Today we\u2019re going to delve into the parser component. Let\u2019s start with some syntactic basics!<\/p>\n
Syntax is the set of rules and processes governing sentence structure in any natural language. It involves things like word order, and constituents <\/strong>(words or phrases that form functional units). One of the most common ways to represent syntactic information (at least as of the 20th<\/sup> century) is with a syntax tree<\/strong>. Traditional syntax trees specify:<\/p>\n Here\u2019s a tree diagram (specifically, a constituency tree) for the sentence, \u201cMy parakeet drinks mimosas in the morning\u201d:<\/p>\n You can see that my parakeet <\/em>forms a larger chunk which is a noun phrase, in the morning <\/em>forms a larger chunk which is a prepositional phrase, drinks mimosas in the morning <\/em>forms an even larger chunk which is a verb phrase, and both the NP and VP combine to form the largest chunk, a full sentence S. Remember that syntax focuses on phrasal order and structure, not meaning or context \u2013 so it can\u2019t tell us why on earth you\u2019re feeding boozy orange juice to your pet bird.<\/p>\n Onto the parsing! Very generally, a parser is a piece of software (often a trained machine learning model) that takes input text, and outputs a parse tree or similar structural representation, based on syntactic rules and statistics learned from its training data.<\/p>\n Syntactic parsers include a component called a Context-Free Grammar<\/strong>, which has:<\/p>\n {S, NP, VP, PP, D, N, A\u2026}<\/p>\n {drinks, parakeet, mimosas, morning, my, in, the}<\/p>\n S \u2192\u00a0NP VP\u00a0 (a sentence S is composed of a noun phrase NP and verb phrase VP)<\/p>\n NP \u2192\u00a0D N\u00a0 (a noun phrase NP is composed of a determiner D and a noun N)<\/p>\n VP \u2192\u00a0VP PP\u00a0 (etc.)<\/p>\n PP \u2192\u00a0P NP<\/p>\n The parser starts at S, and applies its rules successively, until it arrives at the terminal symbols. The resulting parse is the labeled relationships connecting those terminals (i.e. words).<\/p>\n There are two main kinds of syntactic parsers: dependency<\/strong> and constituency<\/strong>. To keep this post to a reasonable length, I\u2019ll focus on dependency only, but constituency parsers output structures similar to the parakeet tree above[3]<\/a>. A dependency parser builds a tree for each input sentence by starting with a sentence root <\/strong>(usually the main verb), and assigning a head<\/a> word to each word, until it gets to the end of the sentence. (Heads link to dependents<\/strong>.) When it\u2019s done, each word has at least one branch, or relationship, with another word. The parser also characterizes each word-word relationship. These are things like: nominal subject of a verb (\u201cnsubj\u201d); object of a verb or a preposition (\u201cdobj\u201d and \u201cpobj,\u201d respectively); conjunction (\u201ccc\u201d for the conjunction word, and \u201cconj\u201d for the elements being conjoined); determiner (\u201cdet\u201d); and adverbial modifier (\u201cadvmod\u201d).<\/p>\n A visualized example will probably help. Taking that same sentence, \u201cMy parakeet drinks mimosas in the morning,\u201d a visualization of the dependency parse might look like this:<\/p>\n Can you spot the root, or main verb? It\u2019s the one without any arrows going towards it: drinks<\/em>. The parser then finds the subject of drinks<\/em>, which is parakeet<\/em>, and labels that relationship \u201cnsubj.\u201d It finds mimosas <\/em>as the direct object of drinks<\/em>, and labels it \u201cdobj.\u201d And so on and so forth.<\/p>\n Let\u2019s look at another example, for a dollop of variety. Here is \u201cMr. Vanderloop had smiled and said hello\u201d:<\/p>\n In this one, the past participle smiled <\/em>is the root\/main verb, which has multiple dependents: its subject Vanderloop<\/em>, its auxiliary (a.k.a. \u201chelping verb\u201d) had<\/em>, its conjunction and<\/em>, and the other verb with which it conjoins, said<\/em>. The subject Vanderloop <\/em>has a dependent Mr.