Introduction to verb phrase structure search patterns.

The following work is aimed at contrasting the structural differences between phrasal verbs and other kinds of verbal structures using a data driven perspective. More specifically, a web based syntactic parser is used to isolate certain verbal constructions based on a fore knowledge of likely possibilities. Phrasal verbs are contrasted structures precisely because they can be ambiguous on the surface and are among the most difficult constructions to account for in a data driven model. An example of the types of structures contrasted by these mechanisms are presented by the following examples:

Both, the researcher bent on describing English's verbal structures and the teacher who tries to teach them are challenged when they need to differentiate between the two basic types of English verbal structures that are examplified in sentences 1&3, because they share a common surface structure. As following analysis and discussion will highlight, that while the difference between the two structures is elusive for any system that is contigent on bottom up processing, previous knowledge of possible structure is a valuable and exciting tool. These sentences examples above are meant to demonstrate that the sentences may look the same, and may have words in the same linear order. Potentially, even the words are identical, yet the structures different. The thrust of these examples is that neither the order of the words, nor anything in the surface structure reveal any of internal differences nor their implications.
1. For the purposes of this analysis items 1and 3 are considered structurally identical. The following discussion will develop the contrast between items 1 and 2 because they are more logically contrastable as well as easier to remember and refer to. Item 3 is mentioned here only to show an example of how the the words in the two kinds of structures can indeed be identical. The contrast here is drawn between 1 where the phrase looked up is operating as a PV, and 2 where looked up is operating as a VP.
The adverbial phrase as the flag raised is not essential in or for this analysis, and is added only to make the examples more coherent.

In language analysis, there are tests or modifications that can distinguish between these two kinds of structures. This analysis will make use of these tests or algorythms. Each of these test is targeted at exploiting features in the phrase structure that verify that the resemblance the phrases share on the surface does not allow one to readily predict the type of underlying structure. Then, the implications the structural differences have on the syntax of possible structures is explored by the following discussion. With further inspec tion, the following text will demonstrate how these structures have very different properties. The application of a few established algoryhms can make the distinctions between the two kinds of sentences obvious. Here this is done to point out the difference between these structures while it isolates and describes as many mutually exclusive situations as possible.

There are three components to this presentation

• The first section reviews and lists the tests used to distinguish phrasal verbs from other kinds of constructions. This also shows the backbone to how one could predict possible structures.
  
• The second section presents a distribution for phrasal verb combinations for twenty English verbs and 71 particles based on their lexical membership in any of five phrasal verb dictionaries. The twenty verbs selected for study here were chosen because they were the most productive verbs found in the lexicons. The 71 particles were chosen because they were all the co-occuring free morphemes that seemed to make sense at the time.
  
• The third section combines the information in the earlier sections through the use of a web based text parser. The parser searches for certain verbs and particles. Once a combination is found it compares it to the phrasal verb combinations put forth in tell section 2.

  This updated parser includes the part of speech sequences listed in section one for each of the verbs combination possibilites listed in section 2 across a corpora of over 700 literary works, amounting to approximately 50 million words released through Project Gutenburg. The updated form of the parser is substantially more powerful and faster. Other features include the ability to tag certain closed class strings including negative polarity words, determiners, and conjunctions.

The underlying structure of phrasal verb sentences make them distinguishable from other kinds of constructions. For quite some time, grammarians have used tests or mutually exclusive environments to help in isolating the phrasal verb cases from verb prepositions combinations. When a test can apply to only phrasal verb then it denotes a mutually exclusive enviornment. Verb and particle combinations found in such a situation can be considered PVs.

Basically any of these tests exploit the fact that the particle and the verb form a single constituent in a phrasal verb, where they don’t in other comparable situations. The syntactic ramifications of being a single constituent are throughouly explored by the following rules. The limitations on a situation are chief the result of its phrase structure. These rules all point out that if the verb and particle are a single constiuent, and as such can undergo various transformations. ***** ********* I may delete the next two paragraphs.. but put the info somewhere else

A brief review of the earlier examples, can note some similarities beteen the two kinds of structures. We can see these sentences are different structurally In the case of 1(b) & 1(c) where they have identical words, there are two distinctly different readings each with its own properties that follow from the different structures.

