told Sally a story.

The sentence in (1) quite obviously describes a TRANSFER event (this is the meaning of “give.v”). In (2), the typical meaning of “kick.v” (to apply force to some object with your foot) is integrated into the larger TRANSFER frame implied by the construction. The sentence in (3) is another obvious case of TRANSFER, but (4) suggests a more metaphorical TRANSFER (“his wishes” are not a physical object transferred between John and Sally), with a verb more frequently used for the physical case. Finally, (5) also describes a metaphorical exchange, in which “a story” is being given to Sally by John; this is part of a larger, well-documented family of metaphors sometimes called the conduit metaphor[5].

So in all the sentences, despite having different verbs, there’s a consistent theme of TRANSFER. This is essentially the idea of a construction. That is, certain syntactic patterns are likely to have certain meanings[6]. Thus, if we wanted to write a grammar[7] to describe the English language, one fairly general construction of this grammar could resemble the following:

CXN Ditransitive-Transitive (+Verb +NP1 +NP2)
MEANING: TRANSFER
[Giver Subject]
Recipient NP1
Object NP2
Action Verb

Now there’s a construction called “Ditransitive-Transfer”, which contains a Verb and two Noun Phrases, and has a meaning of TRANSFER. Furthermore, the different components of meaning (e.g. the elements of the TRANSFER frame) are mapped to the grammatical constituents in a systematic way; the first Noun Phrase is the Recipient, and the second is the Object being transferred. The subject of the sentence is the Giver of the object, and the Verb describes the Action involved in TRANSFER.

Given this sentence and this construction, we could theoretically parse and label the different semantic elements of a sentence taking this form:

John [GIVER] kicked [ACTION] Sally [RECIPIENT] the ball [OBJECT].

Embodied Construction Grammar

There have been many attempts to formalize the notion of construction grammar, all with unique advantages and disadvantages. The formalization I’m most familiar with, and which I’ve worked on developing, is called Embodied Construction Grammar[8] (ECG). ECG is based on the idea that human language understanding is achieved by mentally simulating the events described in language. There’s considerable evidence for the simulation semantics hypothesis, and I recommend beginning with Jerome Feldman’s From Molecule to Metaphor, which describes the theory, related experiments, and ECG, in much more detail.

Crucial to the idea of ECG is that as people go about their lives, they form abstractions of their embodied experiences, from perceptual experiences to sensorimotor ones. Gradually, we construct higher-level schemas, which are still rooted in the low-level embodied schemas. These abstractions are called schemas, and represent the prototypical structure of a type of experience. The notion of TRANSFER described above is a good example, and might be formalized in the following way:

Schema TRANSFER
Roles
GIVER
OBJECT
RECIPIENT
ACTION

 Recall that construction grammar is based on the idea that certain forms are paired with certain meanings. In ECG, we pair these schemas with constructions, which describe how different parts of speech are mapped to the elements in the schema. A more formal version of the Ditransitive construction we defined above might resemble the following:

Construction Ditransitive-Transfer
Subcase of Verb-Phrase
Constituents:
    Verb: Verb
Np1: NP
Np2: NP
Form
Verb before Np1
Np1 before Np2
Meaning: TRANSFER
Np1.m <–>  RECIPIENT
Np2.m <–> OBJECT
Verb.m <–> ACTION
[Subject.m <–> GIVER]

Why This Matters

So we’ve defined a construction, which is neat and all, but how is this useful?

Well, an approach like ECG offers at least several useful qualities:

  1. It’s a formalism that allows linguists to study and capture generalities in form-meaning pairings across a language.
  2. Its emphasis on embodied schemas makes it somewhat more biologically plausible than other, more abstract grammars.
  3. Perhaps its most useful and interesting quality is that it’s computationally implemented. That is, given a grammar of a language defined in ECG, and an input sentence, a program called the ECG Analyzer[9] can produce a semantic parse of that sentence. This is useful for all sorts of things, from providing a language interface to robots and other autonomous agents[10], to detecting and analyzing metaphors in news articles[11].

For more information about the latest implementation of ECG, I suggest you look here, or email me for updates on the current status.

Coming soon is pt. 2, which will discuss how one would build a grammar that allows the natural use of profanity.

Notes/References:

[1] Darío Gutiérrez, E., Levy, R., & Bergen, B. K. (2016). Finding Non-Arbitrary Form-Meaning Systematicity Using String-Metric Learning for Kernel Regression. Acl, (1984), 2379–2388.

[2] Fillmore, C. J. (1988). The Mechanisms of “Construction Grammar.” Annual Meeting of the Berkeley Linguistics Society, 35–55. http://doi.org/10.3765/bls.v14i0.1794

[3] One of the exceptions is iconicity, which I’ll cover in another post.

[4] Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N. (2004). Learning argument structure generalizations. Cognitive Linguistics, 15(3), 289–316. http://doi.org/10.1515/cogl.2004.011

[5] Reddy, M. J. (1979). The Conduit Metaphor: A case of Frame Conflict in our Language about Language. Metaphor and Thought.

[6] Technically, constructions are posited to exist across all levels of language, from phonology to pragmatics, but much of the focus has been on syntactic constructions.

[7] A set of rules that account for the valid expressions people in language can understand and produce.

[8] Bergen, B. K., & Chang, N. (2005). Embodied Construction Grammar in Simulation-Based Language Understanding. Construction Grammars: Cognitive Grounding and Theoretical Extensions, 147–190. http://doi.org/10.1075/cal.3.08ber

[9] Bryant, J. E. (2008). Best-Fit Construction Analysis. Analysis.

[10] Trott, S., Appriou, A., Feldman, J., & Janin, A. (2015). Natural Language Understanding and Communication for Multi-Agent Systems. AAAI Fall Symposium, 137–141.

[11] Narayanan, S. (1997). Knowledge-based Action Representations for Metaphor and Aspect (KARMA). Computer Science, 300.

One thought on “Towards a Profanity-Inclusive Construction Grammar (Pt. 1)

  1. Pingback: Towards a Profanity-Inclusive Construction Grammar (Pt.2) | Sean Trott's Page

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s