Jump to content

Verb Alternations


Kevin Grasso

Recommended Posts

Hi all,

 

I am new to Accordance, and I am hoping to use it on a search I would like to perform on the Biblical Hebrew syntax database. I am trying to find all of the verbs that have "argument alternations". Here is an example from English:

 

"Jim is eating"

 

"Jim is eating apples"

 

Essentially, I need to specify something like the following: all verbs with an optional complement.

 

Does anyone know the best way to do this? What would be ideal for my purposes is if Accordance could spit me out a list of roots with these alternations. That seems like a lot to ask for, but it would be nice if it could be done.

 

It would also be great to find all verbs that alternate between prepositions in their complements, e.g. האמין ב\ל, and all verbs that alternate between one and two objects. I don't know how this would be done, since it seems like all of these would simply be labelled "complement".

 

Finally, I guess this is a question for Robert Holmstedt, but does anyone know the method for distinguishing between complements and adjuncts? I know this is a debatable topic in general linguistics, but I am just wondering why, for example, בגוים is treated as a complement rather than an adjunct in Psalm 110:6 when every other occurrence of דין has a DO marked with את (or unmarked). In English, you can see the difference in "He will judge the nations" vs. "?He will judge among the nations" (latter sentence sounds funny in English, but you get the point).

 

Sorry for all the questions, and thanks in advance for the help!

 

Kevin

Link to comment
Share on other sites

Kevin,

 

Our method has become more nuanced over the years of the project. John Cook has been working on valency for almost a decade and has refined his approach to BH enough that he's begun a valency lexicon project. Our database, in some ways, reflects the progression of sophistication, and so part of the update that will come out very soon (today I'm finishing a set of edits) reflects changes to valency decisions. 

 

On Ps 110:6, I'll paste in an early analysis (it will no doubt change in style, but not in substance as the lexicon project comes back to this verb):

Only one possible monovalent example (Ps 110:6), but some question the text. It is best treated as a bivalent with PP ‘arbitrate between nations.’ Bivalent frequently with ‘people’ (עם, לאמיםas complement (Ps 9:9; 50:4; etc.). Jeremiah alone has a cognate accusative ‘judgment’ (דין; in 21:12 משׁפט) throughout.

 

On searching for "eating" versus "eating apples," since the database is primarily a syntax database, with the semantics of valency included simply because valency sits on the syntax-semantics divide, we do not have verbs themselves categorized ("tagged") in a way that would make the data you want the product of a simple search. You need to determine the verbs that you think have such alternations and then create parallel monovalent (no complement) and bivalent (one complement) searches to find both options. Also, though the valency literature is a bit messy, we do not take complements to be "optional"; that is the definition of an adjunct.

 

Finally, to find the alternations between different types of complements (e.g., NP w or w/o את vs. a ב-PP), you would again need to predetermine the verbs you're looking for, create a search for that verb (1st level -- predicate phrase: 2nd level -- predicate = verb, complement phrase, with complement = noun or preposition). Then you can you use the regular analytics in Accordance to produce a pie chart, etc. Or you can export the list of examples, though you'll need to sort them into the categories yourself. I'm sure Accordance will continue to develop the analytics for the syntax databases (since ours isn't the only one now).  I've attached my search for האמין below. (Note that it caught a couple unrelated and tricky inter-clausal issues that I've since fixed and are in the updated database to be released soon). Also note that the same search without NOUN or Preposition specified in the complement gets the same hits, but then the analytics don't give as much information. Some verbs will, of course, have clausal complements and so this search would not get those particular complements for such verbs.

BivalentVerbBELIEVE.accord.zip

Edited by Robert Holmstedt
Link to comment
Share on other sites

Kevin,

 

It is not too difficult to find examples of complements marked by את OR marked by prepositions, but you won't be able to find verbs that occur in an alternation in a single search. (Let me know if you need details on how to build the Hebrew Construct for these two searches).

 

I would do each search separately, then use analytical tools > analysis to build a list of the verbal root/binyan combos present in each result set. (Make sure you add stem to the analysis so you can see binyan; do this by clicking the little gear in the upper right hand corner of the analysis dialogue). By comparing the two lists you could figure out which verbs occur in both searches. 

