Friday, September 3, 2010

A Change of Mood Revisited (by a professional linguist)!

A few days ago I wrote a post called "A Change of Mood", describing HL7 Mood Codes and comparing them to some existing linguistic moods, mostly for fun.  It looked like book fodder, except that I didn't know what I was going to do with it.  Now I do, so I had a professional linguist review and comment on my post.

I've known Dr. Win Carus since 1992 when I joined the Software Division at Houghton Mifflin to work on spelling correction. He was the brains behind much of the linguisting technology we developed there.  All I really had asked Win for was a little feedback.  This is what I got, in his rather normal expansive manner.  It was too good to keep to myself.

Win's comments appear below.

(a) Etymology: The term "mood" -- like many other incomprehensible English grammatical terms -- has its origin in Latin grammar (Latin 'modus', measure, manner). This is the same etymology for the term "modality" as used in logic. English has borrowed a lot of terms for linguistic and literary phenomena, especially from Latin and Greek, but you will find many French and German technical terms, and even Sanskrit here and there.

"Mood" is a technical linguistic term, so saying "it doesn't make sense" is a bit over the top. Yes, it has an unfamiliar sense, a sense with which your readers are probably not acquainted.

(b) Conflation: This term is used in the Wikipedia article you provide a link to, but I think it needs some clarification.

First, it is not at all uncommon for linguistic forms to carry multiple pieces of linguistic information. A simple example in English: the 's' ending of verbs indicates "third (person) singular (number) present (tense)". Keep in mind, however, that this phenomenon is found rather regularly in "fusional" or "inflecting" (more generally, "synthetic") languages, such as Indo-European languages.

Conflation of this sort is far less common in "agglutinative" languages such as -- Finnish, Hungarian, and Turkish -- in which the grammatical information of a lexeme is constructed by stringing together sequences of grammatical morphs.

And there's another linguistic phenomenon often at play in synthetic languages: "syncretism". This is where a single form can have multiple grammatic functions. For example, the "bare" English verbal form performs the functions of an infinitive and a non-third-person-singular present tense form (for example, the Brown Corpus tag set uses a single "VB" tag for both of these functions while the Penn tag set uses "VB' and "VBP", respectively).

"Unmarked" grammatical categories (such as the indicative mode; see the discussion in the next section) are by their very nature syncretic.

(c) Definitions of "mood": The great linguist Roman Jakobson gave an interesting way of thinking about mood years ago in his "Shifters, Verb Categories and the Russian Verb". He starts by distinguishing between three things:

(1) the narrated event (basically, what the speaker is talking about, the unadorned "just-the-facts-m'am")
(2) the speech event itself (the speech event occurs "here" and "now"; this is clearly a fundamental frame of reference for all sorts of ["deictic" and "indexical"] grammatical and semantic categories such as tense, locational prepositions and personal pronouns)
(3) the participants in the speech event (the speech event is produced by "me" and is directed to "you" and sometimes talks about "him", "her", "it" and "them")

Jakobson describes mood as how the speaker presents the narrated event to the listener, the speaker's overlay on the more or less objective facts of the narrated event. Mood presents the facts from different perspectives, divided most prominently between "real" ('realis') and "unreal" ('irrealis'). The "indicative" form is the basic "real" mood sub-type. (Jakobson also considers the indicative to be a semantically "unmarked" category: indicative forms can be used when speaking about both real and unreal narrative events, such as in the English "I hope that Johnny eats his vegetables.")

There are a wide range of "unreal" mood sub-types: imperative (which can be viewed as a demand or request that a narrative event come into being or be realized); conditional (a counterfactual or hypothetical narrative event); optative (a narrative event we wish or hope to be come the case or be realized); interrogative (asking about the truth of the given narrative event); potential (a narrative event that is considered probably or likely); jussive (requesting or exhorting the listener to accept the narrative event as true and possibly act on that knowledge); and so on (there are lots more, as you've learned already).

Most languages make a grammatical distinction between real and unreal; most languages don't have a large number of grammatical sub-types.

(Intellectual History Note: Jakobson's view on grammatical categories was strongly influenced by both C.S. Peirce's theory of signs, de Saussure's theory of linguistic signs and the Shannon-Weaver theory of communication.

(d) How mood is expressed: As you noted, mood is a property of verbs. But this is really a simplification. It's really a property of verbal linguistic expressions and more generally predications. Mood may be expressed:

(1) by a verb form (for example, indicative and subjunctive forms in Spanish)
(2) by a syntactic pattern (for example, English inverted interrogative word order ["Can you help me?"])
(3) by verb modifiers (for example, "particles" such as the Russian particle "by" used with preterite verb forms to express the subjunctive);
(4) periphrastically (for example, "[I would be very pleased if you would] help me"); and
(5) by differences in verbal intonation ("HELP me!" "Help ME?")

Many grammatical phenomena can be expressed in multiple ways (think, for example, about the ways you can express plurality in English).

American English has almost entirely lost purely morphological mood. Mood is now usually expressed by syntactic pattern, periphrastically or by intotation.

(e) What's the value of grammatical mood to your discussion of HL7 mood? In my opinion, the range of sub-types of grammatical mood does give you starting frame of reference about distinctions between real and unreal narrative events. But you must keep in mind that the distinctions talked about are about how languages have *grammaticalized* mood. You can see this wide range of grammaticalized mood in the two main Wikipedia articles on mood (Grammatical Mood and Eventive Mood). But you must always remember that this is just what it is: a list of grammatical categories.

