I suspect the Google Docs grammar bot is the least useful bot there is. After hundreds of suggestions, I can think of only one instance in which it was right. Is its failure rate so high because it learns from how other people use English, instead of drawing from a basic ruleset?
I’m not saying my grammar is better than everyone else’s but if the bot is learning from how non-native users of the English language construct their sentences, I can see how it would make the suggestions it does, especially about the use of commas and singular/plural referents.
Then again, what I see as failure might be entirely invisible to someone not familiar with, or even interested in, punctuation pedantry. This is where Google Docs’s bot presents an interesting opportunity.
The rules of grammar and punctuation exist to assist the construction and inference of meaning, not to railroad them. However, this definition doesn’t say whether good grammar is simply what most people use and are familiar with or what is derived from a foundational set of rules and guidelines.
Thanks to colonialism, imperialism and industrialism, English has become the world’s official language, but thanks to their inherent political structures, English is also the language of the elite in postcolonial societies that exhibit significant economic inequality.
So those who wield English ‘properly’ – by deploying the rules of grammar and punctuation the way they’re ‘supposed’ to – are also those who have been able to afford a good education. Ergo, deferring to the fundamental ruleset is to flaunt one’s class privilege, and to expect others to do so could play out as a form of linguistic subjugation (think The New Yorker).
On the other hand, the problem with the populist ontology is that it encourages everyone to develop their own styles and patterns based on what they’ve read – after all, there is no one corpus of popular literature – that are very weakly guided by the same logic, if they’re guided by any logic at all. This could render individual pieces difficult to read (or edit).
Now, a question automatically arises: So what? What does each piece employing a different grammar and punctuation style matter as long as you understand what the author is saying? The answer, to me at least, depends on how the piece is going to find itself in the public domain and who is going to read it.
For example, I don’t think anyone would notice if I published such erratic pieces on my blog (although I don’t) – but people will if such pieces show up in a newspaper or a magazine, because newsrooms enforce certain grammatical styles for consistency. Such consistency ensures that:
- Insofar as grammar must assist inference, consistent patterns ensure a regular reader is familiar with the purpose the publication’s styleguide serves in the construction of sentences and paragraphs, which in turn renders the symbols more useful and invisible at the same time;
- The writers, while bringing to bear their own writing styles and voices, still use a ‘minimum common’ style unique to and associated with the publication (and which could ease decision-making for some writers); and
- The publication can reduce the amount of resources expended to train each new member of its copy-editing team
Indeed, I imagine grammatical consistency matters to any professional publication because of the implicit superiority of perfect evenness. But where it gets over the top and unbearable is when its purpose is forgotten, or when it is effected as a display of awareness of, or affiliation to, some elite colonial practice.
Now, while we can agree that the populist definition is less problematic on average, we must also be able to recognise that the use of a ‘minimum common’ remains a good idea if only to protect against the complete dilution of grammatical rules with time. For example, despite the frequency with which it is abused, the comma still serves at least one specific purpose: to demarcate clauses.
In this regard, the Google Docs bot could help streamline the chaos. According to the service’s support documentation, the bot learns its spelling instead of banking exclusively on a dictionary; it’s not hard to extrapolate this behaviour to grammar and syntactic rules as well.
Further, every time you reject the bot’s suggested change, the doc displays the following message: “Thanks for submitting feedback! The suggestion has been automatically ignored.” This isn’t sufficient evidence to conclude that the bot doesn’t learn. For one, the doc doesn’t display a similar message when a suggestion is accepted. For another, Google tracks the following parameters when you’re editing a doc:
customer-type
, customer-id
, customer-name
, storageProvider
, isOwner
, editable
, commentable
, isAnonymousUser
, offlineOptedIn
, serviceWorkerControlled
, zoomFactor
, wasZoomed
, docLocale
, locale
, docsErrorFatal
, isIntegrated
, companion-guest-Keep-status
, companion-guest-Keep-buildLabel
, companion-guest-Tasks-status
, companion-guest-Tasks-buildLabel
, companion-guest-Calendar-status
, companion-guest-Calendar-buildLabel
, companion-expanded
, companion-overlaying-host-content
, spellGrammar
, spellGrammarDetails
, spellGrammarGroup
, spellGrammarFingerprint
Of them, spellGrammar
is set to true
and I assume spellGrammarFingerprint
corresponds to a unique ID.
So assuming further that it learns through individual feedback, the bot must be assimilating a dataset in the background within whose rows and columns an ‘average modal pattern’ could be taking shape. As more and more users accept or reject its suggestions, the mode could become correspondingly more significant and form more of the basis for the bot’s future suggestions.
There are three problems, however.
First, if individual preferences have diverged to such an extent as to disfavour the formation of a single most significant modal style, the bot is unlikely to become useful in a reasonable amount of time or unless it combines user feedback with the preexisting rules of grammar and composition.
Second, Google could have designed each bot to personalise its suggestions according to each account-holder’s writing behaviour. This is quite possible because the more the bot is perceived to be helpful, the likelier its suggestions are to be accepted, and the likelier the user is to continue using Google Docs to compose their pieces.
However, I doubt the bot I encounter on my account is learning from my feedback alone, and it gives me… hope?
Third: if the bot learns only spelling but not grammar and punctuation use, it would be – as I first suspected – the least useful bot there is.