 

Unfortunately, it presents theoretical problems to search the syntax DB for something that is NOT there, so I can't really think of a good way to search for verbs that occur with no complement. I have had the same problem looking for complement/adjunct phrases with no explicit prepositional marking. Maybe someone smarter then me has a good idea.

 

I will let Rob explain the issues with complements/adjuncts, but just note that the distinction can be subjective, so for your research you will need to define for yourself what you want to count as a proper complement. There are certain adjuncts in the DB that I would really count as complements in my framework and vice versa.

 

Pete

 

EDIT: Or you can do it Rob's way. Ha, what timing!

Edited by Peter Bekins
Link to comment
Share on other sites

Thanks for the comments, guys. It sounds like I should talk to John Cook about his work some.

 

Rob, one theoretical question so I can know how you divvy up complements and adjuncts. Take sentences like the following:

 

"Jim broke the window"

 

"The window broke"

 

Under your definition of "adjunct", "Jim" would fall into that category, correct? This is a typical unaccusative where the only underlying necessary argument is the object of the verb (there are, of course, many of these alternations). In the same vein, would you also consider the DO in the above alternation to be an adjunct ("I am eating an apple"), since it too is "optional"?

 

I am just curious because it obviously affects what I would search for in the database. Personally, I wouldn't call those adjuncts, but regardless, I would like to know more precisely how you are using the term.

 

Pete, I might be asking you for how to perform some of these searches, but I will let you know. As I said, I need to tinker with them first and think through the best way to do these searches.

Link to comment
Share on other sites

Kevin,

 

The syntax DB mixes pure syntactic relations with semantic relations into something of a hybrid schema. Rob can better explain the theory/reasoning behind this.

 

The short answer to your question, however, is that a clause is defined as Subject + Predicate so Jim is the syntactic subject and from the perspective of the tagging theory the adjunct/complement distinction doesn't apply. 

 

Pete

Link to comment
Share on other sites

What Pete said.

 

If we went down those paths, the database would be unusable to all but an extreme few (e.g,. those who actually know what an unaccusative is). Our m.o. is "Data primary, theory wise," to which perhaps I should add, "user-in-mind".

 

And, I might add, "Jim" is not syntactically (or semantically) optional in your first clause; this is the transitive alternation with the unaccusative form and the transitive alternate crashes without an overt subject/agent.

 

And on the "eating" versus "eating an apple," you'll need to contact Cook, since we distinguish between monovalency, bivalency with implicit complements, and bivalent with null complements. He discussed this some at our Accordance session at SBL.

Edited by Robert Holmstedt
Link to comment
Share on other sites

Pete, you are right. I forgot that subject was treated as a category of its own, so that actually solves the issue (like I said, I am new to Accordance).

 

Thanks for the feedback from both of you. I think theoretically we would be on the same page if "eating an apple" was some sort of null complement in a bivalent verb. That is helpful.

 

I also didn't mean to sound argumentative. The issues are actually quite complex, and it is helpful to know where you are coming from in the database.

 

Thanks again to you both!

Link to comment
Share on other sites

Kevin,

 

I don't think you were intending to be super precise, but just to make sure, 'Jim is eating' would NOT involve a null complement in a bivalent verb. Null complements are inserted as place holders in pro-drop situations where a referent is highly identifiable but not overtly expressed by a lexical item in a clause. English doesn't allow this in normal clauses, but the idea would be:

 

'My mom gave me a cookie. I left it on the table, but Jim ate (it).'

 

Examples like 'Jim is eating' or 'The window broke' are treated as valency reductions in which one of the arguments is fully deleted. In this case there is no null place-holder in the clause. These deletions are allowed by lexical rules and they produce semantic variations from the base predicate.

 

'Jim ate it' = eat (x, y) where x = 'Jim' and y = 'it'

'Jim was eating' = eat (x) where x = 'Jim' (deletion of y  = -telic)

 

'Jim broke the window' = break (x, y) where x = 'Jim' and y = 'the window'

'The window broke' = break (y) where y = 'the window' (deletion of x = -agent)

 

Long story short, you will not find the eat (x) examples by looking for אכל combined with a null complement. It will have no complement at all, and I am still not 100% sure how to search for these.