The grammar of a language enforces certain ways of expressing certain kinds of information (this is what we think of as its grammar; and, of course, you really don't speak the language correctly or even intelligibly if you don't follow these grammatical prescriptions). But, by its very nature, language provides the linguistic tools to express notions that are not enforced or required by the grammar. There's a wonderfully clear discussion of this point in a recent New York Times article: "Does Your Language Shape How You Think?" by Guy Deutscher.
This is a very condensed presentation of his soon-to-appear book, “Through the Language Glass: Why the World Looks Different in Other Languages". This idea is not really that new. It was already apparent to Peirce that a sign has the property that it can be used as the basis to generate more complex signs (what's called "infinite semiosis").

Outside of the grammatical moods prescribed by a language, there are essentially unlimited and arbitrary ways to express moods in a language.

The discussions of grammatical mood can be used to identify a rich "starter kit" of narrative "moods" in HL7. However, what's most important should be a HL7-specific goal: What are the most useful sub-types (and these will undoubtedly fall into the "unreal" bucket) for describing medical narrative events? Keep in mind that the most important distinction is between "real" and "unreal" moods; and that you can define -- at will and entirely arbitrarily -- how many sub-types you wnat to use. What these grammatical mood descriptions help you to do, in addition, is to determine how one mood sub-type can be distinguished from another.

There is no reason to shoehorn the definitions you make for HL7 moods into grammatical moods. Note that you gave some one-to-many and many-to-one mappings in your attempts. My opinion is that grammatical moods per se don't matter for this work. Your goal should be to make sure that you've set up reasonable distinctions for characterizing medical assertions.

(f) Some Observations about HL7 moods:

(1) Actor/Patient Roles Unspecified: Reading through the list of HL7 moods (, it is unclear who is the actor and who is the acted-upon of these assertions. For instance, a physician proposes a drug to treat the patients disorder and a patient proposes to be more diligent in taking his medications. Is it assumed that only medical staff can be an actor? Is only the physician case allowed?
(2) Heterogeneity: the HL7 moods are extremely heterogenous. (Many clearly could never be matched by something like a grammatical mood.) Take, for instance, OPT and ActMoodPredicate. These are not moods; they are organizational labels for relating moods.
(3) Name/Description Mismatch: ActMoodPromise is characterized thoroughly as a "commitment". Why not call it ActMoodCommitment?
(4) Temporal and Locational Intermixed: The specification of some moods (e.g., EVN, SLOT and APT) have temporal or locational constraints.
(5) Contextual Dependence: Some moods (e.g., ActMoodActRequest) depend on other moods.
(6) Definitional: The mood DEF is a definition (probably some kind of communally accepted and mandated description) that holds independent of time, place or event. This seems to something that stands apart from act-oriented moods.
(7) Act as Uniform Basis for Moods: It seems very odd to base moods uniformaly on an "act". For EXPEC, for instance, how is the prognosis of a condition interpretable as an "act"?

I could write much more about the HL7 moods.


(1) Operational Rules: With respect to distinguishing one type of mood from another, I would suggest that you use "operational" questions for making these distinctions. These are questions that can be answered "yes" or "no" about how a narrative event is being viewed, to place it in one mood or another (sometimes in more than one). For instance, to determine that you're dealing with hortative mood, you could ask: "Is the physician or medical staff [speaker] speaking to the patient [listener] and strongly asking or requesting the patient to do something?" (Here "speaking" stands for communication in general; and there are a wide range of medical "exhortations" -- stopping smoking, taking medicines regularly, and so forth.) Similar formulas with examples could be constructed for each type of mood.

(g) And to help you think more about how we communicate, I'd suggest that you also think about two further subjects:

(1) "speech acts"
To quote from the Wikipedia article on "Speech acts":

"Speech acts can be analysed on three levels: A locutionary act , the performance of an utterance: the actual utterance and its ostensible meaning, comprising phonetic, phatic and rhetic acts corresponding to the verbal, syntactic and semantic aspects of any meaningful utterance; an illocutionary act: the semantic 'illocutionary force' of the utterance, thus its real, intended meaning (see below); and in certain cases a further perlocutionary act: its actual effect, such as persuading, convincing, scaring, enlightening, inspiring, or otherwise getting someone to do or realize something, whether intended or not (Austin 1962)."

This work arose from a short book by J.L. Austin, "How to Do Things With Words".

Thinking about "speech acts" shows how complex the underlying act of communication really is; and that there are a lot of unstated inferences we can draw and implications we can make from even apparently very simple statements.

(2) Grice's Conversational Maxims
Grice is interested in the link between our utterances and what we're talking about. Here's a short description from the Wikipedia article:

"The Maxims are based on his cooperative principle , which states, ‘Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged,’ and is so called because listeners and speakers must speak cooperatively and mutually accept one another to be understood in a particular way. The principle describes how effective communication in conversation is achieved in common social situations and is further broken down into the four Maxims of Quality, Quantity, Relevance and Manner."

These maxims give a different way to analyze how the medical information for an encounter is constructed.


Win Carus is the President and Founder of Information Extraction Systems.  InfoExtract has developed a suite of high-performance, multilingual, adaptive, platform-independent natural language and semantic processing tools which they license and use to develop applications that combine structured, semi-structured and unstructured information. Their medical informatics application suite includes applications for physician referrals, patient selection, medical terminology servers, and patient problem list processing.


Post a Comment