The sentences' ambiguity is clear a lexically based one, because it can be attributed to the presence of a single element, the free morpheme that follows the verb. Throughout this paper this element is called a particle although it is clear, if from only the sentences above, that it can operate as more than one part of speech in a given context. **************

I. Particle shift

Here is an situation where the first example is the best example. The clearest example of the difference between PVs and VPs comes from a transformation often referred to as 'particle shift'. Particle shift points out the syntactic differences between PV and VP unlike any other transformation. In some situations this can mean that the surface structure will outwardly reveal internal phrase structure, unlike the first sentence examples. This situation only happens when then the particle is shifted. The example sentences in a shifted form would look like the following:
b.

(4).a.	He looked up the number as the flag raised.
	He looked the number up as the flag raised.
(5).a.	He looked up the pole as the flag raised. <.td>
b.	*He looked the pole up as the flag raised.

As the following citings from Linder point out, the particle's position relative to the direct object is both widely recognized by the academic community and repeatedly documented by grammarians. Mitchell (1958) reporting that the particle can occur in a pre- or post- nominal position is a good place to start. It, in turn, is similar to Live's (1965) contrast of 'M' construction's noun object in mid position with the 'P' construction's obligatory preposition before the noun. Still, yet another comparable schema is Sroka's position c1 where the particle occurs between the verb and its noun object and his position a1 where the particle is sentence final in active sentences. Yet another documentation of shifted PV forms is exemplified in Spear's dictionary's by the lexical entries that use a 'T" category, indicating that they permit particle shift.

For our purposes, this may most clearly be summed up by the following: Bolinger (1971)

"if the combination is transitive the particle can either precede or follow the noun object ."

For the following data driven analysis, the focus is on the fact Bolinger's statement seems to imply that all transitive PVs allow their particles to shift to the far side of their direct object noun phrase. While, as we'll see, the corresponding VPs enjoy no such freedom.

Clearly, as an algorithm or test, this would seem a simple mechanism for distinguishing transitive VPs from transitive PVs based on their ability to separate. This would hold only if all transitive verbs permit such flexibility with the full range of particles, or at least a predictable subset. The casual impression of most grammarians is that the verbs and particles combine assymetrically to form phrasal verbs.

A possible reason that this distribution appears asymmetric may be the result of other factors confounding factors. As further discussion points out, the factors that influence the separation of verb and particle may be subtle thus further complicating any analysis of PV distribution based on a given verb and the full range of particles. What is clear is that all noun phrases are not regarded as equal to PV constructions.

I.2 Particle Shift: Pronoun Objects

The type and quality the of noun phrase can be shown to influence the productivity of the particle shirt transformation in the following cases where pronouns are substituted for noun phrases:

	(6)a. He looked up the number as the flag raised. 

   	    b*He looked up it as the flag raised. 
	    
	    c. He looked it up as the flag raised. 
	
	(7)a.  He looked up the flagpole as the flag raised. 
	     
	    b.  He looked up it as the flag raised. 
	   
	    c.*He looked it up as the flag raised. 
 

A description of phrasal verb behavior would note that if the direct object is nominalized, a shift is practically obligatory.In a manner consistent with the earlier phenomena, the idiosyncrasies of particle shift's behaviour with pronouns has been documented and analyzed by several researchers. Opinions actually differ as to whether the word order exhibited by examples like 3(c) is obligatory or optional. This provides this study with the opportunity to conject as to wether any phrasal verbs can be found with no shift and a nominalized object as the following would document.

Linder's invaluable observations in this area note point out this interesting descriptive question. First, she notes that researchers like Live, Sroka, Palmer and Mitchell believe that this word order (3c) is obligatory, hence contrastive. If this is indeed the case, this would add yet another mutually exclusive enviorment for phrasal verb syntax.

Only further study will reveal if if this environment is exclusive. On the other hand, Linder says that the majority of people portend the relationship only as a tendency or possibility. Case and point Bolinger's (1971) examples in which the particle precedes the noun. Something about finding a naturally occuring situation.

Another person who documented and analyzed example noun phrase's effect on syntax is Wood(). Her contrast of personal or reflexive pronouns in mid position versus demonstrative and possessive pronouns in the end position makes exclusive situations for the type of noun phrases. I will have to find more info on her
This is most tersely summed in the following quote from Linder:

"Rather than list in isolation the conditions on the relative order of the particles and the pronoun object, it is insightful to view them in the context of general tendencies of VPC, which is the main topic of Wood (1955), van Dongen (1919), Erades (1961) and Chapter Nine of Bolinger (1971). The particle's position relative to full noun object is not entirely arbitrary,... but rather is determined by any of a cluster of formal and semantic factors"

1.3 Particle shift: Heaviness

When considering the reasons why the pronoun noun phrases might operate differently than other kinds of noun phrases, one inevitably notices that the pronouns are usually very small single words especially when contrasted by noun phrases consisting of several words. This might lead some to speculate that heaviness is operating as a confounding element, which limits the size of the phrase that the particle could shift over. As we can see, heaviness does effect the ability of particles to undergo particle shift lessens as the heaviness of the object noun phrase increases. Clearly, the preferred reading favors no shift after a certain point. You are invited to exercis e your particle shift tolerance threshold.