 

Pete

Link to comment
Share on other sites

Thanks again, Pete. I would actually also not consider "eat" to have a null DO/complement in every case when there is no overt one, but I thought it might be how you guys are dealing with the problem of argument deletion (since, if you can delete it, it would be optional, and if optional, an adjunct according to the definition being used - though I don't necessarily ascribe to that definition of adjunct). I now understand how you are using a null complement (which is how I would use that term), but again, that is helpful to know how the database is set up.

 

As for searching, if I can't do it, I will just go through the verbs myself. I'll see what I can do.

 

Again, I appreciate the help and discussion.

Link to comment
Share on other sites

It's neither null or deleted but implicit. This is the direction valency research has gone to solve this issue.

 

In the database, they are tagged as monovalent (no syntactic complement).

Edited by Robert Holmstedt
Link to comment
Share on other sites

To jump in after a weekend away from work . . .

 

I think this discussion demonstrates the confusion entailed by the term "optional," though I myself have been guilty of using it. To provide more insight for you, Kevin, the valency approach we are taking in the database distinguishes between complements that are LICENSED by the verb versus adjuncts that are not. That is, the semantics of the verb dictate to some extent what argument structures or valency patterns the verb may meaningfully appear in and which it cannot.

 

In the case of verbs like 'eat'/אכל, which admit a good deal of variation between bivalent and monovalent argument structures, we are attempting (slowly through additional passes through the data) to distinguish between implicit complements and null complements. As Peter stated, null complements are cases of pro-drop in which an antecedent in the context or pragmatic setting enable recovery of the constituent. The simplest test for a null complement is whether a pronoun can be inserted that has a meaningful antecedent in the context. Implicit complements are *not* understood as deleted complements, contrary to what Peter mentioned and in keeping with Rob's corrective. Instead, they are suggested by the semantics of the verb. So, for example, 'Jim is eating' has not VP complement for the verb, but the reader will infer that Jim probably ate something edible. While there are a small group of linguists that would argue we treat 'eat' w/o the complement and 'eat' with a complement as two distinct lexical entries, most want to retain the intuitive connection between the two verbs.

 

Precisely how the semantics of the verb interact with argument structures, which appear to exist separately at times from the lexical items, is a question I hope to gain some insight from as we begin in earnest taking our stock of data and writing valency dictionary entries. For now, a helpful place to gain a grasp of the issues is Beth Levin and Malka Rappaport-Hovav's book Argument Realization (Cambridge University Press, 2005). If you want more insight into the issue of null versus implicit as we are deciding cases in the database and valency dictionary, e-mail me for a copy of a draft paper on the issue. I'd be happy to send it to you.
 

John

Link to comment
Share on other sites

I see. RRG actually does this slightly differently, so I misunderstood how John was handling it. I assumed from the fact that we originally described the 'Jim ate' type of examples as monovalent that we deleted an argument.

 

In RRG the semantic and syntactic valence are treated separately. So in the lexicon 'to eat' is eat (x,y) with a semantic valence of 2, but the verb is compatible with a syntactic valence of either 1 or 2. 

 

Jim ate fish = eat (Jim, fish); semantic valence = 2 and syntactic valence = 2

Jim ate = eat(Jim, ∅); semantic valence = 2 and syntactic valence = 1

 

The issue is that RRG represents the argument as null (∅) in the argument structure since there is no referent, but I wasn't sure how you have chosen to formalize this since null refers to a syntactic null in Rob's system. This is called an inherent argument in RRG, similar to implicit argument.

 

Pete 

Link to comment
Share on other sites

Yes, Pete, RRG deals with the matter differently. In fact, there are quite a number of ways linguists of various camps deal with the issue. As I mentioned, some linguists argue for a monovalent eat and a bivalent eat. While I don't go that route, neither do I feel the need to delete a complement to get to the monovalent.

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...