<\/em>, with which it forms a compound (proper) noun; said<\/em>\u2019s dependent is the interjection hello<\/em>.<\/p>\n How about our sentence from the beginning, \u201cFind flights from Chicago to New York on February 21\u201d? How might it be parsed? (You can check your hypotheses by typing the sentence into an interactive demo of the displaCy dependency visualizer<\/a>, from which the visualizations above also came[4]<\/a>.) Something to keep in mind here is that English imperative structure leaves the subject \u2013 whoever is being addressed \u2013 implicit.<\/p>\n A slight aside: I\u2019ve chosen simple examples for demonstration, but parsing gets decidedly complicated when input sentences are themselves complicated. Questions, subordinate clauses, coordination (or all three: \u201cWhat\u2019s the name of the movie where the guy drives a flying taxi and saves the human race from aliens?\u201d), and structurally ambiguous sentences (\u201cThe horse raced past the barn fell\u201d) get tricky quickly.<\/p>\n So now we have some parsed output. How is this structured, annotated data useful? Well, one thing you can do with these word relations is identify noun phrases. Identifying noun phrases across sentences helps with another step in the NLP pipeline called Named Entity Recognition<\/strong>, or NER<\/strong>. NER tries to recognize nouns\/noun phrases (names, places, dates, etc.) and label them with categories of concepts from the real world. In our flights <\/em>example, \u201cChicago\u201d and \u201cNew York\u201d should get tagged with some label like CITY or GEOGRAPHIC LOCALE, and \u201cFebruary 21\u201d should get tagged with DATE. Once a text has been automatically annotated for such named entities, information about those entities can then be pulled from a knowledge base (say, Wikipedia).<\/p>\n Having parts of speech and word relations also makes it easier to match up the specifics of a given user command (e.g. \u201cText mom saying I\u2019ll call tonight,\u201d or \u201cShow popular Thai restaurants near me\u201d) with slightly more generalized intents (e.g. Send text<\/em> or Get restaurants<\/em>); machine models can start learning how words typically pattern across the main verb and direct object positions for various commands. Code then uses the more generalized intent to fulfill that request on a device \u2013 be it smartphone, tablet, or home speaker. \u201cFind flights from Chicago to New York on February 21\u201d would hopefully be matched with a more general Get flights <\/em>intent, and the particular noun phrases could be passed to fields for origin, destination, and date.<\/p>\n * * * * *<\/p>\n Before leaving you to your holiday leftovers, I\u2019d like to reiterate that syntactic parsing is only one step in an NLP system. Its parses don\u2019t tell us much about the actual semantics of the linguistic input. Language meaning, however, is a whole other ball of wax, best left for the new year\u2026<\/p>\n <\/p>\n [1]<\/a> There is often terminological confusion between NLP <\/em>and NLU <\/em>(natural language understanding). See this graphic<\/a> for one common breakdown, although I\u2019ve heard the terms used interchangeably as well.<\/p>\n [2]<\/a> If you\u2019re interested to learn about other NLP steps, read this accessible post, Natural Language Processing is Fun!<\/a><\/p>\n [3]<\/a> You can also play around with this interactive demo from Stanford CoreNLP, http:\/\/corenlp.run<\/a>. In the second \u201cAnnotations\u201d field dropdown, make sure you have \u201cconstituency parse\u201d selected.<\/p>\n\n
\n
\n
\n
![\"tree\"](\"https:\/\/i0.wp.com\/linguamonium.com\/wp-content\/uploads\/2018\/12\/tree.png?resize=346%2C224&ssl=1\")
<\/p>\n\n
\n
\n
\n
![\"displacy_parse_parakeet_drinks\"](\"https:\/\/i0.wp.com\/linguamonium.com\/wp-content\/uploads\/2018\/12\/displacy_parse_parakeet_drinks.png?resize=640%2C142&ssl=1\")
<\/p>\n![\"displacy_parse_vanderloop\"](\"https:\/\/i0.wp.com\/linguamonium.com\/wp-content\/uploads\/2018\/12\/displacy_parse_vanderloop.png?resize=640%2C136&ssl=1\")
<\/p>\n