(8). a  He looked up the number. 

     b  He looked the number up. 

(9) a. He looked up the phone number. 

      b. He looked the phone number up.

 (10)a .He looked up the phone number of the theatre. 

        b. He looked the phone number of the theatre up.
 
 (11) a.  He looked up the phone number of the local movie theatre. 
 
         b.  He looked the phone number of the local movie theatre up. 
          
 (12). a  He looked up the phone number of the local,one dollar mantina movie theatre.
              
          b. He looked the phone number of the local, one dollar mantina movie theatre up.

  (13) a.  He looked up the phone number of the local, run-down,  one dollar mantina, movie theatre.
                
          b. He looked the phone number of the local, run-down, one dollar mantina, movie theatre up.

 (14) a  He looked up the phone number of the local, run-down, two-story, one dollar mantina, movie theatre.

        b.  He looked the phone number of the local, run-down, two story, one dollar mantina, movie theatre up.

Another syntactic pattern can be used to distinguish PVs from VPs in a data driven model. Becuase it is it is another mutually exclusive enviroment, nominalization can be a powerful tool for positively identifying PVs. The manner in which PVs nominalize is very productive and very regular. It is possibly the most regular of Pvs transformations. Most PVs are easily made into a nominal through the use of a clear pattern. One would expect any PV that has been nominalized, will have its accompanying NP following separated by an 'of'. If any phrase can be nominalized in this way, as shown in the following examples, then it is a clear indication that that is a PV and not a VP.

 	6. (a) He looked up the number. 

	 (b) His looking up of the number saved him the directory assistance charges. 
	 
  	 (c) He looked up the pole. 
 
	 (d)*His looking up of the pole saved him  from being crushed by the falling flag. 
 

Clearly, trying to nominalize VP just does not work the same way. Nominals are formed by placing an of between a PV and its object NP. A VP can permit no such transformation. make no such construction.

This syntactic pattern was analyzed and identified as a transformation unique to PVs by Fraser when he remarked that PVs' of required a specific placement. Here Fraser identifies this as mutally exlcusive syntactic enviornment of PV as well as a clear template for searching. We can describe this pattern to the parser in the following way.

2.2 Phrasal Verb Transformations: Manner Adverbials

The placement of adverbial information can also be used a clue as to the syntactic category of a questionable combination. This, like the other transformations, is well documented by researchers like Fraser, Legum(i have to find the year) and Mitchell.

In Fraser's analysis of syntactic patterns he notices that only VPs permit the insertion of manner adverbials between the verb and particle where PV do not. Once again the phrases syntactic structure either permits or restricts this sort of productivity. this time the enviornment is a VP one.

	7.(a) *The operator looked directly up the number as the flag raised. 

	(b) The operator looked directly up the pole as the flag raised. 

2.3 Phrasal Verb Transformations: Parenthetical

	8. (a) He looked up, with the efficiency of a computer, the number and address of  the flagpole manufacturers. 
	(b) *He looked up, with the eye of an eagle, the pole so as to assess its straightness. 

This tranformation is difficult to encode into a data driven model because of the reccursion neccessary to allow a phrase insertion.

	9. (a) The operator looked up the number as the flag raised outside in front of the building. 
	
	(b) The number was looked up by the operator as the flag raised outside in front of the building. 
	
	(c) The operator looked up the pole as the flag raised in front of the crowd. 

	(d) *The pole was looked up by the operator as the flag raised in front of the crowd. 
 

2.5 Phrasal Verb Transformations: Quasi-Passive

Fraser's examples come from Kennedy (1920)

• bills figure out to certain amounts
• dirt of a garment will brush off
• dirt of a garment will clean off
• dirt of a garment will rub off
• a chair folds up
• a fresh bone will grind up
• school lets out

2.6 Phrasal Verb Transformations: Relative Clause and Wh questions

3.0 Phrasal Verbs: Tests of Constituency

4.0 Verb + Matrix